program(1.0)
[buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "3520.4.1"}, {"coremlc-version", "3520.5.1"}})]
{
    func main<ios17>(tensor<fp16, [2, 2048, 1, 77]> encoder_hidden_states, tensor<fp32, [2, 1280, 32, 32]> hidden_states_281_cast_fp16, tensor<fp32, [2, 640, 64, 64]> input_113_cast_fp16, tensor<fp32, [2, 640, 32, 32]> input_115_cast_fp16, tensor<fp32, [2, 320, 128, 128]> input_13_cast_fp16, tensor<fp32, [2, 1280, 32, 32]> input_213_cast_fp16, tensor<fp32, [2, 1280, 1, 1]> input_21_cast_fp16, tensor<fp32, [2, 320, 128, 128]> input_29_cast_fp16, tensor<fp32, [2, 2560, 32, 32]> input_423_cast_fp16, tensor<fp32, [2, 320, 128, 128]> input_43_cast_fp16, tensor<fp32, [2, 320, 64, 64]> input_45_cast_fp16, tensor<fp32, [2, 640, 64, 64]> input_79_cast_fp16) {
            tensor<string, []> cast_7_dtype_0 = const()[name = tensor<string, []>("cast_7_dtype_0"), val = tensor<string, []>("fp16")];
            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, []>(64)))];
            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, []>(768)))];
            tensor<string, []> cast_2_dtype_0 = const()[name = tensor<string, []>("cast_2_dtype_0"), val = tensor<string, []>("fp16")];
            tensor<string, []> cast_4_dtype_0 = const()[name = tensor<string, []>("cast_4_dtype_0"), val = tensor<string, []>("fp16")];
            tensor<string, []> cast_3_dtype_0 = const()[name = tensor<string, []>("cast_3_dtype_0"), val = tensor<string, []>("fp16")];
            tensor<string, []> cast_5_dtype_0 = const()[name = tensor<string, []>("cast_5_dtype_0"), val = tensor<string, []>("fp16")];
            tensor<fp16, [640]> add_11_mean_0_to_fp16 = const()[name = tensor<string, []>("add_11_mean_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1472)))];
            tensor<fp16, [640]> add_11_variance_0_to_fp16 = const()[name = tensor<string, []>("add_11_variance_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2816)))];
            tensor<string, []> cast_8_dtype_0 = const()[name = tensor<string, []>("cast_8_dtype_0"), val = tensor<string, []>("fp16")];
            tensor<string, []> cast_9_dtype_0 = const()[name = tensor<string, []>("cast_9_dtype_0"), val = tensor<string, []>("fp16")];
            tensor<string, []> cast_10_dtype_0 = const()[name = tensor<string, []>("cast_10_dtype_0"), val = tensor<string, []>("fp16")];
            tensor<fp16, [1280]> add_23_mean_0_to_fp16 = const()[name = tensor<string, []>("add_23_mean_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4160)))];
            tensor<fp16, [1280]> add_23_variance_0_to_fp16 = const()[name = tensor<string, []>("add_23_variance_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6784)))];
            tensor<string, []> cast_6_dtype_0 = const()[name = tensor<string, []>("cast_6_dtype_0"), val = tensor<string, []>("fp16")];
            tensor<int32, []> var_30355 = const()[name = tensor<string, []>("op_30355"), val = tensor<int32, []>(1)];
            tensor<string, []> cast_0_dtype_0 = const()[name = tensor<string, []>("cast_0_dtype_0"), val = tensor<string, []>("fp16")];
            tensor<fp16, [2560]> add_43_mean_0_to_fp16 = const()[name = tensor<string, []>("add_43_mean_0_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9408)))];
            tensor<fp16, [2560]> add_43_variance_0_to_fp16 = const()[name = tensor<string, []>("add_43_variance_0_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14592)))];
            tensor<string, []> cast_1_dtype_0 = const()[name = tensor<string, []>("cast_1_dtype_0"), val = tensor<string, []>("fp16")];
            tensor<string, []> x_5_pad_type_0 = const()[name = tensor<string, []>("x_5_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> x_5_strides_0 = const()[name = tensor<string, []>("x_5_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> x_5_pad_0 = const()[name = tensor<string, []>("x_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> x_5_dilations_0 = const()[name = tensor<string, []>("x_5_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> x_5_groups_0 = const()[name = tensor<string, []>("x_5_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2560, 1, 1]> up_blocks_0_resnets_0_conv_shortcut_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2457600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(19776))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2477440))), 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, []>(2477632)))];
            tensor<fp16, [2, 2560, 32, 32]> cast_0 = cast(dtype = cast_0_dtype_0, x = input_423_cast_fp16)[name = tensor<string, []>("cast_2")];
            tensor<fp16, [2, 1280, 32, 32]> x_5_cast_fp16 = conv(bias = up_blocks_0_resnets_0_conv_shortcut_bias_to_fp16, dilations = x_5_dilations_0, groups = x_5_groups_0, pad = x_5_pad_0, pad_type = x_5_pad_type_0, strides = x_5_strides_0, weight = up_blocks_0_resnets_0_conv_shortcut_weight_to_fp16_palettized, x = cast_0)[name = tensor<string, []>("x_5_cast_fp16")];
            tensor<fp16, [2, 1280, 32, 32]> cast_1 = cast(dtype = cast_1_dtype_0, x = hidden_states_281_cast_fp16)[name = tensor<string, []>("cast_1")];
            tensor<fp16, [2, 1280, 32, 32]> hidden_states_283_cast_fp16 = add(x = x_5_cast_fp16, y = cast_1)[name = tensor<string, []>("hidden_states_283_cast_fp16")];
            tensor<int32, [5]> reshape_92_shape_0 = const()[name = tensor<string, []>("reshape_92_shape_0"), val = tensor<int32, [5]>([2, 32, 40, 32, 32])];
            tensor<fp16, [2, 32, 40, 32, 32]> reshape_92_cast_fp16 = reshape(shape = reshape_92_shape_0, x = hidden_states_283_cast_fp16)[name = tensor<string, []>("reshape_92_cast_fp16")];
            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_fp16 = reduce_mean(axes = reduce_mean_69_axes_0, keep_dims = reduce_mean_69_keep_dims_0, x = reshape_92_cast_fp16)[name = tensor<string, []>("reduce_mean_69_cast_fp16")];
            tensor<fp16, [2, 32, 40, 32, 32]> sub_46_cast_fp16 = sub(x = reshape_92_cast_fp16, y = reduce_mean_69_cast_fp16)[name = tensor<string, []>("sub_46_cast_fp16")];
            tensor<fp16, [2, 32, 40, 32, 32]> square_23_cast_fp16 = square(x = sub_46_cast_fp16)[name = tensor<string, []>("square_23_cast_fp16")];
            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_fp16 = reduce_mean(axes = reduce_mean_71_axes_0, keep_dims = reduce_mean_71_keep_dims_0, x = square_23_cast_fp16)[name = tensor<string, []>("reduce_mean_71_cast_fp16")];
            tensor<fp16, []> add_46_y_0_to_fp16 = const()[name = tensor<string, []>("add_46_y_0_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_46_cast_fp16 = add(x = reduce_mean_71_cast_fp16, y = add_46_y_0_to_fp16)[name = tensor<string, []>("add_46_cast_fp16")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_23_cast_fp16 = sqrt(x = add_46_cast_fp16)[name = tensor<string, []>("sqrt_23_cast_fp16")];
            tensor<fp16, [2, 32, 40, 32, 32]> real_div_23_cast_fp16 = real_div(x = sub_46_cast_fp16, y = sqrt_23_cast_fp16)[name = tensor<string, []>("real_div_23_cast_fp16")];
            tensor<int32, [4]> reshape_93_shape_0 = const()[name = tensor<string, []>("reshape_93_shape_0"), val = tensor<int32, [4]>([2, 1280, 32, 32])];
            tensor<fp16, [2, 1280, 32, 32]> reshape_93_cast_fp16 = reshape(shape = reshape_93_shape_0, x = real_div_23_cast_fp16)[name = tensor<string, []>("reshape_93_cast_fp16")];
            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, []>(2480256)))];
            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, []>(2482880)))];
            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, 32, 32]> add_47_cast_fp16 = 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_23_mean_0_to_fp16, variance = add_23_variance_0_to_fp16, x = reshape_93_cast_fp16)[name = tensor<string, []>("add_47_cast_fp16")];
            tensor<string, []> hidden_states_285_pad_type_0 = const()[name = tensor<string, []>("hidden_states_285_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> hidden_states_285_strides_0 = const()[name = tensor<string, []>("hidden_states_285_strides_0"), val = tensor<int32, [2]>([1, 1])];
            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<int32, [2]> hidden_states_285_dilations_0 = const()[name = tensor<string, []>("hidden_states_285_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_285_groups_0 = const()[name = tensor<string, []>("hidden_states_285_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_proj_in_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2485504))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3714368))), name = tensor<string, []>("up_blocks_0_attentions_0_proj_in_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_proj_in_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_proj_in_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3714560)))];
            tensor<fp16, [2, 1280, 32, 32]> hidden_states_285_cast_fp16 = conv(bias = up_blocks_0_attentions_0_proj_in_bias_to_fp16, dilations = hidden_states_285_dilations_0, groups = hidden_states_285_groups_0, pad = hidden_states_285_pad_0, pad_type = hidden_states_285_pad_type_0, strides = hidden_states_285_strides_0, weight = up_blocks_0_attentions_0_proj_in_weight_to_fp16_palettized, x = add_47_cast_fp16)[name = tensor<string, []>("hidden_states_285_cast_fp16")];
            tensor<int32, [4]> var_30458 = const()[name = tensor<string, []>("op_30458"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_205_cast_fp16 = reshape(shape = var_30458, x = hidden_states_285_cast_fp16)[name = tensor<string, []>("inputs_205_cast_fp16")];
            tensor<int32, [1]> hidden_states_287_axes_0 = const()[name = tensor<string, []>("hidden_states_287_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_287_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_287_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3717184)))];
            tensor<fp16, [1280]> hidden_states_287_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_287_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3719808)))];
            tensor<fp16, []> var_30474_to_fp16 = const()[name = tensor<string, []>("op_30474_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_287_cast_fp16 = layer_norm(axes = hidden_states_287_axes_0, beta = hidden_states_287_beta_0_to_fp16, epsilon = var_30474_to_fp16, gamma = hidden_states_287_gamma_0_to_fp16, x = inputs_205_cast_fp16)[name = tensor<string, []>("hidden_states_287_cast_fp16")];
            tensor<string, []> q_137_pad_type_0 = const()[name = tensor<string, []>("q_137_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_137_strides_0 = const()[name = tensor<string, []>("q_137_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_137_pad_0 = const()[name = tensor<string, []>("q_137_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_137_dilations_0 = const()[name = tensor<string, []>("q_137_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_137_groups_0 = const()[name = tensor<string, []>("q_137_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3722432))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4951296))), name = tensor<string, []>("up_blocks_0_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, 1024]> q_137_cast_fp16 = conv(dilations = q_137_dilations_0, groups = q_137_groups_0, pad = q_137_pad_0, pad_type = q_137_pad_type_0, strides = q_137_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_287_cast_fp16)[name = tensor<string, []>("q_137_cast_fp16")];
            tensor<string, []> k_273_pad_type_0 = const()[name = tensor<string, []>("k_273_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_273_strides_0 = const()[name = tensor<string, []>("k_273_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_273_pad_0 = const()[name = tensor<string, []>("k_273_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_273_dilations_0 = const()[name = tensor<string, []>("k_273_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_273_groups_0 = const()[name = tensor<string, []>("k_273_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4951488))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6180352))), name = tensor<string, []>("up_blocks_0_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, 1024]> k_273_cast_fp16 = conv(dilations = k_273_dilations_0, groups = k_273_groups_0, pad = k_273_pad_0, pad_type = k_273_pad_type_0, strides = k_273_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_287_cast_fp16)[name = tensor<string, []>("k_273_cast_fp16")];
            tensor<string, []> v_137_pad_type_0 = const()[name = tensor<string, []>("v_137_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_137_strides_0 = const()[name = tensor<string, []>("v_137_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_137_pad_0 = const()[name = tensor<string, []>("v_137_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_137_dilations_0 = const()[name = tensor<string, []>("v_137_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_137_groups_0 = const()[name = tensor<string, []>("v_137_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6180544))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7409408))), name = tensor<string, []>("up_blocks_0_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, 1024]> v_137_cast_fp16 = conv(dilations = v_137_dilations_0, groups = v_137_groups_0, pad = v_137_pad_0, pad_type = v_137_pad_type_0, strides = v_137_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_287_cast_fp16)[name = tensor<string, []>("v_137_cast_fp16")];
            tensor<int32, [4]> var_30507_begin_0 = const()[name = tensor<string, []>("op_30507_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_30507_end_0 = const()[name = tensor<string, []>("op_30507_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_30507_end_mask_0 = const()[name = tensor<string, []>("op_30507_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30507_cast_fp16 = slice_by_index(begin = var_30507_begin_0, end = var_30507_end_0, end_mask = var_30507_end_mask_0, x = q_137_cast_fp16)[name = tensor<string, []>("op_30507_cast_fp16")];
            tensor<int32, [4]> var_30511_begin_0 = const()[name = tensor<string, []>("op_30511_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_30511_end_0 = const()[name = tensor<string, []>("op_30511_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_30511_end_mask_0 = const()[name = tensor<string, []>("op_30511_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30511_cast_fp16 = slice_by_index(begin = var_30511_begin_0, end = var_30511_end_0, end_mask = var_30511_end_mask_0, x = q_137_cast_fp16)[name = tensor<string, []>("op_30511_cast_fp16")];
            tensor<int32, [4]> var_30515_begin_0 = const()[name = tensor<string, []>("op_30515_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_30515_end_0 = const()[name = tensor<string, []>("op_30515_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_30515_end_mask_0 = const()[name = tensor<string, []>("op_30515_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30515_cast_fp16 = slice_by_index(begin = var_30515_begin_0, end = var_30515_end_0, end_mask = var_30515_end_mask_0, x = q_137_cast_fp16)[name = tensor<string, []>("op_30515_cast_fp16")];
            tensor<int32, [4]> var_30519_begin_0 = const()[name = tensor<string, []>("op_30519_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_30519_end_0 = const()[name = tensor<string, []>("op_30519_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_30519_end_mask_0 = const()[name = tensor<string, []>("op_30519_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30519_cast_fp16 = slice_by_index(begin = var_30519_begin_0, end = var_30519_end_0, end_mask = var_30519_end_mask_0, x = q_137_cast_fp16)[name = tensor<string, []>("op_30519_cast_fp16")];
            tensor<int32, [4]> var_30523_begin_0 = const()[name = tensor<string, []>("op_30523_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_30523_end_0 = const()[name = tensor<string, []>("op_30523_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_30523_end_mask_0 = const()[name = tensor<string, []>("op_30523_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30523_cast_fp16 = slice_by_index(begin = var_30523_begin_0, end = var_30523_end_0, end_mask = var_30523_end_mask_0, x = q_137_cast_fp16)[name = tensor<string, []>("op_30523_cast_fp16")];
            tensor<int32, [4]> var_30527_begin_0 = const()[name = tensor<string, []>("op_30527_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_30527_end_0 = const()[name = tensor<string, []>("op_30527_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_30527_end_mask_0 = const()[name = tensor<string, []>("op_30527_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30527_cast_fp16 = slice_by_index(begin = var_30527_begin_0, end = var_30527_end_0, end_mask = var_30527_end_mask_0, x = q_137_cast_fp16)[name = tensor<string, []>("op_30527_cast_fp16")];
            tensor<int32, [4]> var_30531_begin_0 = const()[name = tensor<string, []>("op_30531_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_30531_end_0 = const()[name = tensor<string, []>("op_30531_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_30531_end_mask_0 = const()[name = tensor<string, []>("op_30531_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30531_cast_fp16 = slice_by_index(begin = var_30531_begin_0, end = var_30531_end_0, end_mask = var_30531_end_mask_0, x = q_137_cast_fp16)[name = tensor<string, []>("op_30531_cast_fp16")];
            tensor<int32, [4]> var_30535_begin_0 = const()[name = tensor<string, []>("op_30535_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_30535_end_0 = const()[name = tensor<string, []>("op_30535_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_30535_end_mask_0 = const()[name = tensor<string, []>("op_30535_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30535_cast_fp16 = slice_by_index(begin = var_30535_begin_0, end = var_30535_end_0, end_mask = var_30535_end_mask_0, x = q_137_cast_fp16)[name = tensor<string, []>("op_30535_cast_fp16")];
            tensor<int32, [4]> var_30539_begin_0 = const()[name = tensor<string, []>("op_30539_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_30539_end_0 = const()[name = tensor<string, []>("op_30539_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_30539_end_mask_0 = const()[name = tensor<string, []>("op_30539_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30539_cast_fp16 = slice_by_index(begin = var_30539_begin_0, end = var_30539_end_0, end_mask = var_30539_end_mask_0, x = q_137_cast_fp16)[name = tensor<string, []>("op_30539_cast_fp16")];
            tensor<int32, [4]> var_30543_begin_0 = const()[name = tensor<string, []>("op_30543_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_30543_end_0 = const()[name = tensor<string, []>("op_30543_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_30543_end_mask_0 = const()[name = tensor<string, []>("op_30543_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30543_cast_fp16 = slice_by_index(begin = var_30543_begin_0, end = var_30543_end_0, end_mask = var_30543_end_mask_0, x = q_137_cast_fp16)[name = tensor<string, []>("op_30543_cast_fp16")];
            tensor<int32, [4]> var_30547_begin_0 = const()[name = tensor<string, []>("op_30547_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_30547_end_0 = const()[name = tensor<string, []>("op_30547_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_30547_end_mask_0 = const()[name = tensor<string, []>("op_30547_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30547_cast_fp16 = slice_by_index(begin = var_30547_begin_0, end = var_30547_end_0, end_mask = var_30547_end_mask_0, x = q_137_cast_fp16)[name = tensor<string, []>("op_30547_cast_fp16")];
            tensor<int32, [4]> var_30551_begin_0 = const()[name = tensor<string, []>("op_30551_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_30551_end_0 = const()[name = tensor<string, []>("op_30551_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_30551_end_mask_0 = const()[name = tensor<string, []>("op_30551_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30551_cast_fp16 = slice_by_index(begin = var_30551_begin_0, end = var_30551_end_0, end_mask = var_30551_end_mask_0, x = q_137_cast_fp16)[name = tensor<string, []>("op_30551_cast_fp16")];
            tensor<int32, [4]> var_30555_begin_0 = const()[name = tensor<string, []>("op_30555_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_30555_end_0 = const()[name = tensor<string, []>("op_30555_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_30555_end_mask_0 = const()[name = tensor<string, []>("op_30555_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30555_cast_fp16 = slice_by_index(begin = var_30555_begin_0, end = var_30555_end_0, end_mask = var_30555_end_mask_0, x = q_137_cast_fp16)[name = tensor<string, []>("op_30555_cast_fp16")];
            tensor<int32, [4]> var_30559_begin_0 = const()[name = tensor<string, []>("op_30559_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_30559_end_0 = const()[name = tensor<string, []>("op_30559_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_30559_end_mask_0 = const()[name = tensor<string, []>("op_30559_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30559_cast_fp16 = slice_by_index(begin = var_30559_begin_0, end = var_30559_end_0, end_mask = var_30559_end_mask_0, x = q_137_cast_fp16)[name = tensor<string, []>("op_30559_cast_fp16")];
            tensor<int32, [4]> var_30563_begin_0 = const()[name = tensor<string, []>("op_30563_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_30563_end_0 = const()[name = tensor<string, []>("op_30563_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_30563_end_mask_0 = const()[name = tensor<string, []>("op_30563_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30563_cast_fp16 = slice_by_index(begin = var_30563_begin_0, end = var_30563_end_0, end_mask = var_30563_end_mask_0, x = q_137_cast_fp16)[name = tensor<string, []>("op_30563_cast_fp16")];
            tensor<int32, [4]> var_30567_begin_0 = const()[name = tensor<string, []>("op_30567_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_30567_end_0 = const()[name = tensor<string, []>("op_30567_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_30567_end_mask_0 = const()[name = tensor<string, []>("op_30567_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30567_cast_fp16 = slice_by_index(begin = var_30567_begin_0, end = var_30567_end_0, end_mask = var_30567_end_mask_0, x = q_137_cast_fp16)[name = tensor<string, []>("op_30567_cast_fp16")];
            tensor<int32, [4]> var_30571_begin_0 = const()[name = tensor<string, []>("op_30571_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_30571_end_0 = const()[name = tensor<string, []>("op_30571_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_30571_end_mask_0 = const()[name = tensor<string, []>("op_30571_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30571_cast_fp16 = slice_by_index(begin = var_30571_begin_0, end = var_30571_end_0, end_mask = var_30571_end_mask_0, x = q_137_cast_fp16)[name = tensor<string, []>("op_30571_cast_fp16")];
            tensor<int32, [4]> var_30575_begin_0 = const()[name = tensor<string, []>("op_30575_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_30575_end_0 = const()[name = tensor<string, []>("op_30575_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_30575_end_mask_0 = const()[name = tensor<string, []>("op_30575_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30575_cast_fp16 = slice_by_index(begin = var_30575_begin_0, end = var_30575_end_0, end_mask = var_30575_end_mask_0, x = q_137_cast_fp16)[name = tensor<string, []>("op_30575_cast_fp16")];
            tensor<int32, [4]> var_30579_begin_0 = const()[name = tensor<string, []>("op_30579_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_30579_end_0 = const()[name = tensor<string, []>("op_30579_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_30579_end_mask_0 = const()[name = tensor<string, []>("op_30579_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30579_cast_fp16 = slice_by_index(begin = var_30579_begin_0, end = var_30579_end_0, end_mask = var_30579_end_mask_0, x = q_137_cast_fp16)[name = tensor<string, []>("op_30579_cast_fp16")];
            tensor<int32, [4]> var_30583_begin_0 = const()[name = tensor<string, []>("op_30583_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_30583_end_0 = const()[name = tensor<string, []>("op_30583_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_30583_end_mask_0 = const()[name = tensor<string, []>("op_30583_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30583_cast_fp16 = slice_by_index(begin = var_30583_begin_0, end = var_30583_end_0, end_mask = var_30583_end_mask_0, x = q_137_cast_fp16)[name = tensor<string, []>("op_30583_cast_fp16")];
            tensor<int32, [4]> k_275_perm_0 = const()[name = tensor<string, []>("k_275_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_30590_begin_0 = const()[name = tensor<string, []>("op_30590_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_30590_end_0 = const()[name = tensor<string, []>("op_30590_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_30590_end_mask_0 = const()[name = tensor<string, []>("op_30590_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> k_275_cast_fp16 = transpose(perm = k_275_perm_0, x = k_273_cast_fp16)[name = tensor<string, []>("transpose_71")];
            tensor<fp16, [2, 1024, 1, 64]> var_30590_cast_fp16 = slice_by_index(begin = var_30590_begin_0, end = var_30590_end_0, end_mask = var_30590_end_mask_0, x = k_275_cast_fp16)[name = tensor<string, []>("op_30590_cast_fp16")];
            tensor<int32, [4]> var_30594_begin_0 = const()[name = tensor<string, []>("op_30594_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_30594_end_0 = const()[name = tensor<string, []>("op_30594_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_30594_end_mask_0 = const()[name = tensor<string, []>("op_30594_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_30594_cast_fp16 = slice_by_index(begin = var_30594_begin_0, end = var_30594_end_0, end_mask = var_30594_end_mask_0, x = k_275_cast_fp16)[name = tensor<string, []>("op_30594_cast_fp16")];
            tensor<int32, [4]> var_30598_begin_0 = const()[name = tensor<string, []>("op_30598_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_30598_end_0 = const()[name = tensor<string, []>("op_30598_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_30598_end_mask_0 = const()[name = tensor<string, []>("op_30598_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_30598_cast_fp16 = slice_by_index(begin = var_30598_begin_0, end = var_30598_end_0, end_mask = var_30598_end_mask_0, x = k_275_cast_fp16)[name = tensor<string, []>("op_30598_cast_fp16")];
            tensor<int32, [4]> var_30602_begin_0 = const()[name = tensor<string, []>("op_30602_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_30602_end_0 = const()[name = tensor<string, []>("op_30602_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_30602_end_mask_0 = const()[name = tensor<string, []>("op_30602_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_30602_cast_fp16 = slice_by_index(begin = var_30602_begin_0, end = var_30602_end_0, end_mask = var_30602_end_mask_0, x = k_275_cast_fp16)[name = tensor<string, []>("op_30602_cast_fp16")];
            tensor<int32, [4]> var_30606_begin_0 = const()[name = tensor<string, []>("op_30606_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_30606_end_0 = const()[name = tensor<string, []>("op_30606_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_30606_end_mask_0 = const()[name = tensor<string, []>("op_30606_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_30606_cast_fp16 = slice_by_index(begin = var_30606_begin_0, end = var_30606_end_0, end_mask = var_30606_end_mask_0, x = k_275_cast_fp16)[name = tensor<string, []>("op_30606_cast_fp16")];
            tensor<int32, [4]> var_30610_begin_0 = const()[name = tensor<string, []>("op_30610_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_30610_end_0 = const()[name = tensor<string, []>("op_30610_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_30610_end_mask_0 = const()[name = tensor<string, []>("op_30610_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_30610_cast_fp16 = slice_by_index(begin = var_30610_begin_0, end = var_30610_end_0, end_mask = var_30610_end_mask_0, x = k_275_cast_fp16)[name = tensor<string, []>("op_30610_cast_fp16")];
            tensor<int32, [4]> var_30614_begin_0 = const()[name = tensor<string, []>("op_30614_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_30614_end_0 = const()[name = tensor<string, []>("op_30614_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_30614_end_mask_0 = const()[name = tensor<string, []>("op_30614_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_30614_cast_fp16 = slice_by_index(begin = var_30614_begin_0, end = var_30614_end_0, end_mask = var_30614_end_mask_0, x = k_275_cast_fp16)[name = tensor<string, []>("op_30614_cast_fp16")];
            tensor<int32, [4]> var_30618_begin_0 = const()[name = tensor<string, []>("op_30618_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_30618_end_0 = const()[name = tensor<string, []>("op_30618_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_30618_end_mask_0 = const()[name = tensor<string, []>("op_30618_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_30618_cast_fp16 = slice_by_index(begin = var_30618_begin_0, end = var_30618_end_0, end_mask = var_30618_end_mask_0, x = k_275_cast_fp16)[name = tensor<string, []>("op_30618_cast_fp16")];
            tensor<int32, [4]> var_30622_begin_0 = const()[name = tensor<string, []>("op_30622_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_30622_end_0 = const()[name = tensor<string, []>("op_30622_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_30622_end_mask_0 = const()[name = tensor<string, []>("op_30622_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_30622_cast_fp16 = slice_by_index(begin = var_30622_begin_0, end = var_30622_end_0, end_mask = var_30622_end_mask_0, x = k_275_cast_fp16)[name = tensor<string, []>("op_30622_cast_fp16")];
            tensor<int32, [4]> var_30626_begin_0 = const()[name = tensor<string, []>("op_30626_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_30626_end_0 = const()[name = tensor<string, []>("op_30626_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_30626_end_mask_0 = const()[name = tensor<string, []>("op_30626_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_30626_cast_fp16 = slice_by_index(begin = var_30626_begin_0, end = var_30626_end_0, end_mask = var_30626_end_mask_0, x = k_275_cast_fp16)[name = tensor<string, []>("op_30626_cast_fp16")];
            tensor<int32, [4]> var_30630_begin_0 = const()[name = tensor<string, []>("op_30630_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_30630_end_0 = const()[name = tensor<string, []>("op_30630_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_30630_end_mask_0 = const()[name = tensor<string, []>("op_30630_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_30630_cast_fp16 = slice_by_index(begin = var_30630_begin_0, end = var_30630_end_0, end_mask = var_30630_end_mask_0, x = k_275_cast_fp16)[name = tensor<string, []>("op_30630_cast_fp16")];
            tensor<int32, [4]> var_30634_begin_0 = const()[name = tensor<string, []>("op_30634_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_30634_end_0 = const()[name = tensor<string, []>("op_30634_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_30634_end_mask_0 = const()[name = tensor<string, []>("op_30634_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_30634_cast_fp16 = slice_by_index(begin = var_30634_begin_0, end = var_30634_end_0, end_mask = var_30634_end_mask_0, x = k_275_cast_fp16)[name = tensor<string, []>("op_30634_cast_fp16")];
            tensor<int32, [4]> var_30638_begin_0 = const()[name = tensor<string, []>("op_30638_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_30638_end_0 = const()[name = tensor<string, []>("op_30638_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_30638_end_mask_0 = const()[name = tensor<string, []>("op_30638_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_30638_cast_fp16 = slice_by_index(begin = var_30638_begin_0, end = var_30638_end_0, end_mask = var_30638_end_mask_0, x = k_275_cast_fp16)[name = tensor<string, []>("op_30638_cast_fp16")];
            tensor<int32, [4]> var_30642_begin_0 = const()[name = tensor<string, []>("op_30642_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_30642_end_0 = const()[name = tensor<string, []>("op_30642_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_30642_end_mask_0 = const()[name = tensor<string, []>("op_30642_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_30642_cast_fp16 = slice_by_index(begin = var_30642_begin_0, end = var_30642_end_0, end_mask = var_30642_end_mask_0, x = k_275_cast_fp16)[name = tensor<string, []>("op_30642_cast_fp16")];
            tensor<int32, [4]> var_30646_begin_0 = const()[name = tensor<string, []>("op_30646_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_30646_end_0 = const()[name = tensor<string, []>("op_30646_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_30646_end_mask_0 = const()[name = tensor<string, []>("op_30646_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_30646_cast_fp16 = slice_by_index(begin = var_30646_begin_0, end = var_30646_end_0, end_mask = var_30646_end_mask_0, x = k_275_cast_fp16)[name = tensor<string, []>("op_30646_cast_fp16")];
            tensor<int32, [4]> var_30650_begin_0 = const()[name = tensor<string, []>("op_30650_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_30650_end_0 = const()[name = tensor<string, []>("op_30650_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_30650_end_mask_0 = const()[name = tensor<string, []>("op_30650_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_30650_cast_fp16 = slice_by_index(begin = var_30650_begin_0, end = var_30650_end_0, end_mask = var_30650_end_mask_0, x = k_275_cast_fp16)[name = tensor<string, []>("op_30650_cast_fp16")];
            tensor<int32, [4]> var_30654_begin_0 = const()[name = tensor<string, []>("op_30654_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_30654_end_0 = const()[name = tensor<string, []>("op_30654_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_30654_end_mask_0 = const()[name = tensor<string, []>("op_30654_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_30654_cast_fp16 = slice_by_index(begin = var_30654_begin_0, end = var_30654_end_0, end_mask = var_30654_end_mask_0, x = k_275_cast_fp16)[name = tensor<string, []>("op_30654_cast_fp16")];
            tensor<int32, [4]> var_30658_begin_0 = const()[name = tensor<string, []>("op_30658_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_30658_end_0 = const()[name = tensor<string, []>("op_30658_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_30658_end_mask_0 = const()[name = tensor<string, []>("op_30658_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_30658_cast_fp16 = slice_by_index(begin = var_30658_begin_0, end = var_30658_end_0, end_mask = var_30658_end_mask_0, x = k_275_cast_fp16)[name = tensor<string, []>("op_30658_cast_fp16")];
            tensor<int32, [4]> var_30662_begin_0 = const()[name = tensor<string, []>("op_30662_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_30662_end_0 = const()[name = tensor<string, []>("op_30662_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_30662_end_mask_0 = const()[name = tensor<string, []>("op_30662_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_30662_cast_fp16 = slice_by_index(begin = var_30662_begin_0, end = var_30662_end_0, end_mask = var_30662_end_mask_0, x = k_275_cast_fp16)[name = tensor<string, []>("op_30662_cast_fp16")];
            tensor<int32, [4]> var_30666_begin_0 = const()[name = tensor<string, []>("op_30666_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_30666_end_0 = const()[name = tensor<string, []>("op_30666_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_30666_end_mask_0 = const()[name = tensor<string, []>("op_30666_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_30666_cast_fp16 = slice_by_index(begin = var_30666_begin_0, end = var_30666_end_0, end_mask = var_30666_end_mask_0, x = k_275_cast_fp16)[name = tensor<string, []>("op_30666_cast_fp16")];
            tensor<int32, [4]> var_30668_begin_0 = const()[name = tensor<string, []>("op_30668_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_30668_end_0 = const()[name = tensor<string, []>("op_30668_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_30668_end_mask_0 = const()[name = tensor<string, []>("op_30668_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30668_cast_fp16 = slice_by_index(begin = var_30668_begin_0, end = var_30668_end_0, end_mask = var_30668_end_mask_0, x = v_137_cast_fp16)[name = tensor<string, []>("op_30668_cast_fp16")];
            tensor<int32, [4]> var_30672_begin_0 = const()[name = tensor<string, []>("op_30672_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_30672_end_0 = const()[name = tensor<string, []>("op_30672_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_30672_end_mask_0 = const()[name = tensor<string, []>("op_30672_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30672_cast_fp16 = slice_by_index(begin = var_30672_begin_0, end = var_30672_end_0, end_mask = var_30672_end_mask_0, x = v_137_cast_fp16)[name = tensor<string, []>("op_30672_cast_fp16")];
            tensor<int32, [4]> var_30676_begin_0 = const()[name = tensor<string, []>("op_30676_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_30676_end_0 = const()[name = tensor<string, []>("op_30676_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_30676_end_mask_0 = const()[name = tensor<string, []>("op_30676_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30676_cast_fp16 = slice_by_index(begin = var_30676_begin_0, end = var_30676_end_0, end_mask = var_30676_end_mask_0, x = v_137_cast_fp16)[name = tensor<string, []>("op_30676_cast_fp16")];
            tensor<int32, [4]> var_30680_begin_0 = const()[name = tensor<string, []>("op_30680_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_30680_end_0 = const()[name = tensor<string, []>("op_30680_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_30680_end_mask_0 = const()[name = tensor<string, []>("op_30680_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30680_cast_fp16 = slice_by_index(begin = var_30680_begin_0, end = var_30680_end_0, end_mask = var_30680_end_mask_0, x = v_137_cast_fp16)[name = tensor<string, []>("op_30680_cast_fp16")];
            tensor<int32, [4]> var_30684_begin_0 = const()[name = tensor<string, []>("op_30684_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_30684_end_0 = const()[name = tensor<string, []>("op_30684_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_30684_end_mask_0 = const()[name = tensor<string, []>("op_30684_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30684_cast_fp16 = slice_by_index(begin = var_30684_begin_0, end = var_30684_end_0, end_mask = var_30684_end_mask_0, x = v_137_cast_fp16)[name = tensor<string, []>("op_30684_cast_fp16")];
            tensor<int32, [4]> var_30688_begin_0 = const()[name = tensor<string, []>("op_30688_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_30688_end_0 = const()[name = tensor<string, []>("op_30688_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_30688_end_mask_0 = const()[name = tensor<string, []>("op_30688_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30688_cast_fp16 = slice_by_index(begin = var_30688_begin_0, end = var_30688_end_0, end_mask = var_30688_end_mask_0, x = v_137_cast_fp16)[name = tensor<string, []>("op_30688_cast_fp16")];
            tensor<int32, [4]> var_30692_begin_0 = const()[name = tensor<string, []>("op_30692_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_30692_end_0 = const()[name = tensor<string, []>("op_30692_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_30692_end_mask_0 = const()[name = tensor<string, []>("op_30692_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30692_cast_fp16 = slice_by_index(begin = var_30692_begin_0, end = var_30692_end_0, end_mask = var_30692_end_mask_0, x = v_137_cast_fp16)[name = tensor<string, []>("op_30692_cast_fp16")];
            tensor<int32, [4]> var_30696_begin_0 = const()[name = tensor<string, []>("op_30696_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_30696_end_0 = const()[name = tensor<string, []>("op_30696_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_30696_end_mask_0 = const()[name = tensor<string, []>("op_30696_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30696_cast_fp16 = slice_by_index(begin = var_30696_begin_0, end = var_30696_end_0, end_mask = var_30696_end_mask_0, x = v_137_cast_fp16)[name = tensor<string, []>("op_30696_cast_fp16")];
            tensor<int32, [4]> var_30700_begin_0 = const()[name = tensor<string, []>("op_30700_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_30700_end_0 = const()[name = tensor<string, []>("op_30700_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_30700_end_mask_0 = const()[name = tensor<string, []>("op_30700_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30700_cast_fp16 = slice_by_index(begin = var_30700_begin_0, end = var_30700_end_0, end_mask = var_30700_end_mask_0, x = v_137_cast_fp16)[name = tensor<string, []>("op_30700_cast_fp16")];
            tensor<int32, [4]> var_30704_begin_0 = const()[name = tensor<string, []>("op_30704_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_30704_end_0 = const()[name = tensor<string, []>("op_30704_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_30704_end_mask_0 = const()[name = tensor<string, []>("op_30704_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30704_cast_fp16 = slice_by_index(begin = var_30704_begin_0, end = var_30704_end_0, end_mask = var_30704_end_mask_0, x = v_137_cast_fp16)[name = tensor<string, []>("op_30704_cast_fp16")];
            tensor<int32, [4]> var_30708_begin_0 = const()[name = tensor<string, []>("op_30708_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_30708_end_0 = const()[name = tensor<string, []>("op_30708_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_30708_end_mask_0 = const()[name = tensor<string, []>("op_30708_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30708_cast_fp16 = slice_by_index(begin = var_30708_begin_0, end = var_30708_end_0, end_mask = var_30708_end_mask_0, x = v_137_cast_fp16)[name = tensor<string, []>("op_30708_cast_fp16")];
            tensor<int32, [4]> var_30712_begin_0 = const()[name = tensor<string, []>("op_30712_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_30712_end_0 = const()[name = tensor<string, []>("op_30712_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_30712_end_mask_0 = const()[name = tensor<string, []>("op_30712_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30712_cast_fp16 = slice_by_index(begin = var_30712_begin_0, end = var_30712_end_0, end_mask = var_30712_end_mask_0, x = v_137_cast_fp16)[name = tensor<string, []>("op_30712_cast_fp16")];
            tensor<int32, [4]> var_30716_begin_0 = const()[name = tensor<string, []>("op_30716_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_30716_end_0 = const()[name = tensor<string, []>("op_30716_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_30716_end_mask_0 = const()[name = tensor<string, []>("op_30716_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30716_cast_fp16 = slice_by_index(begin = var_30716_begin_0, end = var_30716_end_0, end_mask = var_30716_end_mask_0, x = v_137_cast_fp16)[name = tensor<string, []>("op_30716_cast_fp16")];
            tensor<int32, [4]> var_30720_begin_0 = const()[name = tensor<string, []>("op_30720_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_30720_end_0 = const()[name = tensor<string, []>("op_30720_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_30720_end_mask_0 = const()[name = tensor<string, []>("op_30720_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30720_cast_fp16 = slice_by_index(begin = var_30720_begin_0, end = var_30720_end_0, end_mask = var_30720_end_mask_0, x = v_137_cast_fp16)[name = tensor<string, []>("op_30720_cast_fp16")];
            tensor<int32, [4]> var_30724_begin_0 = const()[name = tensor<string, []>("op_30724_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_30724_end_0 = const()[name = tensor<string, []>("op_30724_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_30724_end_mask_0 = const()[name = tensor<string, []>("op_30724_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30724_cast_fp16 = slice_by_index(begin = var_30724_begin_0, end = var_30724_end_0, end_mask = var_30724_end_mask_0, x = v_137_cast_fp16)[name = tensor<string, []>("op_30724_cast_fp16")];
            tensor<int32, [4]> var_30728_begin_0 = const()[name = tensor<string, []>("op_30728_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_30728_end_0 = const()[name = tensor<string, []>("op_30728_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_30728_end_mask_0 = const()[name = tensor<string, []>("op_30728_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30728_cast_fp16 = slice_by_index(begin = var_30728_begin_0, end = var_30728_end_0, end_mask = var_30728_end_mask_0, x = v_137_cast_fp16)[name = tensor<string, []>("op_30728_cast_fp16")];
            tensor<int32, [4]> var_30732_begin_0 = const()[name = tensor<string, []>("op_30732_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_30732_end_0 = const()[name = tensor<string, []>("op_30732_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_30732_end_mask_0 = const()[name = tensor<string, []>("op_30732_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30732_cast_fp16 = slice_by_index(begin = var_30732_begin_0, end = var_30732_end_0, end_mask = var_30732_end_mask_0, x = v_137_cast_fp16)[name = tensor<string, []>("op_30732_cast_fp16")];
            tensor<int32, [4]> var_30736_begin_0 = const()[name = tensor<string, []>("op_30736_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_30736_end_0 = const()[name = tensor<string, []>("op_30736_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_30736_end_mask_0 = const()[name = tensor<string, []>("op_30736_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30736_cast_fp16 = slice_by_index(begin = var_30736_begin_0, end = var_30736_end_0, end_mask = var_30736_end_mask_0, x = v_137_cast_fp16)[name = tensor<string, []>("op_30736_cast_fp16")];
            tensor<int32, [4]> var_30740_begin_0 = const()[name = tensor<string, []>("op_30740_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_30740_end_0 = const()[name = tensor<string, []>("op_30740_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_30740_end_mask_0 = const()[name = tensor<string, []>("op_30740_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30740_cast_fp16 = slice_by_index(begin = var_30740_begin_0, end = var_30740_end_0, end_mask = var_30740_end_mask_0, x = v_137_cast_fp16)[name = tensor<string, []>("op_30740_cast_fp16")];
            tensor<int32, [4]> var_30744_begin_0 = const()[name = tensor<string, []>("op_30744_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_30744_end_0 = const()[name = tensor<string, []>("op_30744_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_30744_end_mask_0 = const()[name = tensor<string, []>("op_30744_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30744_cast_fp16 = slice_by_index(begin = var_30744_begin_0, end = var_30744_end_0, end_mask = var_30744_end_mask_0, x = v_137_cast_fp16)[name = tensor<string, []>("op_30744_cast_fp16")];
            tensor<string, []> var_30748_equation_0 = const()[name = tensor<string, []>("op_30748_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30748_cast_fp16 = einsum(equation = var_30748_equation_0, values = (var_30590_cast_fp16, var_30507_cast_fp16))[name = tensor<string, []>("op_30748_cast_fp16")];
            tensor<fp16, []> var_30749_to_fp16 = const()[name = tensor<string, []>("op_30749_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2561_cast_fp16 = mul(x = var_30748_cast_fp16, y = var_30749_to_fp16)[name = tensor<string, []>("aw_2561_cast_fp16")];
            tensor<string, []> var_30752_equation_0 = const()[name = tensor<string, []>("op_30752_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30752_cast_fp16 = einsum(equation = var_30752_equation_0, values = (var_30594_cast_fp16, var_30511_cast_fp16))[name = tensor<string, []>("op_30752_cast_fp16")];
            tensor<fp16, []> var_30753_to_fp16 = const()[name = tensor<string, []>("op_30753_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2563_cast_fp16 = mul(x = var_30752_cast_fp16, y = var_30753_to_fp16)[name = tensor<string, []>("aw_2563_cast_fp16")];
            tensor<string, []> var_30756_equation_0 = const()[name = tensor<string, []>("op_30756_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30756_cast_fp16 = einsum(equation = var_30756_equation_0, values = (var_30598_cast_fp16, var_30515_cast_fp16))[name = tensor<string, []>("op_30756_cast_fp16")];
            tensor<fp16, []> var_30757_to_fp16 = const()[name = tensor<string, []>("op_30757_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2565_cast_fp16 = mul(x = var_30756_cast_fp16, y = var_30757_to_fp16)[name = tensor<string, []>("aw_2565_cast_fp16")];
            tensor<string, []> var_30760_equation_0 = const()[name = tensor<string, []>("op_30760_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30760_cast_fp16 = einsum(equation = var_30760_equation_0, values = (var_30602_cast_fp16, var_30519_cast_fp16))[name = tensor<string, []>("op_30760_cast_fp16")];
            tensor<fp16, []> var_30761_to_fp16 = const()[name = tensor<string, []>("op_30761_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2567_cast_fp16 = mul(x = var_30760_cast_fp16, y = var_30761_to_fp16)[name = tensor<string, []>("aw_2567_cast_fp16")];
            tensor<string, []> var_30764_equation_0 = const()[name = tensor<string, []>("op_30764_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30764_cast_fp16 = einsum(equation = var_30764_equation_0, values = (var_30606_cast_fp16, var_30523_cast_fp16))[name = tensor<string, []>("op_30764_cast_fp16")];
            tensor<fp16, []> var_30765_to_fp16 = const()[name = tensor<string, []>("op_30765_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2569_cast_fp16 = mul(x = var_30764_cast_fp16, y = var_30765_to_fp16)[name = tensor<string, []>("aw_2569_cast_fp16")];
            tensor<string, []> var_30768_equation_0 = const()[name = tensor<string, []>("op_30768_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30768_cast_fp16 = einsum(equation = var_30768_equation_0, values = (var_30610_cast_fp16, var_30527_cast_fp16))[name = tensor<string, []>("op_30768_cast_fp16")];
            tensor<fp16, []> var_30769_to_fp16 = const()[name = tensor<string, []>("op_30769_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2571_cast_fp16 = mul(x = var_30768_cast_fp16, y = var_30769_to_fp16)[name = tensor<string, []>("aw_2571_cast_fp16")];
            tensor<string, []> var_30772_equation_0 = const()[name = tensor<string, []>("op_30772_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30772_cast_fp16 = einsum(equation = var_30772_equation_0, values = (var_30614_cast_fp16, var_30531_cast_fp16))[name = tensor<string, []>("op_30772_cast_fp16")];
            tensor<fp16, []> var_30773_to_fp16 = const()[name = tensor<string, []>("op_30773_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2573_cast_fp16 = mul(x = var_30772_cast_fp16, y = var_30773_to_fp16)[name = tensor<string, []>("aw_2573_cast_fp16")];
            tensor<string, []> var_30776_equation_0 = const()[name = tensor<string, []>("op_30776_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30776_cast_fp16 = einsum(equation = var_30776_equation_0, values = (var_30618_cast_fp16, var_30535_cast_fp16))[name = tensor<string, []>("op_30776_cast_fp16")];
            tensor<fp16, []> var_30777_to_fp16 = const()[name = tensor<string, []>("op_30777_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2575_cast_fp16 = mul(x = var_30776_cast_fp16, y = var_30777_to_fp16)[name = tensor<string, []>("aw_2575_cast_fp16")];
            tensor<string, []> var_30780_equation_0 = const()[name = tensor<string, []>("op_30780_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30780_cast_fp16 = einsum(equation = var_30780_equation_0, values = (var_30622_cast_fp16, var_30539_cast_fp16))[name = tensor<string, []>("op_30780_cast_fp16")];
            tensor<fp16, []> var_30781_to_fp16 = const()[name = tensor<string, []>("op_30781_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2577_cast_fp16 = mul(x = var_30780_cast_fp16, y = var_30781_to_fp16)[name = tensor<string, []>("aw_2577_cast_fp16")];
            tensor<string, []> var_30784_equation_0 = const()[name = tensor<string, []>("op_30784_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30784_cast_fp16 = einsum(equation = var_30784_equation_0, values = (var_30626_cast_fp16, var_30543_cast_fp16))[name = tensor<string, []>("op_30784_cast_fp16")];
            tensor<fp16, []> var_30785_to_fp16 = const()[name = tensor<string, []>("op_30785_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2579_cast_fp16 = mul(x = var_30784_cast_fp16, y = var_30785_to_fp16)[name = tensor<string, []>("aw_2579_cast_fp16")];
            tensor<string, []> var_30788_equation_0 = const()[name = tensor<string, []>("op_30788_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30788_cast_fp16 = einsum(equation = var_30788_equation_0, values = (var_30630_cast_fp16, var_30547_cast_fp16))[name = tensor<string, []>("op_30788_cast_fp16")];
            tensor<fp16, []> var_30789_to_fp16 = const()[name = tensor<string, []>("op_30789_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2581_cast_fp16 = mul(x = var_30788_cast_fp16, y = var_30789_to_fp16)[name = tensor<string, []>("aw_2581_cast_fp16")];
            tensor<string, []> var_30792_equation_0 = const()[name = tensor<string, []>("op_30792_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30792_cast_fp16 = einsum(equation = var_30792_equation_0, values = (var_30634_cast_fp16, var_30551_cast_fp16))[name = tensor<string, []>("op_30792_cast_fp16")];
            tensor<fp16, []> var_30793_to_fp16 = const()[name = tensor<string, []>("op_30793_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2583_cast_fp16 = mul(x = var_30792_cast_fp16, y = var_30793_to_fp16)[name = tensor<string, []>("aw_2583_cast_fp16")];
            tensor<string, []> var_30796_equation_0 = const()[name = tensor<string, []>("op_30796_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30796_cast_fp16 = einsum(equation = var_30796_equation_0, values = (var_30638_cast_fp16, var_30555_cast_fp16))[name = tensor<string, []>("op_30796_cast_fp16")];
            tensor<fp16, []> var_30797_to_fp16 = const()[name = tensor<string, []>("op_30797_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2585_cast_fp16 = mul(x = var_30796_cast_fp16, y = var_30797_to_fp16)[name = tensor<string, []>("aw_2585_cast_fp16")];
            tensor<string, []> var_30800_equation_0 = const()[name = tensor<string, []>("op_30800_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30800_cast_fp16 = einsum(equation = var_30800_equation_0, values = (var_30642_cast_fp16, var_30559_cast_fp16))[name = tensor<string, []>("op_30800_cast_fp16")];
            tensor<fp16, []> var_30801_to_fp16 = const()[name = tensor<string, []>("op_30801_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2587_cast_fp16 = mul(x = var_30800_cast_fp16, y = var_30801_to_fp16)[name = tensor<string, []>("aw_2587_cast_fp16")];
            tensor<string, []> var_30804_equation_0 = const()[name = tensor<string, []>("op_30804_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30804_cast_fp16 = einsum(equation = var_30804_equation_0, values = (var_30646_cast_fp16, var_30563_cast_fp16))[name = tensor<string, []>("op_30804_cast_fp16")];
            tensor<fp16, []> var_30805_to_fp16 = const()[name = tensor<string, []>("op_30805_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2589_cast_fp16 = mul(x = var_30804_cast_fp16, y = var_30805_to_fp16)[name = tensor<string, []>("aw_2589_cast_fp16")];
            tensor<string, []> var_30808_equation_0 = const()[name = tensor<string, []>("op_30808_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30808_cast_fp16 = einsum(equation = var_30808_equation_0, values = (var_30650_cast_fp16, var_30567_cast_fp16))[name = tensor<string, []>("op_30808_cast_fp16")];
            tensor<fp16, []> var_30809_to_fp16 = const()[name = tensor<string, []>("op_30809_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2591_cast_fp16 = mul(x = var_30808_cast_fp16, y = var_30809_to_fp16)[name = tensor<string, []>("aw_2591_cast_fp16")];
            tensor<string, []> var_30812_equation_0 = const()[name = tensor<string, []>("op_30812_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30812_cast_fp16 = einsum(equation = var_30812_equation_0, values = (var_30654_cast_fp16, var_30571_cast_fp16))[name = tensor<string, []>("op_30812_cast_fp16")];
            tensor<fp16, []> var_30813_to_fp16 = const()[name = tensor<string, []>("op_30813_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2593_cast_fp16 = mul(x = var_30812_cast_fp16, y = var_30813_to_fp16)[name = tensor<string, []>("aw_2593_cast_fp16")];
            tensor<string, []> var_30816_equation_0 = const()[name = tensor<string, []>("op_30816_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30816_cast_fp16 = einsum(equation = var_30816_equation_0, values = (var_30658_cast_fp16, var_30575_cast_fp16))[name = tensor<string, []>("op_30816_cast_fp16")];
            tensor<fp16, []> var_30817_to_fp16 = const()[name = tensor<string, []>("op_30817_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2595_cast_fp16 = mul(x = var_30816_cast_fp16, y = var_30817_to_fp16)[name = tensor<string, []>("aw_2595_cast_fp16")];
            tensor<string, []> var_30820_equation_0 = const()[name = tensor<string, []>("op_30820_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30820_cast_fp16 = einsum(equation = var_30820_equation_0, values = (var_30662_cast_fp16, var_30579_cast_fp16))[name = tensor<string, []>("op_30820_cast_fp16")];
            tensor<fp16, []> var_30821_to_fp16 = const()[name = tensor<string, []>("op_30821_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2597_cast_fp16 = mul(x = var_30820_cast_fp16, y = var_30821_to_fp16)[name = tensor<string, []>("aw_2597_cast_fp16")];
            tensor<string, []> var_30824_equation_0 = const()[name = tensor<string, []>("op_30824_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30824_cast_fp16 = einsum(equation = var_30824_equation_0, values = (var_30666_cast_fp16, var_30583_cast_fp16))[name = tensor<string, []>("op_30824_cast_fp16")];
            tensor<fp16, []> var_30825_to_fp16 = const()[name = tensor<string, []>("op_30825_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2599_cast_fp16 = mul(x = var_30824_cast_fp16, y = var_30825_to_fp16)[name = tensor<string, []>("aw_2599_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30827_cast_fp16 = softmax(axis = var_30355, x = aw_2561_cast_fp16)[name = tensor<string, []>("op_30827_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30828_cast_fp16 = softmax(axis = var_30355, x = aw_2563_cast_fp16)[name = tensor<string, []>("op_30828_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30829_cast_fp16 = softmax(axis = var_30355, x = aw_2565_cast_fp16)[name = tensor<string, []>("op_30829_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30830_cast_fp16 = softmax(axis = var_30355, x = aw_2567_cast_fp16)[name = tensor<string, []>("op_30830_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30831_cast_fp16 = softmax(axis = var_30355, x = aw_2569_cast_fp16)[name = tensor<string, []>("op_30831_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30832_cast_fp16 = softmax(axis = var_30355, x = aw_2571_cast_fp16)[name = tensor<string, []>("op_30832_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30833_cast_fp16 = softmax(axis = var_30355, x = aw_2573_cast_fp16)[name = tensor<string, []>("op_30833_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30834_cast_fp16 = softmax(axis = var_30355, x = aw_2575_cast_fp16)[name = tensor<string, []>("op_30834_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30835_cast_fp16 = softmax(axis = var_30355, x = aw_2577_cast_fp16)[name = tensor<string, []>("op_30835_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30836_cast_fp16 = softmax(axis = var_30355, x = aw_2579_cast_fp16)[name = tensor<string, []>("op_30836_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30837_cast_fp16 = softmax(axis = var_30355, x = aw_2581_cast_fp16)[name = tensor<string, []>("op_30837_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30838_cast_fp16 = softmax(axis = var_30355, x = aw_2583_cast_fp16)[name = tensor<string, []>("op_30838_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30839_cast_fp16 = softmax(axis = var_30355, x = aw_2585_cast_fp16)[name = tensor<string, []>("op_30839_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30840_cast_fp16 = softmax(axis = var_30355, x = aw_2587_cast_fp16)[name = tensor<string, []>("op_30840_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30841_cast_fp16 = softmax(axis = var_30355, x = aw_2589_cast_fp16)[name = tensor<string, []>("op_30841_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30842_cast_fp16 = softmax(axis = var_30355, x = aw_2591_cast_fp16)[name = tensor<string, []>("op_30842_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30843_cast_fp16 = softmax(axis = var_30355, x = aw_2593_cast_fp16)[name = tensor<string, []>("op_30843_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30844_cast_fp16 = softmax(axis = var_30355, x = aw_2595_cast_fp16)[name = tensor<string, []>("op_30844_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30845_cast_fp16 = softmax(axis = var_30355, x = aw_2597_cast_fp16)[name = tensor<string, []>("op_30845_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30846_cast_fp16 = softmax(axis = var_30355, x = aw_2599_cast_fp16)[name = tensor<string, []>("op_30846_cast_fp16")];
            tensor<string, []> var_30848_equation_0 = const()[name = tensor<string, []>("op_30848_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30848_cast_fp16 = einsum(equation = var_30848_equation_0, values = (var_30668_cast_fp16, var_30827_cast_fp16))[name = tensor<string, []>("op_30848_cast_fp16")];
            tensor<string, []> var_30850_equation_0 = const()[name = tensor<string, []>("op_30850_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30850_cast_fp16 = einsum(equation = var_30850_equation_0, values = (var_30672_cast_fp16, var_30828_cast_fp16))[name = tensor<string, []>("op_30850_cast_fp16")];
            tensor<string, []> var_30852_equation_0 = const()[name = tensor<string, []>("op_30852_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30852_cast_fp16 = einsum(equation = var_30852_equation_0, values = (var_30676_cast_fp16, var_30829_cast_fp16))[name = tensor<string, []>("op_30852_cast_fp16")];
            tensor<string, []> var_30854_equation_0 = const()[name = tensor<string, []>("op_30854_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30854_cast_fp16 = einsum(equation = var_30854_equation_0, values = (var_30680_cast_fp16, var_30830_cast_fp16))[name = tensor<string, []>("op_30854_cast_fp16")];
            tensor<string, []> var_30856_equation_0 = const()[name = tensor<string, []>("op_30856_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30856_cast_fp16 = einsum(equation = var_30856_equation_0, values = (var_30684_cast_fp16, var_30831_cast_fp16))[name = tensor<string, []>("op_30856_cast_fp16")];
            tensor<string, []> var_30858_equation_0 = const()[name = tensor<string, []>("op_30858_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30858_cast_fp16 = einsum(equation = var_30858_equation_0, values = (var_30688_cast_fp16, var_30832_cast_fp16))[name = tensor<string, []>("op_30858_cast_fp16")];
            tensor<string, []> var_30860_equation_0 = const()[name = tensor<string, []>("op_30860_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30860_cast_fp16 = einsum(equation = var_30860_equation_0, values = (var_30692_cast_fp16, var_30833_cast_fp16))[name = tensor<string, []>("op_30860_cast_fp16")];
            tensor<string, []> var_30862_equation_0 = const()[name = tensor<string, []>("op_30862_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30862_cast_fp16 = einsum(equation = var_30862_equation_0, values = (var_30696_cast_fp16, var_30834_cast_fp16))[name = tensor<string, []>("op_30862_cast_fp16")];
            tensor<string, []> var_30864_equation_0 = const()[name = tensor<string, []>("op_30864_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30864_cast_fp16 = einsum(equation = var_30864_equation_0, values = (var_30700_cast_fp16, var_30835_cast_fp16))[name = tensor<string, []>("op_30864_cast_fp16")];
            tensor<string, []> var_30866_equation_0 = const()[name = tensor<string, []>("op_30866_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30866_cast_fp16 = einsum(equation = var_30866_equation_0, values = (var_30704_cast_fp16, var_30836_cast_fp16))[name = tensor<string, []>("op_30866_cast_fp16")];
            tensor<string, []> var_30868_equation_0 = const()[name = tensor<string, []>("op_30868_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30868_cast_fp16 = einsum(equation = var_30868_equation_0, values = (var_30708_cast_fp16, var_30837_cast_fp16))[name = tensor<string, []>("op_30868_cast_fp16")];
            tensor<string, []> var_30870_equation_0 = const()[name = tensor<string, []>("op_30870_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30870_cast_fp16 = einsum(equation = var_30870_equation_0, values = (var_30712_cast_fp16, var_30838_cast_fp16))[name = tensor<string, []>("op_30870_cast_fp16")];
            tensor<string, []> var_30872_equation_0 = const()[name = tensor<string, []>("op_30872_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30872_cast_fp16 = einsum(equation = var_30872_equation_0, values = (var_30716_cast_fp16, var_30839_cast_fp16))[name = tensor<string, []>("op_30872_cast_fp16")];
            tensor<string, []> var_30874_equation_0 = const()[name = tensor<string, []>("op_30874_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30874_cast_fp16 = einsum(equation = var_30874_equation_0, values = (var_30720_cast_fp16, var_30840_cast_fp16))[name = tensor<string, []>("op_30874_cast_fp16")];
            tensor<string, []> var_30876_equation_0 = const()[name = tensor<string, []>("op_30876_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30876_cast_fp16 = einsum(equation = var_30876_equation_0, values = (var_30724_cast_fp16, var_30841_cast_fp16))[name = tensor<string, []>("op_30876_cast_fp16")];
            tensor<string, []> var_30878_equation_0 = const()[name = tensor<string, []>("op_30878_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30878_cast_fp16 = einsum(equation = var_30878_equation_0, values = (var_30728_cast_fp16, var_30842_cast_fp16))[name = tensor<string, []>("op_30878_cast_fp16")];
            tensor<string, []> var_30880_equation_0 = const()[name = tensor<string, []>("op_30880_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30880_cast_fp16 = einsum(equation = var_30880_equation_0, values = (var_30732_cast_fp16, var_30843_cast_fp16))[name = tensor<string, []>("op_30880_cast_fp16")];
            tensor<string, []> var_30882_equation_0 = const()[name = tensor<string, []>("op_30882_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30882_cast_fp16 = einsum(equation = var_30882_equation_0, values = (var_30736_cast_fp16, var_30844_cast_fp16))[name = tensor<string, []>("op_30882_cast_fp16")];
            tensor<string, []> var_30884_equation_0 = const()[name = tensor<string, []>("op_30884_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30884_cast_fp16 = einsum(equation = var_30884_equation_0, values = (var_30740_cast_fp16, var_30845_cast_fp16))[name = tensor<string, []>("op_30884_cast_fp16")];
            tensor<string, []> var_30886_equation_0 = const()[name = tensor<string, []>("op_30886_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30886_cast_fp16 = einsum(equation = var_30886_equation_0, values = (var_30744_cast_fp16, var_30846_cast_fp16))[name = tensor<string, []>("op_30886_cast_fp16")];
            tensor<bool, []> input_439_interleave_0 = const()[name = tensor<string, []>("input_439_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_439_cast_fp16 = concat(axis = var_30355, interleave = input_439_interleave_0, values = (var_30848_cast_fp16, var_30850_cast_fp16, var_30852_cast_fp16, var_30854_cast_fp16, var_30856_cast_fp16, var_30858_cast_fp16, var_30860_cast_fp16, var_30862_cast_fp16, var_30864_cast_fp16, var_30866_cast_fp16, var_30868_cast_fp16, var_30870_cast_fp16, var_30872_cast_fp16, var_30874_cast_fp16, var_30876_cast_fp16, var_30878_cast_fp16, var_30880_cast_fp16, var_30882_cast_fp16, var_30884_cast_fp16, var_30886_cast_fp16))[name = tensor<string, []>("input_439_cast_fp16")];
            tensor<string, []> var_30896_pad_type_0 = const()[name = tensor<string, []>("op_30896_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_30896_strides_0 = const()[name = tensor<string, []>("op_30896_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_30896_pad_0 = const()[name = tensor<string, []>("op_30896_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_30896_dilations_0 = const()[name = tensor<string, []>("op_30896_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_30896_groups_0 = const()[name = tensor<string, []>("op_30896_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7409600))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8638464))), name = tensor<string, []>("up_blocks_0_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_0_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_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, []>(8638656)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_30896_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_30896_dilations_0, groups = var_30896_groups_0, pad = var_30896_pad_0, pad_type = var_30896_pad_type_0, strides = var_30896_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized, x = input_439_cast_fp16)[name = tensor<string, []>("op_30896_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_207_cast_fp16 = add(x = var_30896_cast_fp16, y = inputs_205_cast_fp16)[name = tensor<string, []>("inputs_207_cast_fp16")];
            tensor<int32, [1]> hidden_states_289_axes_0 = const()[name = tensor<string, []>("hidden_states_289_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_289_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_289_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8641280)))];
            tensor<fp16, [1280]> hidden_states_289_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_289_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8643904)))];
            tensor<fp16, []> var_30906_to_fp16 = const()[name = tensor<string, []>("op_30906_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_289_cast_fp16 = layer_norm(axes = hidden_states_289_axes_0, beta = hidden_states_289_beta_0_to_fp16, epsilon = var_30906_to_fp16, gamma = hidden_states_289_gamma_0_to_fp16, x = inputs_207_cast_fp16)[name = tensor<string, []>("hidden_states_289_cast_fp16")];
            tensor<string, []> q_139_pad_type_0 = const()[name = tensor<string, []>("q_139_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_139_strides_0 = const()[name = tensor<string, []>("q_139_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_139_pad_0 = const()[name = tensor<string, []>("q_139_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_139_dilations_0 = const()[name = tensor<string, []>("q_139_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_139_groups_0 = const()[name = tensor<string, []>("q_139_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8646528))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9875392))), name = tensor<string, []>("up_blocks_0_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, 1024]> q_139_cast_fp16 = conv(dilations = q_139_dilations_0, groups = q_139_groups_0, pad = q_139_pad_0, pad_type = q_139_pad_type_0, strides = q_139_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_289_cast_fp16)[name = tensor<string, []>("q_139_cast_fp16")];
            tensor<string, []> k_277_pad_type_0 = const()[name = tensor<string, []>("k_277_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_277_strides_0 = const()[name = tensor<string, []>("k_277_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_277_pad_0 = const()[name = tensor<string, []>("k_277_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_277_dilations_0 = const()[name = tensor<string, []>("k_277_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_277_groups_0 = const()[name = tensor<string, []>("k_277_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9875584))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11841728))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> k_277_cast_fp16 = conv(dilations = k_277_dilations_0, groups = k_277_groups_0, pad = k_277_pad_0, pad_type = k_277_pad_type_0, strides = k_277_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("k_277_cast_fp16")];
            tensor<string, []> v_139_pad_type_0 = const()[name = tensor<string, []>("v_139_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_139_strides_0 = const()[name = tensor<string, []>("v_139_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_139_pad_0 = const()[name = tensor<string, []>("v_139_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_139_dilations_0 = const()[name = tensor<string, []>("v_139_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_139_groups_0 = const()[name = tensor<string, []>("v_139_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11841920))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13808064))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> v_139_cast_fp16 = conv(dilations = v_139_dilations_0, groups = v_139_groups_0, pad = v_139_pad_0, pad_type = v_139_pad_type_0, strides = v_139_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("v_139_cast_fp16")];
            tensor<int32, [4]> var_30939_begin_0 = const()[name = tensor<string, []>("op_30939_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_30939_end_0 = const()[name = tensor<string, []>("op_30939_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_30939_end_mask_0 = const()[name = tensor<string, []>("op_30939_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30939_cast_fp16 = slice_by_index(begin = var_30939_begin_0, end = var_30939_end_0, end_mask = var_30939_end_mask_0, x = q_139_cast_fp16)[name = tensor<string, []>("op_30939_cast_fp16")];
            tensor<int32, [4]> var_30943_begin_0 = const()[name = tensor<string, []>("op_30943_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_30943_end_0 = const()[name = tensor<string, []>("op_30943_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_30943_end_mask_0 = const()[name = tensor<string, []>("op_30943_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30943_cast_fp16 = slice_by_index(begin = var_30943_begin_0, end = var_30943_end_0, end_mask = var_30943_end_mask_0, x = q_139_cast_fp16)[name = tensor<string, []>("op_30943_cast_fp16")];
            tensor<int32, [4]> var_30947_begin_0 = const()[name = tensor<string, []>("op_30947_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_30947_end_0 = const()[name = tensor<string, []>("op_30947_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_30947_end_mask_0 = const()[name = tensor<string, []>("op_30947_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30947_cast_fp16 = slice_by_index(begin = var_30947_begin_0, end = var_30947_end_0, end_mask = var_30947_end_mask_0, x = q_139_cast_fp16)[name = tensor<string, []>("op_30947_cast_fp16")];
            tensor<int32, [4]> var_30951_begin_0 = const()[name = tensor<string, []>("op_30951_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_30951_end_0 = const()[name = tensor<string, []>("op_30951_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_30951_end_mask_0 = const()[name = tensor<string, []>("op_30951_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30951_cast_fp16 = slice_by_index(begin = var_30951_begin_0, end = var_30951_end_0, end_mask = var_30951_end_mask_0, x = q_139_cast_fp16)[name = tensor<string, []>("op_30951_cast_fp16")];
            tensor<int32, [4]> var_30955_begin_0 = const()[name = tensor<string, []>("op_30955_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_30955_end_0 = const()[name = tensor<string, []>("op_30955_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_30955_end_mask_0 = const()[name = tensor<string, []>("op_30955_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30955_cast_fp16 = slice_by_index(begin = var_30955_begin_0, end = var_30955_end_0, end_mask = var_30955_end_mask_0, x = q_139_cast_fp16)[name = tensor<string, []>("op_30955_cast_fp16")];
            tensor<int32, [4]> var_30959_begin_0 = const()[name = tensor<string, []>("op_30959_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_30959_end_0 = const()[name = tensor<string, []>("op_30959_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_30959_end_mask_0 = const()[name = tensor<string, []>("op_30959_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30959_cast_fp16 = slice_by_index(begin = var_30959_begin_0, end = var_30959_end_0, end_mask = var_30959_end_mask_0, x = q_139_cast_fp16)[name = tensor<string, []>("op_30959_cast_fp16")];
            tensor<int32, [4]> var_30963_begin_0 = const()[name = tensor<string, []>("op_30963_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_30963_end_0 = const()[name = tensor<string, []>("op_30963_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_30963_end_mask_0 = const()[name = tensor<string, []>("op_30963_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30963_cast_fp16 = slice_by_index(begin = var_30963_begin_0, end = var_30963_end_0, end_mask = var_30963_end_mask_0, x = q_139_cast_fp16)[name = tensor<string, []>("op_30963_cast_fp16")];
            tensor<int32, [4]> var_30967_begin_0 = const()[name = tensor<string, []>("op_30967_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_30967_end_0 = const()[name = tensor<string, []>("op_30967_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_30967_end_mask_0 = const()[name = tensor<string, []>("op_30967_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30967_cast_fp16 = slice_by_index(begin = var_30967_begin_0, end = var_30967_end_0, end_mask = var_30967_end_mask_0, x = q_139_cast_fp16)[name = tensor<string, []>("op_30967_cast_fp16")];
            tensor<int32, [4]> var_30971_begin_0 = const()[name = tensor<string, []>("op_30971_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_30971_end_0 = const()[name = tensor<string, []>("op_30971_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_30971_end_mask_0 = const()[name = tensor<string, []>("op_30971_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30971_cast_fp16 = slice_by_index(begin = var_30971_begin_0, end = var_30971_end_0, end_mask = var_30971_end_mask_0, x = q_139_cast_fp16)[name = tensor<string, []>("op_30971_cast_fp16")];
            tensor<int32, [4]> var_30975_begin_0 = const()[name = tensor<string, []>("op_30975_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_30975_end_0 = const()[name = tensor<string, []>("op_30975_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_30975_end_mask_0 = const()[name = tensor<string, []>("op_30975_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30975_cast_fp16 = slice_by_index(begin = var_30975_begin_0, end = var_30975_end_0, end_mask = var_30975_end_mask_0, x = q_139_cast_fp16)[name = tensor<string, []>("op_30975_cast_fp16")];
            tensor<int32, [4]> var_30979_begin_0 = const()[name = tensor<string, []>("op_30979_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_30979_end_0 = const()[name = tensor<string, []>("op_30979_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_30979_end_mask_0 = const()[name = tensor<string, []>("op_30979_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30979_cast_fp16 = slice_by_index(begin = var_30979_begin_0, end = var_30979_end_0, end_mask = var_30979_end_mask_0, x = q_139_cast_fp16)[name = tensor<string, []>("op_30979_cast_fp16")];
            tensor<int32, [4]> var_30983_begin_0 = const()[name = tensor<string, []>("op_30983_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_30983_end_0 = const()[name = tensor<string, []>("op_30983_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_30983_end_mask_0 = const()[name = tensor<string, []>("op_30983_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30983_cast_fp16 = slice_by_index(begin = var_30983_begin_0, end = var_30983_end_0, end_mask = var_30983_end_mask_0, x = q_139_cast_fp16)[name = tensor<string, []>("op_30983_cast_fp16")];
            tensor<int32, [4]> var_30987_begin_0 = const()[name = tensor<string, []>("op_30987_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_30987_end_0 = const()[name = tensor<string, []>("op_30987_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_30987_end_mask_0 = const()[name = tensor<string, []>("op_30987_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30987_cast_fp16 = slice_by_index(begin = var_30987_begin_0, end = var_30987_end_0, end_mask = var_30987_end_mask_0, x = q_139_cast_fp16)[name = tensor<string, []>("op_30987_cast_fp16")];
            tensor<int32, [4]> var_30991_begin_0 = const()[name = tensor<string, []>("op_30991_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_30991_end_0 = const()[name = tensor<string, []>("op_30991_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_30991_end_mask_0 = const()[name = tensor<string, []>("op_30991_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30991_cast_fp16 = slice_by_index(begin = var_30991_begin_0, end = var_30991_end_0, end_mask = var_30991_end_mask_0, x = q_139_cast_fp16)[name = tensor<string, []>("op_30991_cast_fp16")];
            tensor<int32, [4]> var_30995_begin_0 = const()[name = tensor<string, []>("op_30995_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_30995_end_0 = const()[name = tensor<string, []>("op_30995_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_30995_end_mask_0 = const()[name = tensor<string, []>("op_30995_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30995_cast_fp16 = slice_by_index(begin = var_30995_begin_0, end = var_30995_end_0, end_mask = var_30995_end_mask_0, x = q_139_cast_fp16)[name = tensor<string, []>("op_30995_cast_fp16")];
            tensor<int32, [4]> var_30999_begin_0 = const()[name = tensor<string, []>("op_30999_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_30999_end_0 = const()[name = tensor<string, []>("op_30999_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_30999_end_mask_0 = const()[name = tensor<string, []>("op_30999_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30999_cast_fp16 = slice_by_index(begin = var_30999_begin_0, end = var_30999_end_0, end_mask = var_30999_end_mask_0, x = q_139_cast_fp16)[name = tensor<string, []>("op_30999_cast_fp16")];
            tensor<int32, [4]> var_31003_begin_0 = const()[name = tensor<string, []>("op_31003_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_31003_end_0 = const()[name = tensor<string, []>("op_31003_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_31003_end_mask_0 = const()[name = tensor<string, []>("op_31003_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31003_cast_fp16 = slice_by_index(begin = var_31003_begin_0, end = var_31003_end_0, end_mask = var_31003_end_mask_0, x = q_139_cast_fp16)[name = tensor<string, []>("op_31003_cast_fp16")];
            tensor<int32, [4]> var_31007_begin_0 = const()[name = tensor<string, []>("op_31007_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_31007_end_0 = const()[name = tensor<string, []>("op_31007_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_31007_end_mask_0 = const()[name = tensor<string, []>("op_31007_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31007_cast_fp16 = slice_by_index(begin = var_31007_begin_0, end = var_31007_end_0, end_mask = var_31007_end_mask_0, x = q_139_cast_fp16)[name = tensor<string, []>("op_31007_cast_fp16")];
            tensor<int32, [4]> var_31011_begin_0 = const()[name = tensor<string, []>("op_31011_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_31011_end_0 = const()[name = tensor<string, []>("op_31011_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_31011_end_mask_0 = const()[name = tensor<string, []>("op_31011_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31011_cast_fp16 = slice_by_index(begin = var_31011_begin_0, end = var_31011_end_0, end_mask = var_31011_end_mask_0, x = q_139_cast_fp16)[name = tensor<string, []>("op_31011_cast_fp16")];
            tensor<int32, [4]> var_31015_begin_0 = const()[name = tensor<string, []>("op_31015_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_31015_end_0 = const()[name = tensor<string, []>("op_31015_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_31015_end_mask_0 = const()[name = tensor<string, []>("op_31015_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31015_cast_fp16 = slice_by_index(begin = var_31015_begin_0, end = var_31015_end_0, end_mask = var_31015_end_mask_0, x = q_139_cast_fp16)[name = tensor<string, []>("op_31015_cast_fp16")];
            tensor<int32, [4]> k_279_perm_0 = const()[name = tensor<string, []>("k_279_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_31022_begin_0 = const()[name = tensor<string, []>("op_31022_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_31022_end_0 = const()[name = tensor<string, []>("op_31022_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_31022_end_mask_0 = const()[name = tensor<string, []>("op_31022_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> k_279_cast_fp16 = transpose(perm = k_279_perm_0, x = k_277_cast_fp16)[name = tensor<string, []>("transpose_70")];
            tensor<fp16, [2, 77, 1, 64]> var_31022_cast_fp16 = slice_by_index(begin = var_31022_begin_0, end = var_31022_end_0, end_mask = var_31022_end_mask_0, x = k_279_cast_fp16)[name = tensor<string, []>("op_31022_cast_fp16")];
            tensor<int32, [4]> var_31026_begin_0 = const()[name = tensor<string, []>("op_31026_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_31026_end_0 = const()[name = tensor<string, []>("op_31026_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_31026_end_mask_0 = const()[name = tensor<string, []>("op_31026_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31026_cast_fp16 = slice_by_index(begin = var_31026_begin_0, end = var_31026_end_0, end_mask = var_31026_end_mask_0, x = k_279_cast_fp16)[name = tensor<string, []>("op_31026_cast_fp16")];
            tensor<int32, [4]> var_31030_begin_0 = const()[name = tensor<string, []>("op_31030_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_31030_end_0 = const()[name = tensor<string, []>("op_31030_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_31030_end_mask_0 = const()[name = tensor<string, []>("op_31030_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31030_cast_fp16 = slice_by_index(begin = var_31030_begin_0, end = var_31030_end_0, end_mask = var_31030_end_mask_0, x = k_279_cast_fp16)[name = tensor<string, []>("op_31030_cast_fp16")];
            tensor<int32, [4]> var_31034_begin_0 = const()[name = tensor<string, []>("op_31034_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_31034_end_0 = const()[name = tensor<string, []>("op_31034_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_31034_end_mask_0 = const()[name = tensor<string, []>("op_31034_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31034_cast_fp16 = slice_by_index(begin = var_31034_begin_0, end = var_31034_end_0, end_mask = var_31034_end_mask_0, x = k_279_cast_fp16)[name = tensor<string, []>("op_31034_cast_fp16")];
            tensor<int32, [4]> var_31038_begin_0 = const()[name = tensor<string, []>("op_31038_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_31038_end_0 = const()[name = tensor<string, []>("op_31038_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_31038_end_mask_0 = const()[name = tensor<string, []>("op_31038_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31038_cast_fp16 = slice_by_index(begin = var_31038_begin_0, end = var_31038_end_0, end_mask = var_31038_end_mask_0, x = k_279_cast_fp16)[name = tensor<string, []>("op_31038_cast_fp16")];
            tensor<int32, [4]> var_31042_begin_0 = const()[name = tensor<string, []>("op_31042_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_31042_end_0 = const()[name = tensor<string, []>("op_31042_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_31042_end_mask_0 = const()[name = tensor<string, []>("op_31042_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31042_cast_fp16 = slice_by_index(begin = var_31042_begin_0, end = var_31042_end_0, end_mask = var_31042_end_mask_0, x = k_279_cast_fp16)[name = tensor<string, []>("op_31042_cast_fp16")];
            tensor<int32, [4]> var_31046_begin_0 = const()[name = tensor<string, []>("op_31046_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_31046_end_0 = const()[name = tensor<string, []>("op_31046_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_31046_end_mask_0 = const()[name = tensor<string, []>("op_31046_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31046_cast_fp16 = slice_by_index(begin = var_31046_begin_0, end = var_31046_end_0, end_mask = var_31046_end_mask_0, x = k_279_cast_fp16)[name = tensor<string, []>("op_31046_cast_fp16")];
            tensor<int32, [4]> var_31050_begin_0 = const()[name = tensor<string, []>("op_31050_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_31050_end_0 = const()[name = tensor<string, []>("op_31050_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_31050_end_mask_0 = const()[name = tensor<string, []>("op_31050_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31050_cast_fp16 = slice_by_index(begin = var_31050_begin_0, end = var_31050_end_0, end_mask = var_31050_end_mask_0, x = k_279_cast_fp16)[name = tensor<string, []>("op_31050_cast_fp16")];
            tensor<int32, [4]> var_31054_begin_0 = const()[name = tensor<string, []>("op_31054_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_31054_end_0 = const()[name = tensor<string, []>("op_31054_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_31054_end_mask_0 = const()[name = tensor<string, []>("op_31054_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31054_cast_fp16 = slice_by_index(begin = var_31054_begin_0, end = var_31054_end_0, end_mask = var_31054_end_mask_0, x = k_279_cast_fp16)[name = tensor<string, []>("op_31054_cast_fp16")];
            tensor<int32, [4]> var_31058_begin_0 = const()[name = tensor<string, []>("op_31058_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_31058_end_0 = const()[name = tensor<string, []>("op_31058_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_31058_end_mask_0 = const()[name = tensor<string, []>("op_31058_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31058_cast_fp16 = slice_by_index(begin = var_31058_begin_0, end = var_31058_end_0, end_mask = var_31058_end_mask_0, x = k_279_cast_fp16)[name = tensor<string, []>("op_31058_cast_fp16")];
            tensor<int32, [4]> var_31062_begin_0 = const()[name = tensor<string, []>("op_31062_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_31062_end_0 = const()[name = tensor<string, []>("op_31062_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_31062_end_mask_0 = const()[name = tensor<string, []>("op_31062_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31062_cast_fp16 = slice_by_index(begin = var_31062_begin_0, end = var_31062_end_0, end_mask = var_31062_end_mask_0, x = k_279_cast_fp16)[name = tensor<string, []>("op_31062_cast_fp16")];
            tensor<int32, [4]> var_31066_begin_0 = const()[name = tensor<string, []>("op_31066_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_31066_end_0 = const()[name = tensor<string, []>("op_31066_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_31066_end_mask_0 = const()[name = tensor<string, []>("op_31066_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31066_cast_fp16 = slice_by_index(begin = var_31066_begin_0, end = var_31066_end_0, end_mask = var_31066_end_mask_0, x = k_279_cast_fp16)[name = tensor<string, []>("op_31066_cast_fp16")];
            tensor<int32, [4]> var_31070_begin_0 = const()[name = tensor<string, []>("op_31070_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_31070_end_0 = const()[name = tensor<string, []>("op_31070_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_31070_end_mask_0 = const()[name = tensor<string, []>("op_31070_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31070_cast_fp16 = slice_by_index(begin = var_31070_begin_0, end = var_31070_end_0, end_mask = var_31070_end_mask_0, x = k_279_cast_fp16)[name = tensor<string, []>("op_31070_cast_fp16")];
            tensor<int32, [4]> var_31074_begin_0 = const()[name = tensor<string, []>("op_31074_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_31074_end_0 = const()[name = tensor<string, []>("op_31074_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_31074_end_mask_0 = const()[name = tensor<string, []>("op_31074_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31074_cast_fp16 = slice_by_index(begin = var_31074_begin_0, end = var_31074_end_0, end_mask = var_31074_end_mask_0, x = k_279_cast_fp16)[name = tensor<string, []>("op_31074_cast_fp16")];
            tensor<int32, [4]> var_31078_begin_0 = const()[name = tensor<string, []>("op_31078_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_31078_end_0 = const()[name = tensor<string, []>("op_31078_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_31078_end_mask_0 = const()[name = tensor<string, []>("op_31078_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31078_cast_fp16 = slice_by_index(begin = var_31078_begin_0, end = var_31078_end_0, end_mask = var_31078_end_mask_0, x = k_279_cast_fp16)[name = tensor<string, []>("op_31078_cast_fp16")];
            tensor<int32, [4]> var_31082_begin_0 = const()[name = tensor<string, []>("op_31082_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_31082_end_0 = const()[name = tensor<string, []>("op_31082_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_31082_end_mask_0 = const()[name = tensor<string, []>("op_31082_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31082_cast_fp16 = slice_by_index(begin = var_31082_begin_0, end = var_31082_end_0, end_mask = var_31082_end_mask_0, x = k_279_cast_fp16)[name = tensor<string, []>("op_31082_cast_fp16")];
            tensor<int32, [4]> var_31086_begin_0 = const()[name = tensor<string, []>("op_31086_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_31086_end_0 = const()[name = tensor<string, []>("op_31086_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_31086_end_mask_0 = const()[name = tensor<string, []>("op_31086_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31086_cast_fp16 = slice_by_index(begin = var_31086_begin_0, end = var_31086_end_0, end_mask = var_31086_end_mask_0, x = k_279_cast_fp16)[name = tensor<string, []>("op_31086_cast_fp16")];
            tensor<int32, [4]> var_31090_begin_0 = const()[name = tensor<string, []>("op_31090_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_31090_end_0 = const()[name = tensor<string, []>("op_31090_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_31090_end_mask_0 = const()[name = tensor<string, []>("op_31090_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31090_cast_fp16 = slice_by_index(begin = var_31090_begin_0, end = var_31090_end_0, end_mask = var_31090_end_mask_0, x = k_279_cast_fp16)[name = tensor<string, []>("op_31090_cast_fp16")];
            tensor<int32, [4]> var_31094_begin_0 = const()[name = tensor<string, []>("op_31094_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_31094_end_0 = const()[name = tensor<string, []>("op_31094_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_31094_end_mask_0 = const()[name = tensor<string, []>("op_31094_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31094_cast_fp16 = slice_by_index(begin = var_31094_begin_0, end = var_31094_end_0, end_mask = var_31094_end_mask_0, x = k_279_cast_fp16)[name = tensor<string, []>("op_31094_cast_fp16")];
            tensor<int32, [4]> var_31098_begin_0 = const()[name = tensor<string, []>("op_31098_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_31098_end_0 = const()[name = tensor<string, []>("op_31098_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_31098_end_mask_0 = const()[name = tensor<string, []>("op_31098_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31098_cast_fp16 = slice_by_index(begin = var_31098_begin_0, end = var_31098_end_0, end_mask = var_31098_end_mask_0, x = k_279_cast_fp16)[name = tensor<string, []>("op_31098_cast_fp16")];
            tensor<int32, [4]> var_31100_begin_0 = const()[name = tensor<string, []>("op_31100_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_31100_end_0 = const()[name = tensor<string, []>("op_31100_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_31100_end_mask_0 = const()[name = tensor<string, []>("op_31100_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_31100_cast_fp16 = slice_by_index(begin = var_31100_begin_0, end = var_31100_end_0, end_mask = var_31100_end_mask_0, x = v_139_cast_fp16)[name = tensor<string, []>("op_31100_cast_fp16")];
            tensor<int32, [4]> var_31104_begin_0 = const()[name = tensor<string, []>("op_31104_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_31104_end_0 = const()[name = tensor<string, []>("op_31104_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_31104_end_mask_0 = const()[name = tensor<string, []>("op_31104_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_31104_cast_fp16 = slice_by_index(begin = var_31104_begin_0, end = var_31104_end_0, end_mask = var_31104_end_mask_0, x = v_139_cast_fp16)[name = tensor<string, []>("op_31104_cast_fp16")];
            tensor<int32, [4]> var_31108_begin_0 = const()[name = tensor<string, []>("op_31108_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_31108_end_0 = const()[name = tensor<string, []>("op_31108_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_31108_end_mask_0 = const()[name = tensor<string, []>("op_31108_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_31108_cast_fp16 = slice_by_index(begin = var_31108_begin_0, end = var_31108_end_0, end_mask = var_31108_end_mask_0, x = v_139_cast_fp16)[name = tensor<string, []>("op_31108_cast_fp16")];
            tensor<int32, [4]> var_31112_begin_0 = const()[name = tensor<string, []>("op_31112_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_31112_end_0 = const()[name = tensor<string, []>("op_31112_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_31112_end_mask_0 = const()[name = tensor<string, []>("op_31112_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_31112_cast_fp16 = slice_by_index(begin = var_31112_begin_0, end = var_31112_end_0, end_mask = var_31112_end_mask_0, x = v_139_cast_fp16)[name = tensor<string, []>("op_31112_cast_fp16")];
            tensor<int32, [4]> var_31116_begin_0 = const()[name = tensor<string, []>("op_31116_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_31116_end_0 = const()[name = tensor<string, []>("op_31116_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_31116_end_mask_0 = const()[name = tensor<string, []>("op_31116_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_31116_cast_fp16 = slice_by_index(begin = var_31116_begin_0, end = var_31116_end_0, end_mask = var_31116_end_mask_0, x = v_139_cast_fp16)[name = tensor<string, []>("op_31116_cast_fp16")];
            tensor<int32, [4]> var_31120_begin_0 = const()[name = tensor<string, []>("op_31120_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_31120_end_0 = const()[name = tensor<string, []>("op_31120_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_31120_end_mask_0 = const()[name = tensor<string, []>("op_31120_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_31120_cast_fp16 = slice_by_index(begin = var_31120_begin_0, end = var_31120_end_0, end_mask = var_31120_end_mask_0, x = v_139_cast_fp16)[name = tensor<string, []>("op_31120_cast_fp16")];
            tensor<int32, [4]> var_31124_begin_0 = const()[name = tensor<string, []>("op_31124_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_31124_end_0 = const()[name = tensor<string, []>("op_31124_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_31124_end_mask_0 = const()[name = tensor<string, []>("op_31124_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_31124_cast_fp16 = slice_by_index(begin = var_31124_begin_0, end = var_31124_end_0, end_mask = var_31124_end_mask_0, x = v_139_cast_fp16)[name = tensor<string, []>("op_31124_cast_fp16")];
            tensor<int32, [4]> var_31128_begin_0 = const()[name = tensor<string, []>("op_31128_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_31128_end_0 = const()[name = tensor<string, []>("op_31128_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_31128_end_mask_0 = const()[name = tensor<string, []>("op_31128_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_31128_cast_fp16 = slice_by_index(begin = var_31128_begin_0, end = var_31128_end_0, end_mask = var_31128_end_mask_0, x = v_139_cast_fp16)[name = tensor<string, []>("op_31128_cast_fp16")];
            tensor<int32, [4]> var_31132_begin_0 = const()[name = tensor<string, []>("op_31132_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_31132_end_0 = const()[name = tensor<string, []>("op_31132_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_31132_end_mask_0 = const()[name = tensor<string, []>("op_31132_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_31132_cast_fp16 = slice_by_index(begin = var_31132_begin_0, end = var_31132_end_0, end_mask = var_31132_end_mask_0, x = v_139_cast_fp16)[name = tensor<string, []>("op_31132_cast_fp16")];
            tensor<int32, [4]> var_31136_begin_0 = const()[name = tensor<string, []>("op_31136_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_31136_end_0 = const()[name = tensor<string, []>("op_31136_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_31136_end_mask_0 = const()[name = tensor<string, []>("op_31136_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_31136_cast_fp16 = slice_by_index(begin = var_31136_begin_0, end = var_31136_end_0, end_mask = var_31136_end_mask_0, x = v_139_cast_fp16)[name = tensor<string, []>("op_31136_cast_fp16")];
            tensor<int32, [4]> var_31140_begin_0 = const()[name = tensor<string, []>("op_31140_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_31140_end_0 = const()[name = tensor<string, []>("op_31140_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_31140_end_mask_0 = const()[name = tensor<string, []>("op_31140_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_31140_cast_fp16 = slice_by_index(begin = var_31140_begin_0, end = var_31140_end_0, end_mask = var_31140_end_mask_0, x = v_139_cast_fp16)[name = tensor<string, []>("op_31140_cast_fp16")];
            tensor<int32, [4]> var_31144_begin_0 = const()[name = tensor<string, []>("op_31144_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_31144_end_0 = const()[name = tensor<string, []>("op_31144_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_31144_end_mask_0 = const()[name = tensor<string, []>("op_31144_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_31144_cast_fp16 = slice_by_index(begin = var_31144_begin_0, end = var_31144_end_0, end_mask = var_31144_end_mask_0, x = v_139_cast_fp16)[name = tensor<string, []>("op_31144_cast_fp16")];
            tensor<int32, [4]> var_31148_begin_0 = const()[name = tensor<string, []>("op_31148_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_31148_end_0 = const()[name = tensor<string, []>("op_31148_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_31148_end_mask_0 = const()[name = tensor<string, []>("op_31148_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_31148_cast_fp16 = slice_by_index(begin = var_31148_begin_0, end = var_31148_end_0, end_mask = var_31148_end_mask_0, x = v_139_cast_fp16)[name = tensor<string, []>("op_31148_cast_fp16")];
            tensor<int32, [4]> var_31152_begin_0 = const()[name = tensor<string, []>("op_31152_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_31152_end_0 = const()[name = tensor<string, []>("op_31152_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_31152_end_mask_0 = const()[name = tensor<string, []>("op_31152_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_31152_cast_fp16 = slice_by_index(begin = var_31152_begin_0, end = var_31152_end_0, end_mask = var_31152_end_mask_0, x = v_139_cast_fp16)[name = tensor<string, []>("op_31152_cast_fp16")];
            tensor<int32, [4]> var_31156_begin_0 = const()[name = tensor<string, []>("op_31156_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_31156_end_0 = const()[name = tensor<string, []>("op_31156_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_31156_end_mask_0 = const()[name = tensor<string, []>("op_31156_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_31156_cast_fp16 = slice_by_index(begin = var_31156_begin_0, end = var_31156_end_0, end_mask = var_31156_end_mask_0, x = v_139_cast_fp16)[name = tensor<string, []>("op_31156_cast_fp16")];
            tensor<int32, [4]> var_31160_begin_0 = const()[name = tensor<string, []>("op_31160_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_31160_end_0 = const()[name = tensor<string, []>("op_31160_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_31160_end_mask_0 = const()[name = tensor<string, []>("op_31160_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_31160_cast_fp16 = slice_by_index(begin = var_31160_begin_0, end = var_31160_end_0, end_mask = var_31160_end_mask_0, x = v_139_cast_fp16)[name = tensor<string, []>("op_31160_cast_fp16")];
            tensor<int32, [4]> var_31164_begin_0 = const()[name = tensor<string, []>("op_31164_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_31164_end_0 = const()[name = tensor<string, []>("op_31164_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_31164_end_mask_0 = const()[name = tensor<string, []>("op_31164_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_31164_cast_fp16 = slice_by_index(begin = var_31164_begin_0, end = var_31164_end_0, end_mask = var_31164_end_mask_0, x = v_139_cast_fp16)[name = tensor<string, []>("op_31164_cast_fp16")];
            tensor<int32, [4]> var_31168_begin_0 = const()[name = tensor<string, []>("op_31168_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_31168_end_0 = const()[name = tensor<string, []>("op_31168_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_31168_end_mask_0 = const()[name = tensor<string, []>("op_31168_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_31168_cast_fp16 = slice_by_index(begin = var_31168_begin_0, end = var_31168_end_0, end_mask = var_31168_end_mask_0, x = v_139_cast_fp16)[name = tensor<string, []>("op_31168_cast_fp16")];
            tensor<int32, [4]> var_31172_begin_0 = const()[name = tensor<string, []>("op_31172_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_31172_end_0 = const()[name = tensor<string, []>("op_31172_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_31172_end_mask_0 = const()[name = tensor<string, []>("op_31172_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_31172_cast_fp16 = slice_by_index(begin = var_31172_begin_0, end = var_31172_end_0, end_mask = var_31172_end_mask_0, x = v_139_cast_fp16)[name = tensor<string, []>("op_31172_cast_fp16")];
            tensor<int32, [4]> var_31176_begin_0 = const()[name = tensor<string, []>("op_31176_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_31176_end_0 = const()[name = tensor<string, []>("op_31176_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_31176_end_mask_0 = const()[name = tensor<string, []>("op_31176_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_31176_cast_fp16 = slice_by_index(begin = var_31176_begin_0, end = var_31176_end_0, end_mask = var_31176_end_mask_0, x = v_139_cast_fp16)[name = tensor<string, []>("op_31176_cast_fp16")];
            tensor<string, []> var_31180_equation_0 = const()[name = tensor<string, []>("op_31180_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_31180_cast_fp16 = einsum(equation = var_31180_equation_0, values = (var_31022_cast_fp16, var_30939_cast_fp16))[name = tensor<string, []>("op_31180_cast_fp16")];
            tensor<fp16, []> var_31181_to_fp16 = const()[name = tensor<string, []>("op_31181_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2601_cast_fp16 = mul(x = var_31180_cast_fp16, y = var_31181_to_fp16)[name = tensor<string, []>("aw_2601_cast_fp16")];
            tensor<string, []> var_31184_equation_0 = const()[name = tensor<string, []>("op_31184_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_31184_cast_fp16 = einsum(equation = var_31184_equation_0, values = (var_31026_cast_fp16, var_30943_cast_fp16))[name = tensor<string, []>("op_31184_cast_fp16")];
            tensor<fp16, []> var_31185_to_fp16 = const()[name = tensor<string, []>("op_31185_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2603_cast_fp16 = mul(x = var_31184_cast_fp16, y = var_31185_to_fp16)[name = tensor<string, []>("aw_2603_cast_fp16")];
            tensor<string, []> var_31188_equation_0 = const()[name = tensor<string, []>("op_31188_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_31188_cast_fp16 = einsum(equation = var_31188_equation_0, values = (var_31030_cast_fp16, var_30947_cast_fp16))[name = tensor<string, []>("op_31188_cast_fp16")];
            tensor<fp16, []> var_31189_to_fp16 = const()[name = tensor<string, []>("op_31189_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2605_cast_fp16 = mul(x = var_31188_cast_fp16, y = var_31189_to_fp16)[name = tensor<string, []>("aw_2605_cast_fp16")];
            tensor<string, []> var_31192_equation_0 = const()[name = tensor<string, []>("op_31192_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_31192_cast_fp16 = einsum(equation = var_31192_equation_0, values = (var_31034_cast_fp16, var_30951_cast_fp16))[name = tensor<string, []>("op_31192_cast_fp16")];
            tensor<fp16, []> var_31193_to_fp16 = const()[name = tensor<string, []>("op_31193_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2607_cast_fp16 = mul(x = var_31192_cast_fp16, y = var_31193_to_fp16)[name = tensor<string, []>("aw_2607_cast_fp16")];
            tensor<string, []> var_31196_equation_0 = const()[name = tensor<string, []>("op_31196_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_31196_cast_fp16 = einsum(equation = var_31196_equation_0, values = (var_31038_cast_fp16, var_30955_cast_fp16))[name = tensor<string, []>("op_31196_cast_fp16")];
            tensor<fp16, []> var_31197_to_fp16 = const()[name = tensor<string, []>("op_31197_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2609_cast_fp16 = mul(x = var_31196_cast_fp16, y = var_31197_to_fp16)[name = tensor<string, []>("aw_2609_cast_fp16")];
            tensor<string, []> var_31200_equation_0 = const()[name = tensor<string, []>("op_31200_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_31200_cast_fp16 = einsum(equation = var_31200_equation_0, values = (var_31042_cast_fp16, var_30959_cast_fp16))[name = tensor<string, []>("op_31200_cast_fp16")];
            tensor<fp16, []> var_31201_to_fp16 = const()[name = tensor<string, []>("op_31201_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2611_cast_fp16 = mul(x = var_31200_cast_fp16, y = var_31201_to_fp16)[name = tensor<string, []>("aw_2611_cast_fp16")];
            tensor<string, []> var_31204_equation_0 = const()[name = tensor<string, []>("op_31204_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_31204_cast_fp16 = einsum(equation = var_31204_equation_0, values = (var_31046_cast_fp16, var_30963_cast_fp16))[name = tensor<string, []>("op_31204_cast_fp16")];
            tensor<fp16, []> var_31205_to_fp16 = const()[name = tensor<string, []>("op_31205_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2613_cast_fp16 = mul(x = var_31204_cast_fp16, y = var_31205_to_fp16)[name = tensor<string, []>("aw_2613_cast_fp16")];
            tensor<string, []> var_31208_equation_0 = const()[name = tensor<string, []>("op_31208_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_31208_cast_fp16 = einsum(equation = var_31208_equation_0, values = (var_31050_cast_fp16, var_30967_cast_fp16))[name = tensor<string, []>("op_31208_cast_fp16")];
            tensor<fp16, []> var_31209_to_fp16 = const()[name = tensor<string, []>("op_31209_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2615_cast_fp16 = mul(x = var_31208_cast_fp16, y = var_31209_to_fp16)[name = tensor<string, []>("aw_2615_cast_fp16")];
            tensor<string, []> var_31212_equation_0 = const()[name = tensor<string, []>("op_31212_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_31212_cast_fp16 = einsum(equation = var_31212_equation_0, values = (var_31054_cast_fp16, var_30971_cast_fp16))[name = tensor<string, []>("op_31212_cast_fp16")];
            tensor<fp16, []> var_31213_to_fp16 = const()[name = tensor<string, []>("op_31213_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2617_cast_fp16 = mul(x = var_31212_cast_fp16, y = var_31213_to_fp16)[name = tensor<string, []>("aw_2617_cast_fp16")];
            tensor<string, []> var_31216_equation_0 = const()[name = tensor<string, []>("op_31216_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_31216_cast_fp16 = einsum(equation = var_31216_equation_0, values = (var_31058_cast_fp16, var_30975_cast_fp16))[name = tensor<string, []>("op_31216_cast_fp16")];
            tensor<fp16, []> var_31217_to_fp16 = const()[name = tensor<string, []>("op_31217_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2619_cast_fp16 = mul(x = var_31216_cast_fp16, y = var_31217_to_fp16)[name = tensor<string, []>("aw_2619_cast_fp16")];
            tensor<string, []> var_31220_equation_0 = const()[name = tensor<string, []>("op_31220_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_31220_cast_fp16 = einsum(equation = var_31220_equation_0, values = (var_31062_cast_fp16, var_30979_cast_fp16))[name = tensor<string, []>("op_31220_cast_fp16")];
            tensor<fp16, []> var_31221_to_fp16 = const()[name = tensor<string, []>("op_31221_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2621_cast_fp16 = mul(x = var_31220_cast_fp16, y = var_31221_to_fp16)[name = tensor<string, []>("aw_2621_cast_fp16")];
            tensor<string, []> var_31224_equation_0 = const()[name = tensor<string, []>("op_31224_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_31224_cast_fp16 = einsum(equation = var_31224_equation_0, values = (var_31066_cast_fp16, var_30983_cast_fp16))[name = tensor<string, []>("op_31224_cast_fp16")];
            tensor<fp16, []> var_31225_to_fp16 = const()[name = tensor<string, []>("op_31225_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2623_cast_fp16 = mul(x = var_31224_cast_fp16, y = var_31225_to_fp16)[name = tensor<string, []>("aw_2623_cast_fp16")];
            tensor<string, []> var_31228_equation_0 = const()[name = tensor<string, []>("op_31228_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_31228_cast_fp16 = einsum(equation = var_31228_equation_0, values = (var_31070_cast_fp16, var_30987_cast_fp16))[name = tensor<string, []>("op_31228_cast_fp16")];
            tensor<fp16, []> var_31229_to_fp16 = const()[name = tensor<string, []>("op_31229_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2625_cast_fp16 = mul(x = var_31228_cast_fp16, y = var_31229_to_fp16)[name = tensor<string, []>("aw_2625_cast_fp16")];
            tensor<string, []> var_31232_equation_0 = const()[name = tensor<string, []>("op_31232_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_31232_cast_fp16 = einsum(equation = var_31232_equation_0, values = (var_31074_cast_fp16, var_30991_cast_fp16))[name = tensor<string, []>("op_31232_cast_fp16")];
            tensor<fp16, []> var_31233_to_fp16 = const()[name = tensor<string, []>("op_31233_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2627_cast_fp16 = mul(x = var_31232_cast_fp16, y = var_31233_to_fp16)[name = tensor<string, []>("aw_2627_cast_fp16")];
            tensor<string, []> var_31236_equation_0 = const()[name = tensor<string, []>("op_31236_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_31236_cast_fp16 = einsum(equation = var_31236_equation_0, values = (var_31078_cast_fp16, var_30995_cast_fp16))[name = tensor<string, []>("op_31236_cast_fp16")];
            tensor<fp16, []> var_31237_to_fp16 = const()[name = tensor<string, []>("op_31237_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2629_cast_fp16 = mul(x = var_31236_cast_fp16, y = var_31237_to_fp16)[name = tensor<string, []>("aw_2629_cast_fp16")];
            tensor<string, []> var_31240_equation_0 = const()[name = tensor<string, []>("op_31240_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_31240_cast_fp16 = einsum(equation = var_31240_equation_0, values = (var_31082_cast_fp16, var_30999_cast_fp16))[name = tensor<string, []>("op_31240_cast_fp16")];
            tensor<fp16, []> var_31241_to_fp16 = const()[name = tensor<string, []>("op_31241_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2631_cast_fp16 = mul(x = var_31240_cast_fp16, y = var_31241_to_fp16)[name = tensor<string, []>("aw_2631_cast_fp16")];
            tensor<string, []> var_31244_equation_0 = const()[name = tensor<string, []>("op_31244_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_31244_cast_fp16 = einsum(equation = var_31244_equation_0, values = (var_31086_cast_fp16, var_31003_cast_fp16))[name = tensor<string, []>("op_31244_cast_fp16")];
            tensor<fp16, []> var_31245_to_fp16 = const()[name = tensor<string, []>("op_31245_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2633_cast_fp16 = mul(x = var_31244_cast_fp16, y = var_31245_to_fp16)[name = tensor<string, []>("aw_2633_cast_fp16")];
            tensor<string, []> var_31248_equation_0 = const()[name = tensor<string, []>("op_31248_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_31248_cast_fp16 = einsum(equation = var_31248_equation_0, values = (var_31090_cast_fp16, var_31007_cast_fp16))[name = tensor<string, []>("op_31248_cast_fp16")];
            tensor<fp16, []> var_31249_to_fp16 = const()[name = tensor<string, []>("op_31249_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2635_cast_fp16 = mul(x = var_31248_cast_fp16, y = var_31249_to_fp16)[name = tensor<string, []>("aw_2635_cast_fp16")];
            tensor<string, []> var_31252_equation_0 = const()[name = tensor<string, []>("op_31252_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_31252_cast_fp16 = einsum(equation = var_31252_equation_0, values = (var_31094_cast_fp16, var_31011_cast_fp16))[name = tensor<string, []>("op_31252_cast_fp16")];
            tensor<fp16, []> var_31253_to_fp16 = const()[name = tensor<string, []>("op_31253_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2637_cast_fp16 = mul(x = var_31252_cast_fp16, y = var_31253_to_fp16)[name = tensor<string, []>("aw_2637_cast_fp16")];
            tensor<string, []> var_31256_equation_0 = const()[name = tensor<string, []>("op_31256_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_31256_cast_fp16 = einsum(equation = var_31256_equation_0, values = (var_31098_cast_fp16, var_31015_cast_fp16))[name = tensor<string, []>("op_31256_cast_fp16")];
            tensor<fp16, []> var_31257_to_fp16 = const()[name = tensor<string, []>("op_31257_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2639_cast_fp16 = mul(x = var_31256_cast_fp16, y = var_31257_to_fp16)[name = tensor<string, []>("aw_2639_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_31259_cast_fp16 = softmax(axis = var_30355, x = aw_2601_cast_fp16)[name = tensor<string, []>("op_31259_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_31260_cast_fp16 = softmax(axis = var_30355, x = aw_2603_cast_fp16)[name = tensor<string, []>("op_31260_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_31261_cast_fp16 = softmax(axis = var_30355, x = aw_2605_cast_fp16)[name = tensor<string, []>("op_31261_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_31262_cast_fp16 = softmax(axis = var_30355, x = aw_2607_cast_fp16)[name = tensor<string, []>("op_31262_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_31263_cast_fp16 = softmax(axis = var_30355, x = aw_2609_cast_fp16)[name = tensor<string, []>("op_31263_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_31264_cast_fp16 = softmax(axis = var_30355, x = aw_2611_cast_fp16)[name = tensor<string, []>("op_31264_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_31265_cast_fp16 = softmax(axis = var_30355, x = aw_2613_cast_fp16)[name = tensor<string, []>("op_31265_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_31266_cast_fp16 = softmax(axis = var_30355, x = aw_2615_cast_fp16)[name = tensor<string, []>("op_31266_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_31267_cast_fp16 = softmax(axis = var_30355, x = aw_2617_cast_fp16)[name = tensor<string, []>("op_31267_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_31268_cast_fp16 = softmax(axis = var_30355, x = aw_2619_cast_fp16)[name = tensor<string, []>("op_31268_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_31269_cast_fp16 = softmax(axis = var_30355, x = aw_2621_cast_fp16)[name = tensor<string, []>("op_31269_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_31270_cast_fp16 = softmax(axis = var_30355, x = aw_2623_cast_fp16)[name = tensor<string, []>("op_31270_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_31271_cast_fp16 = softmax(axis = var_30355, x = aw_2625_cast_fp16)[name = tensor<string, []>("op_31271_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_31272_cast_fp16 = softmax(axis = var_30355, x = aw_2627_cast_fp16)[name = tensor<string, []>("op_31272_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_31273_cast_fp16 = softmax(axis = var_30355, x = aw_2629_cast_fp16)[name = tensor<string, []>("op_31273_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_31274_cast_fp16 = softmax(axis = var_30355, x = aw_2631_cast_fp16)[name = tensor<string, []>("op_31274_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_31275_cast_fp16 = softmax(axis = var_30355, x = aw_2633_cast_fp16)[name = tensor<string, []>("op_31275_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_31276_cast_fp16 = softmax(axis = var_30355, x = aw_2635_cast_fp16)[name = tensor<string, []>("op_31276_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_31277_cast_fp16 = softmax(axis = var_30355, x = aw_2637_cast_fp16)[name = tensor<string, []>("op_31277_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_31278_cast_fp16 = softmax(axis = var_30355, x = aw_2639_cast_fp16)[name = tensor<string, []>("op_31278_cast_fp16")];
            tensor<string, []> var_31280_equation_0 = const()[name = tensor<string, []>("op_31280_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31280_cast_fp16 = einsum(equation = var_31280_equation_0, values = (var_31100_cast_fp16, var_31259_cast_fp16))[name = tensor<string, []>("op_31280_cast_fp16")];
            tensor<string, []> var_31282_equation_0 = const()[name = tensor<string, []>("op_31282_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31282_cast_fp16 = einsum(equation = var_31282_equation_0, values = (var_31104_cast_fp16, var_31260_cast_fp16))[name = tensor<string, []>("op_31282_cast_fp16")];
            tensor<string, []> var_31284_equation_0 = const()[name = tensor<string, []>("op_31284_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31284_cast_fp16 = einsum(equation = var_31284_equation_0, values = (var_31108_cast_fp16, var_31261_cast_fp16))[name = tensor<string, []>("op_31284_cast_fp16")];
            tensor<string, []> var_31286_equation_0 = const()[name = tensor<string, []>("op_31286_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31286_cast_fp16 = einsum(equation = var_31286_equation_0, values = (var_31112_cast_fp16, var_31262_cast_fp16))[name = tensor<string, []>("op_31286_cast_fp16")];
            tensor<string, []> var_31288_equation_0 = const()[name = tensor<string, []>("op_31288_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31288_cast_fp16 = einsum(equation = var_31288_equation_0, values = (var_31116_cast_fp16, var_31263_cast_fp16))[name = tensor<string, []>("op_31288_cast_fp16")];
            tensor<string, []> var_31290_equation_0 = const()[name = tensor<string, []>("op_31290_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31290_cast_fp16 = einsum(equation = var_31290_equation_0, values = (var_31120_cast_fp16, var_31264_cast_fp16))[name = tensor<string, []>("op_31290_cast_fp16")];
            tensor<string, []> var_31292_equation_0 = const()[name = tensor<string, []>("op_31292_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31292_cast_fp16 = einsum(equation = var_31292_equation_0, values = (var_31124_cast_fp16, var_31265_cast_fp16))[name = tensor<string, []>("op_31292_cast_fp16")];
            tensor<string, []> var_31294_equation_0 = const()[name = tensor<string, []>("op_31294_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31294_cast_fp16 = einsum(equation = var_31294_equation_0, values = (var_31128_cast_fp16, var_31266_cast_fp16))[name = tensor<string, []>("op_31294_cast_fp16")];
            tensor<string, []> var_31296_equation_0 = const()[name = tensor<string, []>("op_31296_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31296_cast_fp16 = einsum(equation = var_31296_equation_0, values = (var_31132_cast_fp16, var_31267_cast_fp16))[name = tensor<string, []>("op_31296_cast_fp16")];
            tensor<string, []> var_31298_equation_0 = const()[name = tensor<string, []>("op_31298_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31298_cast_fp16 = einsum(equation = var_31298_equation_0, values = (var_31136_cast_fp16, var_31268_cast_fp16))[name = tensor<string, []>("op_31298_cast_fp16")];
            tensor<string, []> var_31300_equation_0 = const()[name = tensor<string, []>("op_31300_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31300_cast_fp16 = einsum(equation = var_31300_equation_0, values = (var_31140_cast_fp16, var_31269_cast_fp16))[name = tensor<string, []>("op_31300_cast_fp16")];
            tensor<string, []> var_31302_equation_0 = const()[name = tensor<string, []>("op_31302_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31302_cast_fp16 = einsum(equation = var_31302_equation_0, values = (var_31144_cast_fp16, var_31270_cast_fp16))[name = tensor<string, []>("op_31302_cast_fp16")];
            tensor<string, []> var_31304_equation_0 = const()[name = tensor<string, []>("op_31304_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31304_cast_fp16 = einsum(equation = var_31304_equation_0, values = (var_31148_cast_fp16, var_31271_cast_fp16))[name = tensor<string, []>("op_31304_cast_fp16")];
            tensor<string, []> var_31306_equation_0 = const()[name = tensor<string, []>("op_31306_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31306_cast_fp16 = einsum(equation = var_31306_equation_0, values = (var_31152_cast_fp16, var_31272_cast_fp16))[name = tensor<string, []>("op_31306_cast_fp16")];
            tensor<string, []> var_31308_equation_0 = const()[name = tensor<string, []>("op_31308_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31308_cast_fp16 = einsum(equation = var_31308_equation_0, values = (var_31156_cast_fp16, var_31273_cast_fp16))[name = tensor<string, []>("op_31308_cast_fp16")];
            tensor<string, []> var_31310_equation_0 = const()[name = tensor<string, []>("op_31310_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31310_cast_fp16 = einsum(equation = var_31310_equation_0, values = (var_31160_cast_fp16, var_31274_cast_fp16))[name = tensor<string, []>("op_31310_cast_fp16")];
            tensor<string, []> var_31312_equation_0 = const()[name = tensor<string, []>("op_31312_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31312_cast_fp16 = einsum(equation = var_31312_equation_0, values = (var_31164_cast_fp16, var_31275_cast_fp16))[name = tensor<string, []>("op_31312_cast_fp16")];
            tensor<string, []> var_31314_equation_0 = const()[name = tensor<string, []>("op_31314_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31314_cast_fp16 = einsum(equation = var_31314_equation_0, values = (var_31168_cast_fp16, var_31276_cast_fp16))[name = tensor<string, []>("op_31314_cast_fp16")];
            tensor<string, []> var_31316_equation_0 = const()[name = tensor<string, []>("op_31316_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31316_cast_fp16 = einsum(equation = var_31316_equation_0, values = (var_31172_cast_fp16, var_31277_cast_fp16))[name = tensor<string, []>("op_31316_cast_fp16")];
            tensor<string, []> var_31318_equation_0 = const()[name = tensor<string, []>("op_31318_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31318_cast_fp16 = einsum(equation = var_31318_equation_0, values = (var_31176_cast_fp16, var_31278_cast_fp16))[name = tensor<string, []>("op_31318_cast_fp16")];
            tensor<bool, []> input_441_interleave_0 = const()[name = tensor<string, []>("input_441_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_441_cast_fp16 = concat(axis = var_30355, interleave = input_441_interleave_0, values = (var_31280_cast_fp16, var_31282_cast_fp16, var_31284_cast_fp16, var_31286_cast_fp16, var_31288_cast_fp16, var_31290_cast_fp16, var_31292_cast_fp16, var_31294_cast_fp16, var_31296_cast_fp16, var_31298_cast_fp16, var_31300_cast_fp16, var_31302_cast_fp16, var_31304_cast_fp16, var_31306_cast_fp16, var_31308_cast_fp16, var_31310_cast_fp16, var_31312_cast_fp16, var_31314_cast_fp16, var_31316_cast_fp16, var_31318_cast_fp16))[name = tensor<string, []>("input_441_cast_fp16")];
            tensor<string, []> var_31328_pad_type_0 = const()[name = tensor<string, []>("op_31328_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_31328_strides_0 = const()[name = tensor<string, []>("op_31328_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_31328_pad_0 = const()[name = tensor<string, []>("op_31328_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_31328_dilations_0 = const()[name = tensor<string, []>("op_31328_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_31328_groups_0 = const()[name = tensor<string, []>("op_31328_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13808256))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(15037120))), name = tensor<string, []>("up_blocks_0_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_0_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_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, []>(15037312)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_31328_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_31328_dilations_0, groups = var_31328_groups_0, pad = var_31328_pad_0, pad_type = var_31328_pad_type_0, strides = var_31328_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized, x = input_441_cast_fp16)[name = tensor<string, []>("op_31328_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_209_cast_fp16 = add(x = var_31328_cast_fp16, y = inputs_207_cast_fp16)[name = tensor<string, []>("inputs_209_cast_fp16")];
            tensor<int32, [1]> input_443_axes_0 = const()[name = tensor<string, []>("input_443_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> input_443_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_443_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(15039936)))];
            tensor<fp16, [1280]> input_443_beta_0_to_fp16 = const()[name = tensor<string, []>("input_443_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(15042560)))];
            tensor<fp16, []> var_31338_to_fp16 = const()[name = tensor<string, []>("op_31338_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> input_443_cast_fp16 = layer_norm(axes = input_443_axes_0, beta = input_443_beta_0_to_fp16, epsilon = var_31338_to_fp16, gamma = input_443_gamma_0_to_fp16, x = inputs_209_cast_fp16)[name = tensor<string, []>("input_443_cast_fp16")];
            tensor<string, []> var_31358_pad_type_0 = const()[name = tensor<string, []>("op_31358_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_31358_strides_0 = const()[name = tensor<string, []>("op_31358_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_31358_pad_0 = const()[name = tensor<string, []>("op_31358_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_31358_dilations_0 = const()[name = tensor<string, []>("op_31358_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_31358_groups_0 = const()[name = tensor<string, []>("op_31358_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [9830400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(15045184))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(24875648))), name = tensor<string, []>("up_blocks_0_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_0_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(24875840)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_31358_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_31358_dilations_0, groups = var_31358_groups_0, pad = var_31358_pad_0, pad_type = var_31358_pad_type_0, strides = var_31358_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized, x = input_443_cast_fp16)[name = tensor<string, []>("op_31358_cast_fp16")];
            tensor<int32, [2]> var_31359_split_sizes_0 = const()[name = tensor<string, []>("op_31359_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_31359_axis_0 = const()[name = tensor<string, []>("op_31359_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_31359_cast_fp16_0, tensor<fp16, [2, 5120, 1, 1024]> var_31359_cast_fp16_1 = split(axis = var_31359_axis_0, split_sizes = var_31359_split_sizes_0, x = var_31358_cast_fp16)[name = tensor<string, []>("op_31359_cast_fp16")];
            tensor<string, []> var_31361_mode_0 = const()[name = tensor<string, []>("op_31361_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_31361_cast_fp16 = gelu(mode = var_31361_mode_0, x = var_31359_cast_fp16_1)[name = tensor<string, []>("op_31361_cast_fp16")];
            tensor<fp16, [2, 5120, 1, 1024]> input_445_cast_fp16 = mul(x = var_31359_cast_fp16_0, y = var_31361_cast_fp16)[name = tensor<string, []>("input_445_cast_fp16")];
            tensor<string, []> var_31369_pad_type_0 = const()[name = tensor<string, []>("op_31369_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_31369_strides_0 = const()[name = tensor<string, []>("op_31369_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_31369_pad_0 = const()[name = tensor<string, []>("op_31369_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_31369_dilations_0 = const()[name = tensor<string, []>("op_31369_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_31369_groups_0 = const()[name = tensor<string, []>("op_31369_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [4915200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(24896384))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(29811648))), name = tensor<string, []>("up_blocks_0_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_0_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_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, []>(29811840)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_31369_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_31369_dilations_0, groups = var_31369_groups_0, pad = var_31369_pad_0, pad_type = var_31369_pad_type_0, strides = var_31369_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_445_cast_fp16)[name = tensor<string, []>("op_31369_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_211_cast_fp16 = add(x = var_31369_cast_fp16, y = inputs_209_cast_fp16)[name = tensor<string, []>("inputs_211_cast_fp16")];
            tensor<int32, [1]> hidden_states_293_axes_0 = const()[name = tensor<string, []>("hidden_states_293_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_293_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_293_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(29814464)))];
            tensor<fp16, [1280]> hidden_states_293_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_293_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(29817088)))];
            tensor<fp16, []> var_31385_to_fp16 = const()[name = tensor<string, []>("op_31385_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_293_cast_fp16 = layer_norm(axes = hidden_states_293_axes_0, beta = hidden_states_293_beta_0_to_fp16, epsilon = var_31385_to_fp16, gamma = hidden_states_293_gamma_0_to_fp16, x = inputs_211_cast_fp16)[name = tensor<string, []>("hidden_states_293_cast_fp16")];
            tensor<string, []> q_141_pad_type_0 = const()[name = tensor<string, []>("q_141_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_141_strides_0 = const()[name = tensor<string, []>("q_141_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_141_pad_0 = const()[name = tensor<string, []>("q_141_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_141_dilations_0 = const()[name = tensor<string, []>("q_141_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_141_groups_0 = const()[name = tensor<string, []>("q_141_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_1_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(29819712))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(31048576))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_1_attn1_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_141_cast_fp16 = conv(dilations = q_141_dilations_0, groups = q_141_groups_0, pad = q_141_pad_0, pad_type = q_141_pad_type_0, strides = q_141_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_1_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_293_cast_fp16)[name = tensor<string, []>("q_141_cast_fp16")];
            tensor<string, []> k_281_pad_type_0 = const()[name = tensor<string, []>("k_281_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_281_strides_0 = const()[name = tensor<string, []>("k_281_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_281_pad_0 = const()[name = tensor<string, []>("k_281_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_281_dilations_0 = const()[name = tensor<string, []>("k_281_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_281_groups_0 = const()[name = tensor<string, []>("k_281_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_1_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(31048768))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(32277632))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_1_attn1_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> k_281_cast_fp16 = conv(dilations = k_281_dilations_0, groups = k_281_groups_0, pad = k_281_pad_0, pad_type = k_281_pad_type_0, strides = k_281_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_1_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_293_cast_fp16)[name = tensor<string, []>("k_281_cast_fp16")];
            tensor<string, []> v_141_pad_type_0 = const()[name = tensor<string, []>("v_141_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_141_strides_0 = const()[name = tensor<string, []>("v_141_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_141_pad_0 = const()[name = tensor<string, []>("v_141_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_141_dilations_0 = const()[name = tensor<string, []>("v_141_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_141_groups_0 = const()[name = tensor<string, []>("v_141_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_1_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(32277824))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(33506688))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_1_attn1_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> v_141_cast_fp16 = conv(dilations = v_141_dilations_0, groups = v_141_groups_0, pad = v_141_pad_0, pad_type = v_141_pad_type_0, strides = v_141_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_1_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_293_cast_fp16)[name = tensor<string, []>("v_141_cast_fp16")];
            tensor<int32, [4]> var_31418_begin_0 = const()[name = tensor<string, []>("op_31418_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_31418_end_0 = const()[name = tensor<string, []>("op_31418_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_31418_end_mask_0 = const()[name = tensor<string, []>("op_31418_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31418_cast_fp16 = slice_by_index(begin = var_31418_begin_0, end = var_31418_end_0, end_mask = var_31418_end_mask_0, x = q_141_cast_fp16)[name = tensor<string, []>("op_31418_cast_fp16")];
            tensor<int32, [4]> var_31422_begin_0 = const()[name = tensor<string, []>("op_31422_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_31422_end_0 = const()[name = tensor<string, []>("op_31422_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_31422_end_mask_0 = const()[name = tensor<string, []>("op_31422_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31422_cast_fp16 = slice_by_index(begin = var_31422_begin_0, end = var_31422_end_0, end_mask = var_31422_end_mask_0, x = q_141_cast_fp16)[name = tensor<string, []>("op_31422_cast_fp16")];
            tensor<int32, [4]> var_31426_begin_0 = const()[name = tensor<string, []>("op_31426_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_31426_end_0 = const()[name = tensor<string, []>("op_31426_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_31426_end_mask_0 = const()[name = tensor<string, []>("op_31426_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31426_cast_fp16 = slice_by_index(begin = var_31426_begin_0, end = var_31426_end_0, end_mask = var_31426_end_mask_0, x = q_141_cast_fp16)[name = tensor<string, []>("op_31426_cast_fp16")];
            tensor<int32, [4]> var_31430_begin_0 = const()[name = tensor<string, []>("op_31430_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_31430_end_0 = const()[name = tensor<string, []>("op_31430_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_31430_end_mask_0 = const()[name = tensor<string, []>("op_31430_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31430_cast_fp16 = slice_by_index(begin = var_31430_begin_0, end = var_31430_end_0, end_mask = var_31430_end_mask_0, x = q_141_cast_fp16)[name = tensor<string, []>("op_31430_cast_fp16")];
            tensor<int32, [4]> var_31434_begin_0 = const()[name = tensor<string, []>("op_31434_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_31434_end_0 = const()[name = tensor<string, []>("op_31434_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_31434_end_mask_0 = const()[name = tensor<string, []>("op_31434_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31434_cast_fp16 = slice_by_index(begin = var_31434_begin_0, end = var_31434_end_0, end_mask = var_31434_end_mask_0, x = q_141_cast_fp16)[name = tensor<string, []>("op_31434_cast_fp16")];
            tensor<int32, [4]> var_31438_begin_0 = const()[name = tensor<string, []>("op_31438_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_31438_end_0 = const()[name = tensor<string, []>("op_31438_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_31438_end_mask_0 = const()[name = tensor<string, []>("op_31438_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31438_cast_fp16 = slice_by_index(begin = var_31438_begin_0, end = var_31438_end_0, end_mask = var_31438_end_mask_0, x = q_141_cast_fp16)[name = tensor<string, []>("op_31438_cast_fp16")];
            tensor<int32, [4]> var_31442_begin_0 = const()[name = tensor<string, []>("op_31442_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_31442_end_0 = const()[name = tensor<string, []>("op_31442_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_31442_end_mask_0 = const()[name = tensor<string, []>("op_31442_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31442_cast_fp16 = slice_by_index(begin = var_31442_begin_0, end = var_31442_end_0, end_mask = var_31442_end_mask_0, x = q_141_cast_fp16)[name = tensor<string, []>("op_31442_cast_fp16")];
            tensor<int32, [4]> var_31446_begin_0 = const()[name = tensor<string, []>("op_31446_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_31446_end_0 = const()[name = tensor<string, []>("op_31446_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_31446_end_mask_0 = const()[name = tensor<string, []>("op_31446_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31446_cast_fp16 = slice_by_index(begin = var_31446_begin_0, end = var_31446_end_0, end_mask = var_31446_end_mask_0, x = q_141_cast_fp16)[name = tensor<string, []>("op_31446_cast_fp16")];
            tensor<int32, [4]> var_31450_begin_0 = const()[name = tensor<string, []>("op_31450_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_31450_end_0 = const()[name = tensor<string, []>("op_31450_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_31450_end_mask_0 = const()[name = tensor<string, []>("op_31450_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31450_cast_fp16 = slice_by_index(begin = var_31450_begin_0, end = var_31450_end_0, end_mask = var_31450_end_mask_0, x = q_141_cast_fp16)[name = tensor<string, []>("op_31450_cast_fp16")];
            tensor<int32, [4]> var_31454_begin_0 = const()[name = tensor<string, []>("op_31454_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_31454_end_0 = const()[name = tensor<string, []>("op_31454_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_31454_end_mask_0 = const()[name = tensor<string, []>("op_31454_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31454_cast_fp16 = slice_by_index(begin = var_31454_begin_0, end = var_31454_end_0, end_mask = var_31454_end_mask_0, x = q_141_cast_fp16)[name = tensor<string, []>("op_31454_cast_fp16")];
            tensor<int32, [4]> var_31458_begin_0 = const()[name = tensor<string, []>("op_31458_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_31458_end_0 = const()[name = tensor<string, []>("op_31458_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_31458_end_mask_0 = const()[name = tensor<string, []>("op_31458_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31458_cast_fp16 = slice_by_index(begin = var_31458_begin_0, end = var_31458_end_0, end_mask = var_31458_end_mask_0, x = q_141_cast_fp16)[name = tensor<string, []>("op_31458_cast_fp16")];
            tensor<int32, [4]> var_31462_begin_0 = const()[name = tensor<string, []>("op_31462_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_31462_end_0 = const()[name = tensor<string, []>("op_31462_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_31462_end_mask_0 = const()[name = tensor<string, []>("op_31462_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31462_cast_fp16 = slice_by_index(begin = var_31462_begin_0, end = var_31462_end_0, end_mask = var_31462_end_mask_0, x = q_141_cast_fp16)[name = tensor<string, []>("op_31462_cast_fp16")];
            tensor<int32, [4]> var_31466_begin_0 = const()[name = tensor<string, []>("op_31466_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_31466_end_0 = const()[name = tensor<string, []>("op_31466_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_31466_end_mask_0 = const()[name = tensor<string, []>("op_31466_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31466_cast_fp16 = slice_by_index(begin = var_31466_begin_0, end = var_31466_end_0, end_mask = var_31466_end_mask_0, x = q_141_cast_fp16)[name = tensor<string, []>("op_31466_cast_fp16")];
            tensor<int32, [4]> var_31470_begin_0 = const()[name = tensor<string, []>("op_31470_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_31470_end_0 = const()[name = tensor<string, []>("op_31470_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_31470_end_mask_0 = const()[name = tensor<string, []>("op_31470_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31470_cast_fp16 = slice_by_index(begin = var_31470_begin_0, end = var_31470_end_0, end_mask = var_31470_end_mask_0, x = q_141_cast_fp16)[name = tensor<string, []>("op_31470_cast_fp16")];
            tensor<int32, [4]> var_31474_begin_0 = const()[name = tensor<string, []>("op_31474_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_31474_end_0 = const()[name = tensor<string, []>("op_31474_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_31474_end_mask_0 = const()[name = tensor<string, []>("op_31474_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31474_cast_fp16 = slice_by_index(begin = var_31474_begin_0, end = var_31474_end_0, end_mask = var_31474_end_mask_0, x = q_141_cast_fp16)[name = tensor<string, []>("op_31474_cast_fp16")];
            tensor<int32, [4]> var_31478_begin_0 = const()[name = tensor<string, []>("op_31478_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_31478_end_0 = const()[name = tensor<string, []>("op_31478_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_31478_end_mask_0 = const()[name = tensor<string, []>("op_31478_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31478_cast_fp16 = slice_by_index(begin = var_31478_begin_0, end = var_31478_end_0, end_mask = var_31478_end_mask_0, x = q_141_cast_fp16)[name = tensor<string, []>("op_31478_cast_fp16")];
            tensor<int32, [4]> var_31482_begin_0 = const()[name = tensor<string, []>("op_31482_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_31482_end_0 = const()[name = tensor<string, []>("op_31482_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_31482_end_mask_0 = const()[name = tensor<string, []>("op_31482_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31482_cast_fp16 = slice_by_index(begin = var_31482_begin_0, end = var_31482_end_0, end_mask = var_31482_end_mask_0, x = q_141_cast_fp16)[name = tensor<string, []>("op_31482_cast_fp16")];
            tensor<int32, [4]> var_31486_begin_0 = const()[name = tensor<string, []>("op_31486_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_31486_end_0 = const()[name = tensor<string, []>("op_31486_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_31486_end_mask_0 = const()[name = tensor<string, []>("op_31486_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31486_cast_fp16 = slice_by_index(begin = var_31486_begin_0, end = var_31486_end_0, end_mask = var_31486_end_mask_0, x = q_141_cast_fp16)[name = tensor<string, []>("op_31486_cast_fp16")];
            tensor<int32, [4]> var_31490_begin_0 = const()[name = tensor<string, []>("op_31490_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_31490_end_0 = const()[name = tensor<string, []>("op_31490_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_31490_end_mask_0 = const()[name = tensor<string, []>("op_31490_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31490_cast_fp16 = slice_by_index(begin = var_31490_begin_0, end = var_31490_end_0, end_mask = var_31490_end_mask_0, x = q_141_cast_fp16)[name = tensor<string, []>("op_31490_cast_fp16")];
            tensor<int32, [4]> var_31494_begin_0 = const()[name = tensor<string, []>("op_31494_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_31494_end_0 = const()[name = tensor<string, []>("op_31494_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_31494_end_mask_0 = const()[name = tensor<string, []>("op_31494_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31494_cast_fp16 = slice_by_index(begin = var_31494_begin_0, end = var_31494_end_0, end_mask = var_31494_end_mask_0, x = q_141_cast_fp16)[name = tensor<string, []>("op_31494_cast_fp16")];
            tensor<int32, [4]> k_283_perm_0 = const()[name = tensor<string, []>("k_283_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_31501_begin_0 = const()[name = tensor<string, []>("op_31501_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_31501_end_0 = const()[name = tensor<string, []>("op_31501_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_31501_end_mask_0 = const()[name = tensor<string, []>("op_31501_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> k_283_cast_fp16 = transpose(perm = k_283_perm_0, x = k_281_cast_fp16)[name = tensor<string, []>("transpose_69")];
            tensor<fp16, [2, 1024, 1, 64]> var_31501_cast_fp16 = slice_by_index(begin = var_31501_begin_0, end = var_31501_end_0, end_mask = var_31501_end_mask_0, x = k_283_cast_fp16)[name = tensor<string, []>("op_31501_cast_fp16")];
            tensor<int32, [4]> var_31505_begin_0 = const()[name = tensor<string, []>("op_31505_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_31505_end_0 = const()[name = tensor<string, []>("op_31505_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_31505_end_mask_0 = const()[name = tensor<string, []>("op_31505_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_31505_cast_fp16 = slice_by_index(begin = var_31505_begin_0, end = var_31505_end_0, end_mask = var_31505_end_mask_0, x = k_283_cast_fp16)[name = tensor<string, []>("op_31505_cast_fp16")];
            tensor<int32, [4]> var_31509_begin_0 = const()[name = tensor<string, []>("op_31509_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_31509_end_0 = const()[name = tensor<string, []>("op_31509_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_31509_end_mask_0 = const()[name = tensor<string, []>("op_31509_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_31509_cast_fp16 = slice_by_index(begin = var_31509_begin_0, end = var_31509_end_0, end_mask = var_31509_end_mask_0, x = k_283_cast_fp16)[name = tensor<string, []>("op_31509_cast_fp16")];
            tensor<int32, [4]> var_31513_begin_0 = const()[name = tensor<string, []>("op_31513_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_31513_end_0 = const()[name = tensor<string, []>("op_31513_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_31513_end_mask_0 = const()[name = tensor<string, []>("op_31513_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_31513_cast_fp16 = slice_by_index(begin = var_31513_begin_0, end = var_31513_end_0, end_mask = var_31513_end_mask_0, x = k_283_cast_fp16)[name = tensor<string, []>("op_31513_cast_fp16")];
            tensor<int32, [4]> var_31517_begin_0 = const()[name = tensor<string, []>("op_31517_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_31517_end_0 = const()[name = tensor<string, []>("op_31517_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_31517_end_mask_0 = const()[name = tensor<string, []>("op_31517_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_31517_cast_fp16 = slice_by_index(begin = var_31517_begin_0, end = var_31517_end_0, end_mask = var_31517_end_mask_0, x = k_283_cast_fp16)[name = tensor<string, []>("op_31517_cast_fp16")];
            tensor<int32, [4]> var_31521_begin_0 = const()[name = tensor<string, []>("op_31521_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_31521_end_0 = const()[name = tensor<string, []>("op_31521_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_31521_end_mask_0 = const()[name = tensor<string, []>("op_31521_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_31521_cast_fp16 = slice_by_index(begin = var_31521_begin_0, end = var_31521_end_0, end_mask = var_31521_end_mask_0, x = k_283_cast_fp16)[name = tensor<string, []>("op_31521_cast_fp16")];
            tensor<int32, [4]> var_31525_begin_0 = const()[name = tensor<string, []>("op_31525_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_31525_end_0 = const()[name = tensor<string, []>("op_31525_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_31525_end_mask_0 = const()[name = tensor<string, []>("op_31525_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_31525_cast_fp16 = slice_by_index(begin = var_31525_begin_0, end = var_31525_end_0, end_mask = var_31525_end_mask_0, x = k_283_cast_fp16)[name = tensor<string, []>("op_31525_cast_fp16")];
            tensor<int32, [4]> var_31529_begin_0 = const()[name = tensor<string, []>("op_31529_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_31529_end_0 = const()[name = tensor<string, []>("op_31529_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_31529_end_mask_0 = const()[name = tensor<string, []>("op_31529_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_31529_cast_fp16 = slice_by_index(begin = var_31529_begin_0, end = var_31529_end_0, end_mask = var_31529_end_mask_0, x = k_283_cast_fp16)[name = tensor<string, []>("op_31529_cast_fp16")];
            tensor<int32, [4]> var_31533_begin_0 = const()[name = tensor<string, []>("op_31533_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_31533_end_0 = const()[name = tensor<string, []>("op_31533_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_31533_end_mask_0 = const()[name = tensor<string, []>("op_31533_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_31533_cast_fp16 = slice_by_index(begin = var_31533_begin_0, end = var_31533_end_0, end_mask = var_31533_end_mask_0, x = k_283_cast_fp16)[name = tensor<string, []>("op_31533_cast_fp16")];
            tensor<int32, [4]> var_31537_begin_0 = const()[name = tensor<string, []>("op_31537_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_31537_end_0 = const()[name = tensor<string, []>("op_31537_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_31537_end_mask_0 = const()[name = tensor<string, []>("op_31537_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_31537_cast_fp16 = slice_by_index(begin = var_31537_begin_0, end = var_31537_end_0, end_mask = var_31537_end_mask_0, x = k_283_cast_fp16)[name = tensor<string, []>("op_31537_cast_fp16")];
            tensor<int32, [4]> var_31541_begin_0 = const()[name = tensor<string, []>("op_31541_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_31541_end_0 = const()[name = tensor<string, []>("op_31541_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_31541_end_mask_0 = const()[name = tensor<string, []>("op_31541_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_31541_cast_fp16 = slice_by_index(begin = var_31541_begin_0, end = var_31541_end_0, end_mask = var_31541_end_mask_0, x = k_283_cast_fp16)[name = tensor<string, []>("op_31541_cast_fp16")];
            tensor<int32, [4]> var_31545_begin_0 = const()[name = tensor<string, []>("op_31545_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_31545_end_0 = const()[name = tensor<string, []>("op_31545_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_31545_end_mask_0 = const()[name = tensor<string, []>("op_31545_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_31545_cast_fp16 = slice_by_index(begin = var_31545_begin_0, end = var_31545_end_0, end_mask = var_31545_end_mask_0, x = k_283_cast_fp16)[name = tensor<string, []>("op_31545_cast_fp16")];
            tensor<int32, [4]> var_31549_begin_0 = const()[name = tensor<string, []>("op_31549_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_31549_end_0 = const()[name = tensor<string, []>("op_31549_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_31549_end_mask_0 = const()[name = tensor<string, []>("op_31549_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_31549_cast_fp16 = slice_by_index(begin = var_31549_begin_0, end = var_31549_end_0, end_mask = var_31549_end_mask_0, x = k_283_cast_fp16)[name = tensor<string, []>("op_31549_cast_fp16")];
            tensor<int32, [4]> var_31553_begin_0 = const()[name = tensor<string, []>("op_31553_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_31553_end_0 = const()[name = tensor<string, []>("op_31553_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_31553_end_mask_0 = const()[name = tensor<string, []>("op_31553_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_31553_cast_fp16 = slice_by_index(begin = var_31553_begin_0, end = var_31553_end_0, end_mask = var_31553_end_mask_0, x = k_283_cast_fp16)[name = tensor<string, []>("op_31553_cast_fp16")];
            tensor<int32, [4]> var_31557_begin_0 = const()[name = tensor<string, []>("op_31557_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_31557_end_0 = const()[name = tensor<string, []>("op_31557_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_31557_end_mask_0 = const()[name = tensor<string, []>("op_31557_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_31557_cast_fp16 = slice_by_index(begin = var_31557_begin_0, end = var_31557_end_0, end_mask = var_31557_end_mask_0, x = k_283_cast_fp16)[name = tensor<string, []>("op_31557_cast_fp16")];
            tensor<int32, [4]> var_31561_begin_0 = const()[name = tensor<string, []>("op_31561_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_31561_end_0 = const()[name = tensor<string, []>("op_31561_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_31561_end_mask_0 = const()[name = tensor<string, []>("op_31561_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_31561_cast_fp16 = slice_by_index(begin = var_31561_begin_0, end = var_31561_end_0, end_mask = var_31561_end_mask_0, x = k_283_cast_fp16)[name = tensor<string, []>("op_31561_cast_fp16")];
            tensor<int32, [4]> var_31565_begin_0 = const()[name = tensor<string, []>("op_31565_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_31565_end_0 = const()[name = tensor<string, []>("op_31565_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_31565_end_mask_0 = const()[name = tensor<string, []>("op_31565_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_31565_cast_fp16 = slice_by_index(begin = var_31565_begin_0, end = var_31565_end_0, end_mask = var_31565_end_mask_0, x = k_283_cast_fp16)[name = tensor<string, []>("op_31565_cast_fp16")];
            tensor<int32, [4]> var_31569_begin_0 = const()[name = tensor<string, []>("op_31569_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_31569_end_0 = const()[name = tensor<string, []>("op_31569_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_31569_end_mask_0 = const()[name = tensor<string, []>("op_31569_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_31569_cast_fp16 = slice_by_index(begin = var_31569_begin_0, end = var_31569_end_0, end_mask = var_31569_end_mask_0, x = k_283_cast_fp16)[name = tensor<string, []>("op_31569_cast_fp16")];
            tensor<int32, [4]> var_31573_begin_0 = const()[name = tensor<string, []>("op_31573_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_31573_end_0 = const()[name = tensor<string, []>("op_31573_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_31573_end_mask_0 = const()[name = tensor<string, []>("op_31573_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_31573_cast_fp16 = slice_by_index(begin = var_31573_begin_0, end = var_31573_end_0, end_mask = var_31573_end_mask_0, x = k_283_cast_fp16)[name = tensor<string, []>("op_31573_cast_fp16")];
            tensor<int32, [4]> var_31577_begin_0 = const()[name = tensor<string, []>("op_31577_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_31577_end_0 = const()[name = tensor<string, []>("op_31577_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_31577_end_mask_0 = const()[name = tensor<string, []>("op_31577_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_31577_cast_fp16 = slice_by_index(begin = var_31577_begin_0, end = var_31577_end_0, end_mask = var_31577_end_mask_0, x = k_283_cast_fp16)[name = tensor<string, []>("op_31577_cast_fp16")];
            tensor<int32, [4]> var_31579_begin_0 = const()[name = tensor<string, []>("op_31579_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_31579_end_0 = const()[name = tensor<string, []>("op_31579_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_31579_end_mask_0 = const()[name = tensor<string, []>("op_31579_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31579_cast_fp16 = slice_by_index(begin = var_31579_begin_0, end = var_31579_end_0, end_mask = var_31579_end_mask_0, x = v_141_cast_fp16)[name = tensor<string, []>("op_31579_cast_fp16")];
            tensor<int32, [4]> var_31583_begin_0 = const()[name = tensor<string, []>("op_31583_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_31583_end_0 = const()[name = tensor<string, []>("op_31583_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_31583_end_mask_0 = const()[name = tensor<string, []>("op_31583_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31583_cast_fp16 = slice_by_index(begin = var_31583_begin_0, end = var_31583_end_0, end_mask = var_31583_end_mask_0, x = v_141_cast_fp16)[name = tensor<string, []>("op_31583_cast_fp16")];
            tensor<int32, [4]> var_31587_begin_0 = const()[name = tensor<string, []>("op_31587_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_31587_end_0 = const()[name = tensor<string, []>("op_31587_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_31587_end_mask_0 = const()[name = tensor<string, []>("op_31587_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31587_cast_fp16 = slice_by_index(begin = var_31587_begin_0, end = var_31587_end_0, end_mask = var_31587_end_mask_0, x = v_141_cast_fp16)[name = tensor<string, []>("op_31587_cast_fp16")];
            tensor<int32, [4]> var_31591_begin_0 = const()[name = tensor<string, []>("op_31591_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_31591_end_0 = const()[name = tensor<string, []>("op_31591_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_31591_end_mask_0 = const()[name = tensor<string, []>("op_31591_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31591_cast_fp16 = slice_by_index(begin = var_31591_begin_0, end = var_31591_end_0, end_mask = var_31591_end_mask_0, x = v_141_cast_fp16)[name = tensor<string, []>("op_31591_cast_fp16")];
            tensor<int32, [4]> var_31595_begin_0 = const()[name = tensor<string, []>("op_31595_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_31595_end_0 = const()[name = tensor<string, []>("op_31595_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_31595_end_mask_0 = const()[name = tensor<string, []>("op_31595_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31595_cast_fp16 = slice_by_index(begin = var_31595_begin_0, end = var_31595_end_0, end_mask = var_31595_end_mask_0, x = v_141_cast_fp16)[name = tensor<string, []>("op_31595_cast_fp16")];
            tensor<int32, [4]> var_31599_begin_0 = const()[name = tensor<string, []>("op_31599_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_31599_end_0 = const()[name = tensor<string, []>("op_31599_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_31599_end_mask_0 = const()[name = tensor<string, []>("op_31599_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31599_cast_fp16 = slice_by_index(begin = var_31599_begin_0, end = var_31599_end_0, end_mask = var_31599_end_mask_0, x = v_141_cast_fp16)[name = tensor<string, []>("op_31599_cast_fp16")];
            tensor<int32, [4]> var_31603_begin_0 = const()[name = tensor<string, []>("op_31603_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_31603_end_0 = const()[name = tensor<string, []>("op_31603_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_31603_end_mask_0 = const()[name = tensor<string, []>("op_31603_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31603_cast_fp16 = slice_by_index(begin = var_31603_begin_0, end = var_31603_end_0, end_mask = var_31603_end_mask_0, x = v_141_cast_fp16)[name = tensor<string, []>("op_31603_cast_fp16")];
            tensor<int32, [4]> var_31607_begin_0 = const()[name = tensor<string, []>("op_31607_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_31607_end_0 = const()[name = tensor<string, []>("op_31607_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_31607_end_mask_0 = const()[name = tensor<string, []>("op_31607_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31607_cast_fp16 = slice_by_index(begin = var_31607_begin_0, end = var_31607_end_0, end_mask = var_31607_end_mask_0, x = v_141_cast_fp16)[name = tensor<string, []>("op_31607_cast_fp16")];
            tensor<int32, [4]> var_31611_begin_0 = const()[name = tensor<string, []>("op_31611_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_31611_end_0 = const()[name = tensor<string, []>("op_31611_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_31611_end_mask_0 = const()[name = tensor<string, []>("op_31611_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31611_cast_fp16 = slice_by_index(begin = var_31611_begin_0, end = var_31611_end_0, end_mask = var_31611_end_mask_0, x = v_141_cast_fp16)[name = tensor<string, []>("op_31611_cast_fp16")];
            tensor<int32, [4]> var_31615_begin_0 = const()[name = tensor<string, []>("op_31615_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_31615_end_0 = const()[name = tensor<string, []>("op_31615_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_31615_end_mask_0 = const()[name = tensor<string, []>("op_31615_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31615_cast_fp16 = slice_by_index(begin = var_31615_begin_0, end = var_31615_end_0, end_mask = var_31615_end_mask_0, x = v_141_cast_fp16)[name = tensor<string, []>("op_31615_cast_fp16")];
            tensor<int32, [4]> var_31619_begin_0 = const()[name = tensor<string, []>("op_31619_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_31619_end_0 = const()[name = tensor<string, []>("op_31619_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_31619_end_mask_0 = const()[name = tensor<string, []>("op_31619_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31619_cast_fp16 = slice_by_index(begin = var_31619_begin_0, end = var_31619_end_0, end_mask = var_31619_end_mask_0, x = v_141_cast_fp16)[name = tensor<string, []>("op_31619_cast_fp16")];
            tensor<int32, [4]> var_31623_begin_0 = const()[name = tensor<string, []>("op_31623_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_31623_end_0 = const()[name = tensor<string, []>("op_31623_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_31623_end_mask_0 = const()[name = tensor<string, []>("op_31623_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31623_cast_fp16 = slice_by_index(begin = var_31623_begin_0, end = var_31623_end_0, end_mask = var_31623_end_mask_0, x = v_141_cast_fp16)[name = tensor<string, []>("op_31623_cast_fp16")];
            tensor<int32, [4]> var_31627_begin_0 = const()[name = tensor<string, []>("op_31627_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_31627_end_0 = const()[name = tensor<string, []>("op_31627_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_31627_end_mask_0 = const()[name = tensor<string, []>("op_31627_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31627_cast_fp16 = slice_by_index(begin = var_31627_begin_0, end = var_31627_end_0, end_mask = var_31627_end_mask_0, x = v_141_cast_fp16)[name = tensor<string, []>("op_31627_cast_fp16")];
            tensor<int32, [4]> var_31631_begin_0 = const()[name = tensor<string, []>("op_31631_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_31631_end_0 = const()[name = tensor<string, []>("op_31631_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_31631_end_mask_0 = const()[name = tensor<string, []>("op_31631_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31631_cast_fp16 = slice_by_index(begin = var_31631_begin_0, end = var_31631_end_0, end_mask = var_31631_end_mask_0, x = v_141_cast_fp16)[name = tensor<string, []>("op_31631_cast_fp16")];
            tensor<int32, [4]> var_31635_begin_0 = const()[name = tensor<string, []>("op_31635_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_31635_end_0 = const()[name = tensor<string, []>("op_31635_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_31635_end_mask_0 = const()[name = tensor<string, []>("op_31635_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31635_cast_fp16 = slice_by_index(begin = var_31635_begin_0, end = var_31635_end_0, end_mask = var_31635_end_mask_0, x = v_141_cast_fp16)[name = tensor<string, []>("op_31635_cast_fp16")];
            tensor<int32, [4]> var_31639_begin_0 = const()[name = tensor<string, []>("op_31639_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_31639_end_0 = const()[name = tensor<string, []>("op_31639_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_31639_end_mask_0 = const()[name = tensor<string, []>("op_31639_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31639_cast_fp16 = slice_by_index(begin = var_31639_begin_0, end = var_31639_end_0, end_mask = var_31639_end_mask_0, x = v_141_cast_fp16)[name = tensor<string, []>("op_31639_cast_fp16")];
            tensor<int32, [4]> var_31643_begin_0 = const()[name = tensor<string, []>("op_31643_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_31643_end_0 = const()[name = tensor<string, []>("op_31643_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_31643_end_mask_0 = const()[name = tensor<string, []>("op_31643_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31643_cast_fp16 = slice_by_index(begin = var_31643_begin_0, end = var_31643_end_0, end_mask = var_31643_end_mask_0, x = v_141_cast_fp16)[name = tensor<string, []>("op_31643_cast_fp16")];
            tensor<int32, [4]> var_31647_begin_0 = const()[name = tensor<string, []>("op_31647_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_31647_end_0 = const()[name = tensor<string, []>("op_31647_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_31647_end_mask_0 = const()[name = tensor<string, []>("op_31647_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31647_cast_fp16 = slice_by_index(begin = var_31647_begin_0, end = var_31647_end_0, end_mask = var_31647_end_mask_0, x = v_141_cast_fp16)[name = tensor<string, []>("op_31647_cast_fp16")];
            tensor<int32, [4]> var_31651_begin_0 = const()[name = tensor<string, []>("op_31651_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_31651_end_0 = const()[name = tensor<string, []>("op_31651_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_31651_end_mask_0 = const()[name = tensor<string, []>("op_31651_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31651_cast_fp16 = slice_by_index(begin = var_31651_begin_0, end = var_31651_end_0, end_mask = var_31651_end_mask_0, x = v_141_cast_fp16)[name = tensor<string, []>("op_31651_cast_fp16")];
            tensor<int32, [4]> var_31655_begin_0 = const()[name = tensor<string, []>("op_31655_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_31655_end_0 = const()[name = tensor<string, []>("op_31655_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_31655_end_mask_0 = const()[name = tensor<string, []>("op_31655_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31655_cast_fp16 = slice_by_index(begin = var_31655_begin_0, end = var_31655_end_0, end_mask = var_31655_end_mask_0, x = v_141_cast_fp16)[name = tensor<string, []>("op_31655_cast_fp16")];
            tensor<string, []> var_31659_equation_0 = const()[name = tensor<string, []>("op_31659_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31659_cast_fp16 = einsum(equation = var_31659_equation_0, values = (var_31501_cast_fp16, var_31418_cast_fp16))[name = tensor<string, []>("op_31659_cast_fp16")];
            tensor<fp16, []> var_31660_to_fp16 = const()[name = tensor<string, []>("op_31660_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2641_cast_fp16 = mul(x = var_31659_cast_fp16, y = var_31660_to_fp16)[name = tensor<string, []>("aw_2641_cast_fp16")];
            tensor<string, []> var_31663_equation_0 = const()[name = tensor<string, []>("op_31663_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31663_cast_fp16 = einsum(equation = var_31663_equation_0, values = (var_31505_cast_fp16, var_31422_cast_fp16))[name = tensor<string, []>("op_31663_cast_fp16")];
            tensor<fp16, []> var_31664_to_fp16 = const()[name = tensor<string, []>("op_31664_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2643_cast_fp16 = mul(x = var_31663_cast_fp16, y = var_31664_to_fp16)[name = tensor<string, []>("aw_2643_cast_fp16")];
            tensor<string, []> var_31667_equation_0 = const()[name = tensor<string, []>("op_31667_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31667_cast_fp16 = einsum(equation = var_31667_equation_0, values = (var_31509_cast_fp16, var_31426_cast_fp16))[name = tensor<string, []>("op_31667_cast_fp16")];
            tensor<fp16, []> var_31668_to_fp16 = const()[name = tensor<string, []>("op_31668_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2645_cast_fp16 = mul(x = var_31667_cast_fp16, y = var_31668_to_fp16)[name = tensor<string, []>("aw_2645_cast_fp16")];
            tensor<string, []> var_31671_equation_0 = const()[name = tensor<string, []>("op_31671_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31671_cast_fp16 = einsum(equation = var_31671_equation_0, values = (var_31513_cast_fp16, var_31430_cast_fp16))[name = tensor<string, []>("op_31671_cast_fp16")];
            tensor<fp16, []> var_31672_to_fp16 = const()[name = tensor<string, []>("op_31672_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2647_cast_fp16 = mul(x = var_31671_cast_fp16, y = var_31672_to_fp16)[name = tensor<string, []>("aw_2647_cast_fp16")];
            tensor<string, []> var_31675_equation_0 = const()[name = tensor<string, []>("op_31675_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31675_cast_fp16 = einsum(equation = var_31675_equation_0, values = (var_31517_cast_fp16, var_31434_cast_fp16))[name = tensor<string, []>("op_31675_cast_fp16")];
            tensor<fp16, []> var_31676_to_fp16 = const()[name = tensor<string, []>("op_31676_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2649_cast_fp16 = mul(x = var_31675_cast_fp16, y = var_31676_to_fp16)[name = tensor<string, []>("aw_2649_cast_fp16")];
            tensor<string, []> var_31679_equation_0 = const()[name = tensor<string, []>("op_31679_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31679_cast_fp16 = einsum(equation = var_31679_equation_0, values = (var_31521_cast_fp16, var_31438_cast_fp16))[name = tensor<string, []>("op_31679_cast_fp16")];
            tensor<fp16, []> var_31680_to_fp16 = const()[name = tensor<string, []>("op_31680_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2651_cast_fp16 = mul(x = var_31679_cast_fp16, y = var_31680_to_fp16)[name = tensor<string, []>("aw_2651_cast_fp16")];
            tensor<string, []> var_31683_equation_0 = const()[name = tensor<string, []>("op_31683_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31683_cast_fp16 = einsum(equation = var_31683_equation_0, values = (var_31525_cast_fp16, var_31442_cast_fp16))[name = tensor<string, []>("op_31683_cast_fp16")];
            tensor<fp16, []> var_31684_to_fp16 = const()[name = tensor<string, []>("op_31684_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2653_cast_fp16 = mul(x = var_31683_cast_fp16, y = var_31684_to_fp16)[name = tensor<string, []>("aw_2653_cast_fp16")];
            tensor<string, []> var_31687_equation_0 = const()[name = tensor<string, []>("op_31687_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31687_cast_fp16 = einsum(equation = var_31687_equation_0, values = (var_31529_cast_fp16, var_31446_cast_fp16))[name = tensor<string, []>("op_31687_cast_fp16")];
            tensor<fp16, []> var_31688_to_fp16 = const()[name = tensor<string, []>("op_31688_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2655_cast_fp16 = mul(x = var_31687_cast_fp16, y = var_31688_to_fp16)[name = tensor<string, []>("aw_2655_cast_fp16")];
            tensor<string, []> var_31691_equation_0 = const()[name = tensor<string, []>("op_31691_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31691_cast_fp16 = einsum(equation = var_31691_equation_0, values = (var_31533_cast_fp16, var_31450_cast_fp16))[name = tensor<string, []>("op_31691_cast_fp16")];
            tensor<fp16, []> var_31692_to_fp16 = const()[name = tensor<string, []>("op_31692_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2657_cast_fp16 = mul(x = var_31691_cast_fp16, y = var_31692_to_fp16)[name = tensor<string, []>("aw_2657_cast_fp16")];
            tensor<string, []> var_31695_equation_0 = const()[name = tensor<string, []>("op_31695_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31695_cast_fp16 = einsum(equation = var_31695_equation_0, values = (var_31537_cast_fp16, var_31454_cast_fp16))[name = tensor<string, []>("op_31695_cast_fp16")];
            tensor<fp16, []> var_31696_to_fp16 = const()[name = tensor<string, []>("op_31696_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2659_cast_fp16 = mul(x = var_31695_cast_fp16, y = var_31696_to_fp16)[name = tensor<string, []>("aw_2659_cast_fp16")];
            tensor<string, []> var_31699_equation_0 = const()[name = tensor<string, []>("op_31699_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31699_cast_fp16 = einsum(equation = var_31699_equation_0, values = (var_31541_cast_fp16, var_31458_cast_fp16))[name = tensor<string, []>("op_31699_cast_fp16")];
            tensor<fp16, []> var_31700_to_fp16 = const()[name = tensor<string, []>("op_31700_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2661_cast_fp16 = mul(x = var_31699_cast_fp16, y = var_31700_to_fp16)[name = tensor<string, []>("aw_2661_cast_fp16")];
            tensor<string, []> var_31703_equation_0 = const()[name = tensor<string, []>("op_31703_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31703_cast_fp16 = einsum(equation = var_31703_equation_0, values = (var_31545_cast_fp16, var_31462_cast_fp16))[name = tensor<string, []>("op_31703_cast_fp16")];
            tensor<fp16, []> var_31704_to_fp16 = const()[name = tensor<string, []>("op_31704_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2663_cast_fp16 = mul(x = var_31703_cast_fp16, y = var_31704_to_fp16)[name = tensor<string, []>("aw_2663_cast_fp16")];
            tensor<string, []> var_31707_equation_0 = const()[name = tensor<string, []>("op_31707_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31707_cast_fp16 = einsum(equation = var_31707_equation_0, values = (var_31549_cast_fp16, var_31466_cast_fp16))[name = tensor<string, []>("op_31707_cast_fp16")];
            tensor<fp16, []> var_31708_to_fp16 = const()[name = tensor<string, []>("op_31708_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2665_cast_fp16 = mul(x = var_31707_cast_fp16, y = var_31708_to_fp16)[name = tensor<string, []>("aw_2665_cast_fp16")];
            tensor<string, []> var_31711_equation_0 = const()[name = tensor<string, []>("op_31711_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31711_cast_fp16 = einsum(equation = var_31711_equation_0, values = (var_31553_cast_fp16, var_31470_cast_fp16))[name = tensor<string, []>("op_31711_cast_fp16")];
            tensor<fp16, []> var_31712_to_fp16 = const()[name = tensor<string, []>("op_31712_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2667_cast_fp16 = mul(x = var_31711_cast_fp16, y = var_31712_to_fp16)[name = tensor<string, []>("aw_2667_cast_fp16")];
            tensor<string, []> var_31715_equation_0 = const()[name = tensor<string, []>("op_31715_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31715_cast_fp16 = einsum(equation = var_31715_equation_0, values = (var_31557_cast_fp16, var_31474_cast_fp16))[name = tensor<string, []>("op_31715_cast_fp16")];
            tensor<fp16, []> var_31716_to_fp16 = const()[name = tensor<string, []>("op_31716_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2669_cast_fp16 = mul(x = var_31715_cast_fp16, y = var_31716_to_fp16)[name = tensor<string, []>("aw_2669_cast_fp16")];
            tensor<string, []> var_31719_equation_0 = const()[name = tensor<string, []>("op_31719_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31719_cast_fp16 = einsum(equation = var_31719_equation_0, values = (var_31561_cast_fp16, var_31478_cast_fp16))[name = tensor<string, []>("op_31719_cast_fp16")];
            tensor<fp16, []> var_31720_to_fp16 = const()[name = tensor<string, []>("op_31720_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2671_cast_fp16 = mul(x = var_31719_cast_fp16, y = var_31720_to_fp16)[name = tensor<string, []>("aw_2671_cast_fp16")];
            tensor<string, []> var_31723_equation_0 = const()[name = tensor<string, []>("op_31723_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31723_cast_fp16 = einsum(equation = var_31723_equation_0, values = (var_31565_cast_fp16, var_31482_cast_fp16))[name = tensor<string, []>("op_31723_cast_fp16")];
            tensor<fp16, []> var_31724_to_fp16 = const()[name = tensor<string, []>("op_31724_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2673_cast_fp16 = mul(x = var_31723_cast_fp16, y = var_31724_to_fp16)[name = tensor<string, []>("aw_2673_cast_fp16")];
            tensor<string, []> var_31727_equation_0 = const()[name = tensor<string, []>("op_31727_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31727_cast_fp16 = einsum(equation = var_31727_equation_0, values = (var_31569_cast_fp16, var_31486_cast_fp16))[name = tensor<string, []>("op_31727_cast_fp16")];
            tensor<fp16, []> var_31728_to_fp16 = const()[name = tensor<string, []>("op_31728_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2675_cast_fp16 = mul(x = var_31727_cast_fp16, y = var_31728_to_fp16)[name = tensor<string, []>("aw_2675_cast_fp16")];
            tensor<string, []> var_31731_equation_0 = const()[name = tensor<string, []>("op_31731_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31731_cast_fp16 = einsum(equation = var_31731_equation_0, values = (var_31573_cast_fp16, var_31490_cast_fp16))[name = tensor<string, []>("op_31731_cast_fp16")];
            tensor<fp16, []> var_31732_to_fp16 = const()[name = tensor<string, []>("op_31732_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2677_cast_fp16 = mul(x = var_31731_cast_fp16, y = var_31732_to_fp16)[name = tensor<string, []>("aw_2677_cast_fp16")];
            tensor<string, []> var_31735_equation_0 = const()[name = tensor<string, []>("op_31735_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31735_cast_fp16 = einsum(equation = var_31735_equation_0, values = (var_31577_cast_fp16, var_31494_cast_fp16))[name = tensor<string, []>("op_31735_cast_fp16")];
            tensor<fp16, []> var_31736_to_fp16 = const()[name = tensor<string, []>("op_31736_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2679_cast_fp16 = mul(x = var_31735_cast_fp16, y = var_31736_to_fp16)[name = tensor<string, []>("aw_2679_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31738_cast_fp16 = softmax(axis = var_30355, x = aw_2641_cast_fp16)[name = tensor<string, []>("op_31738_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31739_cast_fp16 = softmax(axis = var_30355, x = aw_2643_cast_fp16)[name = tensor<string, []>("op_31739_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31740_cast_fp16 = softmax(axis = var_30355, x = aw_2645_cast_fp16)[name = tensor<string, []>("op_31740_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31741_cast_fp16 = softmax(axis = var_30355, x = aw_2647_cast_fp16)[name = tensor<string, []>("op_31741_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31742_cast_fp16 = softmax(axis = var_30355, x = aw_2649_cast_fp16)[name = tensor<string, []>("op_31742_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31743_cast_fp16 = softmax(axis = var_30355, x = aw_2651_cast_fp16)[name = tensor<string, []>("op_31743_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31744_cast_fp16 = softmax(axis = var_30355, x = aw_2653_cast_fp16)[name = tensor<string, []>("op_31744_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31745_cast_fp16 = softmax(axis = var_30355, x = aw_2655_cast_fp16)[name = tensor<string, []>("op_31745_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31746_cast_fp16 = softmax(axis = var_30355, x = aw_2657_cast_fp16)[name = tensor<string, []>("op_31746_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31747_cast_fp16 = softmax(axis = var_30355, x = aw_2659_cast_fp16)[name = tensor<string, []>("op_31747_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31748_cast_fp16 = softmax(axis = var_30355, x = aw_2661_cast_fp16)[name = tensor<string, []>("op_31748_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31749_cast_fp16 = softmax(axis = var_30355, x = aw_2663_cast_fp16)[name = tensor<string, []>("op_31749_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31750_cast_fp16 = softmax(axis = var_30355, x = aw_2665_cast_fp16)[name = tensor<string, []>("op_31750_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31751_cast_fp16 = softmax(axis = var_30355, x = aw_2667_cast_fp16)[name = tensor<string, []>("op_31751_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31752_cast_fp16 = softmax(axis = var_30355, x = aw_2669_cast_fp16)[name = tensor<string, []>("op_31752_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31753_cast_fp16 = softmax(axis = var_30355, x = aw_2671_cast_fp16)[name = tensor<string, []>("op_31753_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31754_cast_fp16 = softmax(axis = var_30355, x = aw_2673_cast_fp16)[name = tensor<string, []>("op_31754_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31755_cast_fp16 = softmax(axis = var_30355, x = aw_2675_cast_fp16)[name = tensor<string, []>("op_31755_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31756_cast_fp16 = softmax(axis = var_30355, x = aw_2677_cast_fp16)[name = tensor<string, []>("op_31756_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31757_cast_fp16 = softmax(axis = var_30355, x = aw_2679_cast_fp16)[name = tensor<string, []>("op_31757_cast_fp16")];
            tensor<string, []> var_31759_equation_0 = const()[name = tensor<string, []>("op_31759_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31759_cast_fp16 = einsum(equation = var_31759_equation_0, values = (var_31579_cast_fp16, var_31738_cast_fp16))[name = tensor<string, []>("op_31759_cast_fp16")];
            tensor<string, []> var_31761_equation_0 = const()[name = tensor<string, []>("op_31761_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31761_cast_fp16 = einsum(equation = var_31761_equation_0, values = (var_31583_cast_fp16, var_31739_cast_fp16))[name = tensor<string, []>("op_31761_cast_fp16")];
            tensor<string, []> var_31763_equation_0 = const()[name = tensor<string, []>("op_31763_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31763_cast_fp16 = einsum(equation = var_31763_equation_0, values = (var_31587_cast_fp16, var_31740_cast_fp16))[name = tensor<string, []>("op_31763_cast_fp16")];
            tensor<string, []> var_31765_equation_0 = const()[name = tensor<string, []>("op_31765_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31765_cast_fp16 = einsum(equation = var_31765_equation_0, values = (var_31591_cast_fp16, var_31741_cast_fp16))[name = tensor<string, []>("op_31765_cast_fp16")];
            tensor<string, []> var_31767_equation_0 = const()[name = tensor<string, []>("op_31767_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31767_cast_fp16 = einsum(equation = var_31767_equation_0, values = (var_31595_cast_fp16, var_31742_cast_fp16))[name = tensor<string, []>("op_31767_cast_fp16")];
            tensor<string, []> var_31769_equation_0 = const()[name = tensor<string, []>("op_31769_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31769_cast_fp16 = einsum(equation = var_31769_equation_0, values = (var_31599_cast_fp16, var_31743_cast_fp16))[name = tensor<string, []>("op_31769_cast_fp16")];
            tensor<string, []> var_31771_equation_0 = const()[name = tensor<string, []>("op_31771_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31771_cast_fp16 = einsum(equation = var_31771_equation_0, values = (var_31603_cast_fp16, var_31744_cast_fp16))[name = tensor<string, []>("op_31771_cast_fp16")];
            tensor<string, []> var_31773_equation_0 = const()[name = tensor<string, []>("op_31773_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31773_cast_fp16 = einsum(equation = var_31773_equation_0, values = (var_31607_cast_fp16, var_31745_cast_fp16))[name = tensor<string, []>("op_31773_cast_fp16")];
            tensor<string, []> var_31775_equation_0 = const()[name = tensor<string, []>("op_31775_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31775_cast_fp16 = einsum(equation = var_31775_equation_0, values = (var_31611_cast_fp16, var_31746_cast_fp16))[name = tensor<string, []>("op_31775_cast_fp16")];
            tensor<string, []> var_31777_equation_0 = const()[name = tensor<string, []>("op_31777_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31777_cast_fp16 = einsum(equation = var_31777_equation_0, values = (var_31615_cast_fp16, var_31747_cast_fp16))[name = tensor<string, []>("op_31777_cast_fp16")];
            tensor<string, []> var_31779_equation_0 = const()[name = tensor<string, []>("op_31779_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31779_cast_fp16 = einsum(equation = var_31779_equation_0, values = (var_31619_cast_fp16, var_31748_cast_fp16))[name = tensor<string, []>("op_31779_cast_fp16")];
            tensor<string, []> var_31781_equation_0 = const()[name = tensor<string, []>("op_31781_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31781_cast_fp16 = einsum(equation = var_31781_equation_0, values = (var_31623_cast_fp16, var_31749_cast_fp16))[name = tensor<string, []>("op_31781_cast_fp16")];
            tensor<string, []> var_31783_equation_0 = const()[name = tensor<string, []>("op_31783_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31783_cast_fp16 = einsum(equation = var_31783_equation_0, values = (var_31627_cast_fp16, var_31750_cast_fp16))[name = tensor<string, []>("op_31783_cast_fp16")];
            tensor<string, []> var_31785_equation_0 = const()[name = tensor<string, []>("op_31785_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31785_cast_fp16 = einsum(equation = var_31785_equation_0, values = (var_31631_cast_fp16, var_31751_cast_fp16))[name = tensor<string, []>("op_31785_cast_fp16")];
            tensor<string, []> var_31787_equation_0 = const()[name = tensor<string, []>("op_31787_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31787_cast_fp16 = einsum(equation = var_31787_equation_0, values = (var_31635_cast_fp16, var_31752_cast_fp16))[name = tensor<string, []>("op_31787_cast_fp16")];
            tensor<string, []> var_31789_equation_0 = const()[name = tensor<string, []>("op_31789_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31789_cast_fp16 = einsum(equation = var_31789_equation_0, values = (var_31639_cast_fp16, var_31753_cast_fp16))[name = tensor<string, []>("op_31789_cast_fp16")];
            tensor<string, []> var_31791_equation_0 = const()[name = tensor<string, []>("op_31791_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31791_cast_fp16 = einsum(equation = var_31791_equation_0, values = (var_31643_cast_fp16, var_31754_cast_fp16))[name = tensor<string, []>("op_31791_cast_fp16")];
            tensor<string, []> var_31793_equation_0 = const()[name = tensor<string, []>("op_31793_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31793_cast_fp16 = einsum(equation = var_31793_equation_0, values = (var_31647_cast_fp16, var_31755_cast_fp16))[name = tensor<string, []>("op_31793_cast_fp16")];
            tensor<string, []> var_31795_equation_0 = const()[name = tensor<string, []>("op_31795_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31795_cast_fp16 = einsum(equation = var_31795_equation_0, values = (var_31651_cast_fp16, var_31756_cast_fp16))[name = tensor<string, []>("op_31795_cast_fp16")];
            tensor<string, []> var_31797_equation_0 = const()[name = tensor<string, []>("op_31797_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31797_cast_fp16 = einsum(equation = var_31797_equation_0, values = (var_31655_cast_fp16, var_31757_cast_fp16))[name = tensor<string, []>("op_31797_cast_fp16")];
            tensor<bool, []> input_447_interleave_0 = const()[name = tensor<string, []>("input_447_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_447_cast_fp16 = concat(axis = var_30355, interleave = input_447_interleave_0, values = (var_31759_cast_fp16, var_31761_cast_fp16, var_31763_cast_fp16, var_31765_cast_fp16, var_31767_cast_fp16, var_31769_cast_fp16, var_31771_cast_fp16, var_31773_cast_fp16, var_31775_cast_fp16, var_31777_cast_fp16, var_31779_cast_fp16, var_31781_cast_fp16, var_31783_cast_fp16, var_31785_cast_fp16, var_31787_cast_fp16, var_31789_cast_fp16, var_31791_cast_fp16, var_31793_cast_fp16, var_31795_cast_fp16, var_31797_cast_fp16))[name = tensor<string, []>("input_447_cast_fp16")];
            tensor<string, []> var_31807_pad_type_0 = const()[name = tensor<string, []>("op_31807_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_31807_strides_0 = const()[name = tensor<string, []>("op_31807_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_31807_pad_0 = const()[name = tensor<string, []>("op_31807_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_31807_dilations_0 = const()[name = tensor<string, []>("op_31807_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_31807_groups_0 = const()[name = tensor<string, []>("op_31807_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_1_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(33506880))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34735744))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_1_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_1_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_1_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34735936)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_31807_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_1_attn1_to_out_0_bias_to_fp16, dilations = var_31807_dilations_0, groups = var_31807_groups_0, pad = var_31807_pad_0, pad_type = var_31807_pad_type_0, strides = var_31807_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_1_attn1_to_out_0_weight_to_fp16_palettized, x = input_447_cast_fp16)[name = tensor<string, []>("op_31807_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_213_cast_fp16 = add(x = var_31807_cast_fp16, y = inputs_211_cast_fp16)[name = tensor<string, []>("inputs_213_cast_fp16")];
            tensor<int32, [1]> hidden_states_295_axes_0 = const()[name = tensor<string, []>("hidden_states_295_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_295_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_295_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34738560)))];
            tensor<fp16, [1280]> hidden_states_295_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_295_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34741184)))];
            tensor<fp16, []> var_31817_to_fp16 = const()[name = tensor<string, []>("op_31817_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_295_cast_fp16 = layer_norm(axes = hidden_states_295_axes_0, beta = hidden_states_295_beta_0_to_fp16, epsilon = var_31817_to_fp16, gamma = hidden_states_295_gamma_0_to_fp16, x = inputs_213_cast_fp16)[name = tensor<string, []>("hidden_states_295_cast_fp16")];
            tensor<string, []> q_143_pad_type_0 = const()[name = tensor<string, []>("q_143_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_143_strides_0 = const()[name = tensor<string, []>("q_143_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_143_pad_0 = const()[name = tensor<string, []>("q_143_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_143_dilations_0 = const()[name = tensor<string, []>("q_143_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_143_groups_0 = const()[name = tensor<string, []>("q_143_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_1_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34743808))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35972672))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_1_attn2_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_143_cast_fp16 = conv(dilations = q_143_dilations_0, groups = q_143_groups_0, pad = q_143_pad_0, pad_type = q_143_pad_type_0, strides = q_143_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_1_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_295_cast_fp16)[name = tensor<string, []>("q_143_cast_fp16")];
            tensor<string, []> k_285_pad_type_0 = const()[name = tensor<string, []>("k_285_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_285_strides_0 = const()[name = tensor<string, []>("k_285_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_285_pad_0 = const()[name = tensor<string, []>("k_285_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_285_dilations_0 = const()[name = tensor<string, []>("k_285_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_285_groups_0 = const()[name = tensor<string, []>("k_285_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_1_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35972864))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(37939008))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_1_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> k_285_cast_fp16 = conv(dilations = k_285_dilations_0, groups = k_285_groups_0, pad = k_285_pad_0, pad_type = k_285_pad_type_0, strides = k_285_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_1_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("k_285_cast_fp16")];
            tensor<string, []> v_143_pad_type_0 = const()[name = tensor<string, []>("v_143_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_143_strides_0 = const()[name = tensor<string, []>("v_143_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_143_pad_0 = const()[name = tensor<string, []>("v_143_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_143_dilations_0 = const()[name = tensor<string, []>("v_143_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_143_groups_0 = const()[name = tensor<string, []>("v_143_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_1_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(37939200))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39905344))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_1_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> v_143_cast_fp16 = conv(dilations = v_143_dilations_0, groups = v_143_groups_0, pad = v_143_pad_0, pad_type = v_143_pad_type_0, strides = v_143_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_1_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("v_143_cast_fp16")];
            tensor<int32, [4]> var_31850_begin_0 = const()[name = tensor<string, []>("op_31850_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_31850_end_0 = const()[name = tensor<string, []>("op_31850_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_31850_end_mask_0 = const()[name = tensor<string, []>("op_31850_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31850_cast_fp16 = slice_by_index(begin = var_31850_begin_0, end = var_31850_end_0, end_mask = var_31850_end_mask_0, x = q_143_cast_fp16)[name = tensor<string, []>("op_31850_cast_fp16")];
            tensor<int32, [4]> var_31854_begin_0 = const()[name = tensor<string, []>("op_31854_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_31854_end_0 = const()[name = tensor<string, []>("op_31854_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_31854_end_mask_0 = const()[name = tensor<string, []>("op_31854_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31854_cast_fp16 = slice_by_index(begin = var_31854_begin_0, end = var_31854_end_0, end_mask = var_31854_end_mask_0, x = q_143_cast_fp16)[name = tensor<string, []>("op_31854_cast_fp16")];
            tensor<int32, [4]> var_31858_begin_0 = const()[name = tensor<string, []>("op_31858_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_31858_end_0 = const()[name = tensor<string, []>("op_31858_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_31858_end_mask_0 = const()[name = tensor<string, []>("op_31858_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31858_cast_fp16 = slice_by_index(begin = var_31858_begin_0, end = var_31858_end_0, end_mask = var_31858_end_mask_0, x = q_143_cast_fp16)[name = tensor<string, []>("op_31858_cast_fp16")];
            tensor<int32, [4]> var_31862_begin_0 = const()[name = tensor<string, []>("op_31862_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_31862_end_0 = const()[name = tensor<string, []>("op_31862_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_31862_end_mask_0 = const()[name = tensor<string, []>("op_31862_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31862_cast_fp16 = slice_by_index(begin = var_31862_begin_0, end = var_31862_end_0, end_mask = var_31862_end_mask_0, x = q_143_cast_fp16)[name = tensor<string, []>("op_31862_cast_fp16")];
            tensor<int32, [4]> var_31866_begin_0 = const()[name = tensor<string, []>("op_31866_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_31866_end_0 = const()[name = tensor<string, []>("op_31866_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_31866_end_mask_0 = const()[name = tensor<string, []>("op_31866_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31866_cast_fp16 = slice_by_index(begin = var_31866_begin_0, end = var_31866_end_0, end_mask = var_31866_end_mask_0, x = q_143_cast_fp16)[name = tensor<string, []>("op_31866_cast_fp16")];
            tensor<int32, [4]> var_31870_begin_0 = const()[name = tensor<string, []>("op_31870_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_31870_end_0 = const()[name = tensor<string, []>("op_31870_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_31870_end_mask_0 = const()[name = tensor<string, []>("op_31870_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31870_cast_fp16 = slice_by_index(begin = var_31870_begin_0, end = var_31870_end_0, end_mask = var_31870_end_mask_0, x = q_143_cast_fp16)[name = tensor<string, []>("op_31870_cast_fp16")];
            tensor<int32, [4]> var_31874_begin_0 = const()[name = tensor<string, []>("op_31874_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_31874_end_0 = const()[name = tensor<string, []>("op_31874_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_31874_end_mask_0 = const()[name = tensor<string, []>("op_31874_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31874_cast_fp16 = slice_by_index(begin = var_31874_begin_0, end = var_31874_end_0, end_mask = var_31874_end_mask_0, x = q_143_cast_fp16)[name = tensor<string, []>("op_31874_cast_fp16")];
            tensor<int32, [4]> var_31878_begin_0 = const()[name = tensor<string, []>("op_31878_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_31878_end_0 = const()[name = tensor<string, []>("op_31878_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_31878_end_mask_0 = const()[name = tensor<string, []>("op_31878_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31878_cast_fp16 = slice_by_index(begin = var_31878_begin_0, end = var_31878_end_0, end_mask = var_31878_end_mask_0, x = q_143_cast_fp16)[name = tensor<string, []>("op_31878_cast_fp16")];
            tensor<int32, [4]> var_31882_begin_0 = const()[name = tensor<string, []>("op_31882_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_31882_end_0 = const()[name = tensor<string, []>("op_31882_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_31882_end_mask_0 = const()[name = tensor<string, []>("op_31882_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31882_cast_fp16 = slice_by_index(begin = var_31882_begin_0, end = var_31882_end_0, end_mask = var_31882_end_mask_0, x = q_143_cast_fp16)[name = tensor<string, []>("op_31882_cast_fp16")];
            tensor<int32, [4]> var_31886_begin_0 = const()[name = tensor<string, []>("op_31886_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_31886_end_0 = const()[name = tensor<string, []>("op_31886_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_31886_end_mask_0 = const()[name = tensor<string, []>("op_31886_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31886_cast_fp16 = slice_by_index(begin = var_31886_begin_0, end = var_31886_end_0, end_mask = var_31886_end_mask_0, x = q_143_cast_fp16)[name = tensor<string, []>("op_31886_cast_fp16")];
            tensor<int32, [4]> var_31890_begin_0 = const()[name = tensor<string, []>("op_31890_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_31890_end_0 = const()[name = tensor<string, []>("op_31890_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_31890_end_mask_0 = const()[name = tensor<string, []>("op_31890_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31890_cast_fp16 = slice_by_index(begin = var_31890_begin_0, end = var_31890_end_0, end_mask = var_31890_end_mask_0, x = q_143_cast_fp16)[name = tensor<string, []>("op_31890_cast_fp16")];
            tensor<int32, [4]> var_31894_begin_0 = const()[name = tensor<string, []>("op_31894_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_31894_end_0 = const()[name = tensor<string, []>("op_31894_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_31894_end_mask_0 = const()[name = tensor<string, []>("op_31894_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31894_cast_fp16 = slice_by_index(begin = var_31894_begin_0, end = var_31894_end_0, end_mask = var_31894_end_mask_0, x = q_143_cast_fp16)[name = tensor<string, []>("op_31894_cast_fp16")];
            tensor<int32, [4]> var_31898_begin_0 = const()[name = tensor<string, []>("op_31898_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_31898_end_0 = const()[name = tensor<string, []>("op_31898_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_31898_end_mask_0 = const()[name = tensor<string, []>("op_31898_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31898_cast_fp16 = slice_by_index(begin = var_31898_begin_0, end = var_31898_end_0, end_mask = var_31898_end_mask_0, x = q_143_cast_fp16)[name = tensor<string, []>("op_31898_cast_fp16")];
            tensor<int32, [4]> var_31902_begin_0 = const()[name = tensor<string, []>("op_31902_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_31902_end_0 = const()[name = tensor<string, []>("op_31902_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_31902_end_mask_0 = const()[name = tensor<string, []>("op_31902_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31902_cast_fp16 = slice_by_index(begin = var_31902_begin_0, end = var_31902_end_0, end_mask = var_31902_end_mask_0, x = q_143_cast_fp16)[name = tensor<string, []>("op_31902_cast_fp16")];
            tensor<int32, [4]> var_31906_begin_0 = const()[name = tensor<string, []>("op_31906_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_31906_end_0 = const()[name = tensor<string, []>("op_31906_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_31906_end_mask_0 = const()[name = tensor<string, []>("op_31906_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31906_cast_fp16 = slice_by_index(begin = var_31906_begin_0, end = var_31906_end_0, end_mask = var_31906_end_mask_0, x = q_143_cast_fp16)[name = tensor<string, []>("op_31906_cast_fp16")];
            tensor<int32, [4]> var_31910_begin_0 = const()[name = tensor<string, []>("op_31910_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_31910_end_0 = const()[name = tensor<string, []>("op_31910_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_31910_end_mask_0 = const()[name = tensor<string, []>("op_31910_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31910_cast_fp16 = slice_by_index(begin = var_31910_begin_0, end = var_31910_end_0, end_mask = var_31910_end_mask_0, x = q_143_cast_fp16)[name = tensor<string, []>("op_31910_cast_fp16")];
            tensor<int32, [4]> var_31914_begin_0 = const()[name = tensor<string, []>("op_31914_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_31914_end_0 = const()[name = tensor<string, []>("op_31914_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_31914_end_mask_0 = const()[name = tensor<string, []>("op_31914_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31914_cast_fp16 = slice_by_index(begin = var_31914_begin_0, end = var_31914_end_0, end_mask = var_31914_end_mask_0, x = q_143_cast_fp16)[name = tensor<string, []>("op_31914_cast_fp16")];
            tensor<int32, [4]> var_31918_begin_0 = const()[name = tensor<string, []>("op_31918_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_31918_end_0 = const()[name = tensor<string, []>("op_31918_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_31918_end_mask_0 = const()[name = tensor<string, []>("op_31918_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31918_cast_fp16 = slice_by_index(begin = var_31918_begin_0, end = var_31918_end_0, end_mask = var_31918_end_mask_0, x = q_143_cast_fp16)[name = tensor<string, []>("op_31918_cast_fp16")];
            tensor<int32, [4]> var_31922_begin_0 = const()[name = tensor<string, []>("op_31922_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_31922_end_0 = const()[name = tensor<string, []>("op_31922_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_31922_end_mask_0 = const()[name = tensor<string, []>("op_31922_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31922_cast_fp16 = slice_by_index(begin = var_31922_begin_0, end = var_31922_end_0, end_mask = var_31922_end_mask_0, x = q_143_cast_fp16)[name = tensor<string, []>("op_31922_cast_fp16")];
            tensor<int32, [4]> var_31926_begin_0 = const()[name = tensor<string, []>("op_31926_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_31926_end_0 = const()[name = tensor<string, []>("op_31926_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_31926_end_mask_0 = const()[name = tensor<string, []>("op_31926_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31926_cast_fp16 = slice_by_index(begin = var_31926_begin_0, end = var_31926_end_0, end_mask = var_31926_end_mask_0, x = q_143_cast_fp16)[name = tensor<string, []>("op_31926_cast_fp16")];
            tensor<int32, [4]> k_287_perm_0 = const()[name = tensor<string, []>("k_287_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_31933_begin_0 = const()[name = tensor<string, []>("op_31933_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_31933_end_0 = const()[name = tensor<string, []>("op_31933_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_31933_end_mask_0 = const()[name = tensor<string, []>("op_31933_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> k_287_cast_fp16 = transpose(perm = k_287_perm_0, x = k_285_cast_fp16)[name = tensor<string, []>("transpose_68")];
            tensor<fp16, [2, 77, 1, 64]> var_31933_cast_fp16 = slice_by_index(begin = var_31933_begin_0, end = var_31933_end_0, end_mask = var_31933_end_mask_0, x = k_287_cast_fp16)[name = tensor<string, []>("op_31933_cast_fp16")];
            tensor<int32, [4]> var_31937_begin_0 = const()[name = tensor<string, []>("op_31937_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_31937_end_0 = const()[name = tensor<string, []>("op_31937_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_31937_end_mask_0 = const()[name = tensor<string, []>("op_31937_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31937_cast_fp16 = slice_by_index(begin = var_31937_begin_0, end = var_31937_end_0, end_mask = var_31937_end_mask_0, x = k_287_cast_fp16)[name = tensor<string, []>("op_31937_cast_fp16")];
            tensor<int32, [4]> var_31941_begin_0 = const()[name = tensor<string, []>("op_31941_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_31941_end_0 = const()[name = tensor<string, []>("op_31941_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_31941_end_mask_0 = const()[name = tensor<string, []>("op_31941_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31941_cast_fp16 = slice_by_index(begin = var_31941_begin_0, end = var_31941_end_0, end_mask = var_31941_end_mask_0, x = k_287_cast_fp16)[name = tensor<string, []>("op_31941_cast_fp16")];
            tensor<int32, [4]> var_31945_begin_0 = const()[name = tensor<string, []>("op_31945_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_31945_end_0 = const()[name = tensor<string, []>("op_31945_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_31945_end_mask_0 = const()[name = tensor<string, []>("op_31945_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31945_cast_fp16 = slice_by_index(begin = var_31945_begin_0, end = var_31945_end_0, end_mask = var_31945_end_mask_0, x = k_287_cast_fp16)[name = tensor<string, []>("op_31945_cast_fp16")];
            tensor<int32, [4]> var_31949_begin_0 = const()[name = tensor<string, []>("op_31949_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_31949_end_0 = const()[name = tensor<string, []>("op_31949_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_31949_end_mask_0 = const()[name = tensor<string, []>("op_31949_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31949_cast_fp16 = slice_by_index(begin = var_31949_begin_0, end = var_31949_end_0, end_mask = var_31949_end_mask_0, x = k_287_cast_fp16)[name = tensor<string, []>("op_31949_cast_fp16")];
            tensor<int32, [4]> var_31953_begin_0 = const()[name = tensor<string, []>("op_31953_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_31953_end_0 = const()[name = tensor<string, []>("op_31953_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_31953_end_mask_0 = const()[name = tensor<string, []>("op_31953_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31953_cast_fp16 = slice_by_index(begin = var_31953_begin_0, end = var_31953_end_0, end_mask = var_31953_end_mask_0, x = k_287_cast_fp16)[name = tensor<string, []>("op_31953_cast_fp16")];
            tensor<int32, [4]> var_31957_begin_0 = const()[name = tensor<string, []>("op_31957_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_31957_end_0 = const()[name = tensor<string, []>("op_31957_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_31957_end_mask_0 = const()[name = tensor<string, []>("op_31957_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31957_cast_fp16 = slice_by_index(begin = var_31957_begin_0, end = var_31957_end_0, end_mask = var_31957_end_mask_0, x = k_287_cast_fp16)[name = tensor<string, []>("op_31957_cast_fp16")];
            tensor<int32, [4]> var_31961_begin_0 = const()[name = tensor<string, []>("op_31961_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_31961_end_0 = const()[name = tensor<string, []>("op_31961_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_31961_end_mask_0 = const()[name = tensor<string, []>("op_31961_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31961_cast_fp16 = slice_by_index(begin = var_31961_begin_0, end = var_31961_end_0, end_mask = var_31961_end_mask_0, x = k_287_cast_fp16)[name = tensor<string, []>("op_31961_cast_fp16")];
            tensor<int32, [4]> var_31965_begin_0 = const()[name = tensor<string, []>("op_31965_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_31965_end_0 = const()[name = tensor<string, []>("op_31965_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_31965_end_mask_0 = const()[name = tensor<string, []>("op_31965_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31965_cast_fp16 = slice_by_index(begin = var_31965_begin_0, end = var_31965_end_0, end_mask = var_31965_end_mask_0, x = k_287_cast_fp16)[name = tensor<string, []>("op_31965_cast_fp16")];
            tensor<int32, [4]> var_31969_begin_0 = const()[name = tensor<string, []>("op_31969_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_31969_end_0 = const()[name = tensor<string, []>("op_31969_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_31969_end_mask_0 = const()[name = tensor<string, []>("op_31969_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31969_cast_fp16 = slice_by_index(begin = var_31969_begin_0, end = var_31969_end_0, end_mask = var_31969_end_mask_0, x = k_287_cast_fp16)[name = tensor<string, []>("op_31969_cast_fp16")];
            tensor<int32, [4]> var_31973_begin_0 = const()[name = tensor<string, []>("op_31973_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_31973_end_0 = const()[name = tensor<string, []>("op_31973_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_31973_end_mask_0 = const()[name = tensor<string, []>("op_31973_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31973_cast_fp16 = slice_by_index(begin = var_31973_begin_0, end = var_31973_end_0, end_mask = var_31973_end_mask_0, x = k_287_cast_fp16)[name = tensor<string, []>("op_31973_cast_fp16")];
            tensor<int32, [4]> var_31977_begin_0 = const()[name = tensor<string, []>("op_31977_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_31977_end_0 = const()[name = tensor<string, []>("op_31977_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_31977_end_mask_0 = const()[name = tensor<string, []>("op_31977_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31977_cast_fp16 = slice_by_index(begin = var_31977_begin_0, end = var_31977_end_0, end_mask = var_31977_end_mask_0, x = k_287_cast_fp16)[name = tensor<string, []>("op_31977_cast_fp16")];
            tensor<int32, [4]> var_31981_begin_0 = const()[name = tensor<string, []>("op_31981_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_31981_end_0 = const()[name = tensor<string, []>("op_31981_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_31981_end_mask_0 = const()[name = tensor<string, []>("op_31981_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31981_cast_fp16 = slice_by_index(begin = var_31981_begin_0, end = var_31981_end_0, end_mask = var_31981_end_mask_0, x = k_287_cast_fp16)[name = tensor<string, []>("op_31981_cast_fp16")];
            tensor<int32, [4]> var_31985_begin_0 = const()[name = tensor<string, []>("op_31985_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_31985_end_0 = const()[name = tensor<string, []>("op_31985_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_31985_end_mask_0 = const()[name = tensor<string, []>("op_31985_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31985_cast_fp16 = slice_by_index(begin = var_31985_begin_0, end = var_31985_end_0, end_mask = var_31985_end_mask_0, x = k_287_cast_fp16)[name = tensor<string, []>("op_31985_cast_fp16")];
            tensor<int32, [4]> var_31989_begin_0 = const()[name = tensor<string, []>("op_31989_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_31989_end_0 = const()[name = tensor<string, []>("op_31989_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_31989_end_mask_0 = const()[name = tensor<string, []>("op_31989_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31989_cast_fp16 = slice_by_index(begin = var_31989_begin_0, end = var_31989_end_0, end_mask = var_31989_end_mask_0, x = k_287_cast_fp16)[name = tensor<string, []>("op_31989_cast_fp16")];
            tensor<int32, [4]> var_31993_begin_0 = const()[name = tensor<string, []>("op_31993_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_31993_end_0 = const()[name = tensor<string, []>("op_31993_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_31993_end_mask_0 = const()[name = tensor<string, []>("op_31993_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31993_cast_fp16 = slice_by_index(begin = var_31993_begin_0, end = var_31993_end_0, end_mask = var_31993_end_mask_0, x = k_287_cast_fp16)[name = tensor<string, []>("op_31993_cast_fp16")];
            tensor<int32, [4]> var_31997_begin_0 = const()[name = tensor<string, []>("op_31997_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_31997_end_0 = const()[name = tensor<string, []>("op_31997_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_31997_end_mask_0 = const()[name = tensor<string, []>("op_31997_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31997_cast_fp16 = slice_by_index(begin = var_31997_begin_0, end = var_31997_end_0, end_mask = var_31997_end_mask_0, x = k_287_cast_fp16)[name = tensor<string, []>("op_31997_cast_fp16")];
            tensor<int32, [4]> var_32001_begin_0 = const()[name = tensor<string, []>("op_32001_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_32001_end_0 = const()[name = tensor<string, []>("op_32001_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_32001_end_mask_0 = const()[name = tensor<string, []>("op_32001_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_32001_cast_fp16 = slice_by_index(begin = var_32001_begin_0, end = var_32001_end_0, end_mask = var_32001_end_mask_0, x = k_287_cast_fp16)[name = tensor<string, []>("op_32001_cast_fp16")];
            tensor<int32, [4]> var_32005_begin_0 = const()[name = tensor<string, []>("op_32005_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_32005_end_0 = const()[name = tensor<string, []>("op_32005_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_32005_end_mask_0 = const()[name = tensor<string, []>("op_32005_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_32005_cast_fp16 = slice_by_index(begin = var_32005_begin_0, end = var_32005_end_0, end_mask = var_32005_end_mask_0, x = k_287_cast_fp16)[name = tensor<string, []>("op_32005_cast_fp16")];
            tensor<int32, [4]> var_32009_begin_0 = const()[name = tensor<string, []>("op_32009_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_32009_end_0 = const()[name = tensor<string, []>("op_32009_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_32009_end_mask_0 = const()[name = tensor<string, []>("op_32009_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_32009_cast_fp16 = slice_by_index(begin = var_32009_begin_0, end = var_32009_end_0, end_mask = var_32009_end_mask_0, x = k_287_cast_fp16)[name = tensor<string, []>("op_32009_cast_fp16")];
            tensor<int32, [4]> var_32011_begin_0 = const()[name = tensor<string, []>("op_32011_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_32011_end_0 = const()[name = tensor<string, []>("op_32011_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_32011_end_mask_0 = const()[name = tensor<string, []>("op_32011_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32011_cast_fp16 = slice_by_index(begin = var_32011_begin_0, end = var_32011_end_0, end_mask = var_32011_end_mask_0, x = v_143_cast_fp16)[name = tensor<string, []>("op_32011_cast_fp16")];
            tensor<int32, [4]> var_32015_begin_0 = const()[name = tensor<string, []>("op_32015_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_32015_end_0 = const()[name = tensor<string, []>("op_32015_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_32015_end_mask_0 = const()[name = tensor<string, []>("op_32015_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32015_cast_fp16 = slice_by_index(begin = var_32015_begin_0, end = var_32015_end_0, end_mask = var_32015_end_mask_0, x = v_143_cast_fp16)[name = tensor<string, []>("op_32015_cast_fp16")];
            tensor<int32, [4]> var_32019_begin_0 = const()[name = tensor<string, []>("op_32019_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_32019_end_0 = const()[name = tensor<string, []>("op_32019_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_32019_end_mask_0 = const()[name = tensor<string, []>("op_32019_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32019_cast_fp16 = slice_by_index(begin = var_32019_begin_0, end = var_32019_end_0, end_mask = var_32019_end_mask_0, x = v_143_cast_fp16)[name = tensor<string, []>("op_32019_cast_fp16")];
            tensor<int32, [4]> var_32023_begin_0 = const()[name = tensor<string, []>("op_32023_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_32023_end_0 = const()[name = tensor<string, []>("op_32023_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_32023_end_mask_0 = const()[name = tensor<string, []>("op_32023_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32023_cast_fp16 = slice_by_index(begin = var_32023_begin_0, end = var_32023_end_0, end_mask = var_32023_end_mask_0, x = v_143_cast_fp16)[name = tensor<string, []>("op_32023_cast_fp16")];
            tensor<int32, [4]> var_32027_begin_0 = const()[name = tensor<string, []>("op_32027_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_32027_end_0 = const()[name = tensor<string, []>("op_32027_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_32027_end_mask_0 = const()[name = tensor<string, []>("op_32027_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32027_cast_fp16 = slice_by_index(begin = var_32027_begin_0, end = var_32027_end_0, end_mask = var_32027_end_mask_0, x = v_143_cast_fp16)[name = tensor<string, []>("op_32027_cast_fp16")];
            tensor<int32, [4]> var_32031_begin_0 = const()[name = tensor<string, []>("op_32031_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_32031_end_0 = const()[name = tensor<string, []>("op_32031_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_32031_end_mask_0 = const()[name = tensor<string, []>("op_32031_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32031_cast_fp16 = slice_by_index(begin = var_32031_begin_0, end = var_32031_end_0, end_mask = var_32031_end_mask_0, x = v_143_cast_fp16)[name = tensor<string, []>("op_32031_cast_fp16")];
            tensor<int32, [4]> var_32035_begin_0 = const()[name = tensor<string, []>("op_32035_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_32035_end_0 = const()[name = tensor<string, []>("op_32035_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_32035_end_mask_0 = const()[name = tensor<string, []>("op_32035_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32035_cast_fp16 = slice_by_index(begin = var_32035_begin_0, end = var_32035_end_0, end_mask = var_32035_end_mask_0, x = v_143_cast_fp16)[name = tensor<string, []>("op_32035_cast_fp16")];
            tensor<int32, [4]> var_32039_begin_0 = const()[name = tensor<string, []>("op_32039_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_32039_end_0 = const()[name = tensor<string, []>("op_32039_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_32039_end_mask_0 = const()[name = tensor<string, []>("op_32039_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32039_cast_fp16 = slice_by_index(begin = var_32039_begin_0, end = var_32039_end_0, end_mask = var_32039_end_mask_0, x = v_143_cast_fp16)[name = tensor<string, []>("op_32039_cast_fp16")];
            tensor<int32, [4]> var_32043_begin_0 = const()[name = tensor<string, []>("op_32043_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_32043_end_0 = const()[name = tensor<string, []>("op_32043_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_32043_end_mask_0 = const()[name = tensor<string, []>("op_32043_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32043_cast_fp16 = slice_by_index(begin = var_32043_begin_0, end = var_32043_end_0, end_mask = var_32043_end_mask_0, x = v_143_cast_fp16)[name = tensor<string, []>("op_32043_cast_fp16")];
            tensor<int32, [4]> var_32047_begin_0 = const()[name = tensor<string, []>("op_32047_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_32047_end_0 = const()[name = tensor<string, []>("op_32047_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_32047_end_mask_0 = const()[name = tensor<string, []>("op_32047_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32047_cast_fp16 = slice_by_index(begin = var_32047_begin_0, end = var_32047_end_0, end_mask = var_32047_end_mask_0, x = v_143_cast_fp16)[name = tensor<string, []>("op_32047_cast_fp16")];
            tensor<int32, [4]> var_32051_begin_0 = const()[name = tensor<string, []>("op_32051_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_32051_end_0 = const()[name = tensor<string, []>("op_32051_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_32051_end_mask_0 = const()[name = tensor<string, []>("op_32051_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32051_cast_fp16 = slice_by_index(begin = var_32051_begin_0, end = var_32051_end_0, end_mask = var_32051_end_mask_0, x = v_143_cast_fp16)[name = tensor<string, []>("op_32051_cast_fp16")];
            tensor<int32, [4]> var_32055_begin_0 = const()[name = tensor<string, []>("op_32055_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_32055_end_0 = const()[name = tensor<string, []>("op_32055_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_32055_end_mask_0 = const()[name = tensor<string, []>("op_32055_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32055_cast_fp16 = slice_by_index(begin = var_32055_begin_0, end = var_32055_end_0, end_mask = var_32055_end_mask_0, x = v_143_cast_fp16)[name = tensor<string, []>("op_32055_cast_fp16")];
            tensor<int32, [4]> var_32059_begin_0 = const()[name = tensor<string, []>("op_32059_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_32059_end_0 = const()[name = tensor<string, []>("op_32059_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_32059_end_mask_0 = const()[name = tensor<string, []>("op_32059_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32059_cast_fp16 = slice_by_index(begin = var_32059_begin_0, end = var_32059_end_0, end_mask = var_32059_end_mask_0, x = v_143_cast_fp16)[name = tensor<string, []>("op_32059_cast_fp16")];
            tensor<int32, [4]> var_32063_begin_0 = const()[name = tensor<string, []>("op_32063_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_32063_end_0 = const()[name = tensor<string, []>("op_32063_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_32063_end_mask_0 = const()[name = tensor<string, []>("op_32063_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32063_cast_fp16 = slice_by_index(begin = var_32063_begin_0, end = var_32063_end_0, end_mask = var_32063_end_mask_0, x = v_143_cast_fp16)[name = tensor<string, []>("op_32063_cast_fp16")];
            tensor<int32, [4]> var_32067_begin_0 = const()[name = tensor<string, []>("op_32067_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_32067_end_0 = const()[name = tensor<string, []>("op_32067_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_32067_end_mask_0 = const()[name = tensor<string, []>("op_32067_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32067_cast_fp16 = slice_by_index(begin = var_32067_begin_0, end = var_32067_end_0, end_mask = var_32067_end_mask_0, x = v_143_cast_fp16)[name = tensor<string, []>("op_32067_cast_fp16")];
            tensor<int32, [4]> var_32071_begin_0 = const()[name = tensor<string, []>("op_32071_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_32071_end_0 = const()[name = tensor<string, []>("op_32071_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_32071_end_mask_0 = const()[name = tensor<string, []>("op_32071_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32071_cast_fp16 = slice_by_index(begin = var_32071_begin_0, end = var_32071_end_0, end_mask = var_32071_end_mask_0, x = v_143_cast_fp16)[name = tensor<string, []>("op_32071_cast_fp16")];
            tensor<int32, [4]> var_32075_begin_0 = const()[name = tensor<string, []>("op_32075_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_32075_end_0 = const()[name = tensor<string, []>("op_32075_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_32075_end_mask_0 = const()[name = tensor<string, []>("op_32075_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32075_cast_fp16 = slice_by_index(begin = var_32075_begin_0, end = var_32075_end_0, end_mask = var_32075_end_mask_0, x = v_143_cast_fp16)[name = tensor<string, []>("op_32075_cast_fp16")];
            tensor<int32, [4]> var_32079_begin_0 = const()[name = tensor<string, []>("op_32079_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_32079_end_0 = const()[name = tensor<string, []>("op_32079_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_32079_end_mask_0 = const()[name = tensor<string, []>("op_32079_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32079_cast_fp16 = slice_by_index(begin = var_32079_begin_0, end = var_32079_end_0, end_mask = var_32079_end_mask_0, x = v_143_cast_fp16)[name = tensor<string, []>("op_32079_cast_fp16")];
            tensor<int32, [4]> var_32083_begin_0 = const()[name = tensor<string, []>("op_32083_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_32083_end_0 = const()[name = tensor<string, []>("op_32083_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_32083_end_mask_0 = const()[name = tensor<string, []>("op_32083_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32083_cast_fp16 = slice_by_index(begin = var_32083_begin_0, end = var_32083_end_0, end_mask = var_32083_end_mask_0, x = v_143_cast_fp16)[name = tensor<string, []>("op_32083_cast_fp16")];
            tensor<int32, [4]> var_32087_begin_0 = const()[name = tensor<string, []>("op_32087_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_32087_end_0 = const()[name = tensor<string, []>("op_32087_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_32087_end_mask_0 = const()[name = tensor<string, []>("op_32087_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32087_cast_fp16 = slice_by_index(begin = var_32087_begin_0, end = var_32087_end_0, end_mask = var_32087_end_mask_0, x = v_143_cast_fp16)[name = tensor<string, []>("op_32087_cast_fp16")];
            tensor<string, []> var_32091_equation_0 = const()[name = tensor<string, []>("op_32091_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_32091_cast_fp16 = einsum(equation = var_32091_equation_0, values = (var_31933_cast_fp16, var_31850_cast_fp16))[name = tensor<string, []>("op_32091_cast_fp16")];
            tensor<fp16, []> var_32092_to_fp16 = const()[name = tensor<string, []>("op_32092_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2681_cast_fp16 = mul(x = var_32091_cast_fp16, y = var_32092_to_fp16)[name = tensor<string, []>("aw_2681_cast_fp16")];
            tensor<string, []> var_32095_equation_0 = const()[name = tensor<string, []>("op_32095_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_32095_cast_fp16 = einsum(equation = var_32095_equation_0, values = (var_31937_cast_fp16, var_31854_cast_fp16))[name = tensor<string, []>("op_32095_cast_fp16")];
            tensor<fp16, []> var_32096_to_fp16 = const()[name = tensor<string, []>("op_32096_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2683_cast_fp16 = mul(x = var_32095_cast_fp16, y = var_32096_to_fp16)[name = tensor<string, []>("aw_2683_cast_fp16")];
            tensor<string, []> var_32099_equation_0 = const()[name = tensor<string, []>("op_32099_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_32099_cast_fp16 = einsum(equation = var_32099_equation_0, values = (var_31941_cast_fp16, var_31858_cast_fp16))[name = tensor<string, []>("op_32099_cast_fp16")];
            tensor<fp16, []> var_32100_to_fp16 = const()[name = tensor<string, []>("op_32100_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2685_cast_fp16 = mul(x = var_32099_cast_fp16, y = var_32100_to_fp16)[name = tensor<string, []>("aw_2685_cast_fp16")];
            tensor<string, []> var_32103_equation_0 = const()[name = tensor<string, []>("op_32103_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_32103_cast_fp16 = einsum(equation = var_32103_equation_0, values = (var_31945_cast_fp16, var_31862_cast_fp16))[name = tensor<string, []>("op_32103_cast_fp16")];
            tensor<fp16, []> var_32104_to_fp16 = const()[name = tensor<string, []>("op_32104_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2687_cast_fp16 = mul(x = var_32103_cast_fp16, y = var_32104_to_fp16)[name = tensor<string, []>("aw_2687_cast_fp16")];
            tensor<string, []> var_32107_equation_0 = const()[name = tensor<string, []>("op_32107_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_32107_cast_fp16 = einsum(equation = var_32107_equation_0, values = (var_31949_cast_fp16, var_31866_cast_fp16))[name = tensor<string, []>("op_32107_cast_fp16")];
            tensor<fp16, []> var_32108_to_fp16 = const()[name = tensor<string, []>("op_32108_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2689_cast_fp16 = mul(x = var_32107_cast_fp16, y = var_32108_to_fp16)[name = tensor<string, []>("aw_2689_cast_fp16")];
            tensor<string, []> var_32111_equation_0 = const()[name = tensor<string, []>("op_32111_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_32111_cast_fp16 = einsum(equation = var_32111_equation_0, values = (var_31953_cast_fp16, var_31870_cast_fp16))[name = tensor<string, []>("op_32111_cast_fp16")];
            tensor<fp16, []> var_32112_to_fp16 = const()[name = tensor<string, []>("op_32112_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2691_cast_fp16 = mul(x = var_32111_cast_fp16, y = var_32112_to_fp16)[name = tensor<string, []>("aw_2691_cast_fp16")];
            tensor<string, []> var_32115_equation_0 = const()[name = tensor<string, []>("op_32115_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_32115_cast_fp16 = einsum(equation = var_32115_equation_0, values = (var_31957_cast_fp16, var_31874_cast_fp16))[name = tensor<string, []>("op_32115_cast_fp16")];
            tensor<fp16, []> var_32116_to_fp16 = const()[name = tensor<string, []>("op_32116_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2693_cast_fp16 = mul(x = var_32115_cast_fp16, y = var_32116_to_fp16)[name = tensor<string, []>("aw_2693_cast_fp16")];
            tensor<string, []> var_32119_equation_0 = const()[name = tensor<string, []>("op_32119_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_32119_cast_fp16 = einsum(equation = var_32119_equation_0, values = (var_31961_cast_fp16, var_31878_cast_fp16))[name = tensor<string, []>("op_32119_cast_fp16")];
            tensor<fp16, []> var_32120_to_fp16 = const()[name = tensor<string, []>("op_32120_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2695_cast_fp16 = mul(x = var_32119_cast_fp16, y = var_32120_to_fp16)[name = tensor<string, []>("aw_2695_cast_fp16")];
            tensor<string, []> var_32123_equation_0 = const()[name = tensor<string, []>("op_32123_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_32123_cast_fp16 = einsum(equation = var_32123_equation_0, values = (var_31965_cast_fp16, var_31882_cast_fp16))[name = tensor<string, []>("op_32123_cast_fp16")];
            tensor<fp16, []> var_32124_to_fp16 = const()[name = tensor<string, []>("op_32124_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2697_cast_fp16 = mul(x = var_32123_cast_fp16, y = var_32124_to_fp16)[name = tensor<string, []>("aw_2697_cast_fp16")];
            tensor<string, []> var_32127_equation_0 = const()[name = tensor<string, []>("op_32127_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_32127_cast_fp16 = einsum(equation = var_32127_equation_0, values = (var_31969_cast_fp16, var_31886_cast_fp16))[name = tensor<string, []>("op_32127_cast_fp16")];
            tensor<fp16, []> var_32128_to_fp16 = const()[name = tensor<string, []>("op_32128_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2699_cast_fp16 = mul(x = var_32127_cast_fp16, y = var_32128_to_fp16)[name = tensor<string, []>("aw_2699_cast_fp16")];
            tensor<string, []> var_32131_equation_0 = const()[name = tensor<string, []>("op_32131_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_32131_cast_fp16 = einsum(equation = var_32131_equation_0, values = (var_31973_cast_fp16, var_31890_cast_fp16))[name = tensor<string, []>("op_32131_cast_fp16")];
            tensor<fp16, []> var_32132_to_fp16 = const()[name = tensor<string, []>("op_32132_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2701_cast_fp16 = mul(x = var_32131_cast_fp16, y = var_32132_to_fp16)[name = tensor<string, []>("aw_2701_cast_fp16")];
            tensor<string, []> var_32135_equation_0 = const()[name = tensor<string, []>("op_32135_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_32135_cast_fp16 = einsum(equation = var_32135_equation_0, values = (var_31977_cast_fp16, var_31894_cast_fp16))[name = tensor<string, []>("op_32135_cast_fp16")];
            tensor<fp16, []> var_32136_to_fp16 = const()[name = tensor<string, []>("op_32136_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2703_cast_fp16 = mul(x = var_32135_cast_fp16, y = var_32136_to_fp16)[name = tensor<string, []>("aw_2703_cast_fp16")];
            tensor<string, []> var_32139_equation_0 = const()[name = tensor<string, []>("op_32139_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_32139_cast_fp16 = einsum(equation = var_32139_equation_0, values = (var_31981_cast_fp16, var_31898_cast_fp16))[name = tensor<string, []>("op_32139_cast_fp16")];
            tensor<fp16, []> var_32140_to_fp16 = const()[name = tensor<string, []>("op_32140_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2705_cast_fp16 = mul(x = var_32139_cast_fp16, y = var_32140_to_fp16)[name = tensor<string, []>("aw_2705_cast_fp16")];
            tensor<string, []> var_32143_equation_0 = const()[name = tensor<string, []>("op_32143_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_32143_cast_fp16 = einsum(equation = var_32143_equation_0, values = (var_31985_cast_fp16, var_31902_cast_fp16))[name = tensor<string, []>("op_32143_cast_fp16")];
            tensor<fp16, []> var_32144_to_fp16 = const()[name = tensor<string, []>("op_32144_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2707_cast_fp16 = mul(x = var_32143_cast_fp16, y = var_32144_to_fp16)[name = tensor<string, []>("aw_2707_cast_fp16")];
            tensor<string, []> var_32147_equation_0 = const()[name = tensor<string, []>("op_32147_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_32147_cast_fp16 = einsum(equation = var_32147_equation_0, values = (var_31989_cast_fp16, var_31906_cast_fp16))[name = tensor<string, []>("op_32147_cast_fp16")];
            tensor<fp16, []> var_32148_to_fp16 = const()[name = tensor<string, []>("op_32148_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2709_cast_fp16 = mul(x = var_32147_cast_fp16, y = var_32148_to_fp16)[name = tensor<string, []>("aw_2709_cast_fp16")];
            tensor<string, []> var_32151_equation_0 = const()[name = tensor<string, []>("op_32151_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_32151_cast_fp16 = einsum(equation = var_32151_equation_0, values = (var_31993_cast_fp16, var_31910_cast_fp16))[name = tensor<string, []>("op_32151_cast_fp16")];
            tensor<fp16, []> var_32152_to_fp16 = const()[name = tensor<string, []>("op_32152_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2711_cast_fp16 = mul(x = var_32151_cast_fp16, y = var_32152_to_fp16)[name = tensor<string, []>("aw_2711_cast_fp16")];
            tensor<string, []> var_32155_equation_0 = const()[name = tensor<string, []>("op_32155_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_32155_cast_fp16 = einsum(equation = var_32155_equation_0, values = (var_31997_cast_fp16, var_31914_cast_fp16))[name = tensor<string, []>("op_32155_cast_fp16")];
            tensor<fp16, []> var_32156_to_fp16 = const()[name = tensor<string, []>("op_32156_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2713_cast_fp16 = mul(x = var_32155_cast_fp16, y = var_32156_to_fp16)[name = tensor<string, []>("aw_2713_cast_fp16")];
            tensor<string, []> var_32159_equation_0 = const()[name = tensor<string, []>("op_32159_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_32159_cast_fp16 = einsum(equation = var_32159_equation_0, values = (var_32001_cast_fp16, var_31918_cast_fp16))[name = tensor<string, []>("op_32159_cast_fp16")];
            tensor<fp16, []> var_32160_to_fp16 = const()[name = tensor<string, []>("op_32160_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2715_cast_fp16 = mul(x = var_32159_cast_fp16, y = var_32160_to_fp16)[name = tensor<string, []>("aw_2715_cast_fp16")];
            tensor<string, []> var_32163_equation_0 = const()[name = tensor<string, []>("op_32163_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_32163_cast_fp16 = einsum(equation = var_32163_equation_0, values = (var_32005_cast_fp16, var_31922_cast_fp16))[name = tensor<string, []>("op_32163_cast_fp16")];
            tensor<fp16, []> var_32164_to_fp16 = const()[name = tensor<string, []>("op_32164_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2717_cast_fp16 = mul(x = var_32163_cast_fp16, y = var_32164_to_fp16)[name = tensor<string, []>("aw_2717_cast_fp16")];
            tensor<string, []> var_32167_equation_0 = const()[name = tensor<string, []>("op_32167_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_32167_cast_fp16 = einsum(equation = var_32167_equation_0, values = (var_32009_cast_fp16, var_31926_cast_fp16))[name = tensor<string, []>("op_32167_cast_fp16")];
            tensor<fp16, []> var_32168_to_fp16 = const()[name = tensor<string, []>("op_32168_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2719_cast_fp16 = mul(x = var_32167_cast_fp16, y = var_32168_to_fp16)[name = tensor<string, []>("aw_2719_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_32170_cast_fp16 = softmax(axis = var_30355, x = aw_2681_cast_fp16)[name = tensor<string, []>("op_32170_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_32171_cast_fp16 = softmax(axis = var_30355, x = aw_2683_cast_fp16)[name = tensor<string, []>("op_32171_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_32172_cast_fp16 = softmax(axis = var_30355, x = aw_2685_cast_fp16)[name = tensor<string, []>("op_32172_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_32173_cast_fp16 = softmax(axis = var_30355, x = aw_2687_cast_fp16)[name = tensor<string, []>("op_32173_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_32174_cast_fp16 = softmax(axis = var_30355, x = aw_2689_cast_fp16)[name = tensor<string, []>("op_32174_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_32175_cast_fp16 = softmax(axis = var_30355, x = aw_2691_cast_fp16)[name = tensor<string, []>("op_32175_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_32176_cast_fp16 = softmax(axis = var_30355, x = aw_2693_cast_fp16)[name = tensor<string, []>("op_32176_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_32177_cast_fp16 = softmax(axis = var_30355, x = aw_2695_cast_fp16)[name = tensor<string, []>("op_32177_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_32178_cast_fp16 = softmax(axis = var_30355, x = aw_2697_cast_fp16)[name = tensor<string, []>("op_32178_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_32179_cast_fp16 = softmax(axis = var_30355, x = aw_2699_cast_fp16)[name = tensor<string, []>("op_32179_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_32180_cast_fp16 = softmax(axis = var_30355, x = aw_2701_cast_fp16)[name = tensor<string, []>("op_32180_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_32181_cast_fp16 = softmax(axis = var_30355, x = aw_2703_cast_fp16)[name = tensor<string, []>("op_32181_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_32182_cast_fp16 = softmax(axis = var_30355, x = aw_2705_cast_fp16)[name = tensor<string, []>("op_32182_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_32183_cast_fp16 = softmax(axis = var_30355, x = aw_2707_cast_fp16)[name = tensor<string, []>("op_32183_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_32184_cast_fp16 = softmax(axis = var_30355, x = aw_2709_cast_fp16)[name = tensor<string, []>("op_32184_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_32185_cast_fp16 = softmax(axis = var_30355, x = aw_2711_cast_fp16)[name = tensor<string, []>("op_32185_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_32186_cast_fp16 = softmax(axis = var_30355, x = aw_2713_cast_fp16)[name = tensor<string, []>("op_32186_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_32187_cast_fp16 = softmax(axis = var_30355, x = aw_2715_cast_fp16)[name = tensor<string, []>("op_32187_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_32188_cast_fp16 = softmax(axis = var_30355, x = aw_2717_cast_fp16)[name = tensor<string, []>("op_32188_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_32189_cast_fp16 = softmax(axis = var_30355, x = aw_2719_cast_fp16)[name = tensor<string, []>("op_32189_cast_fp16")];
            tensor<string, []> var_32191_equation_0 = const()[name = tensor<string, []>("op_32191_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32191_cast_fp16 = einsum(equation = var_32191_equation_0, values = (var_32011_cast_fp16, var_32170_cast_fp16))[name = tensor<string, []>("op_32191_cast_fp16")];
            tensor<string, []> var_32193_equation_0 = const()[name = tensor<string, []>("op_32193_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32193_cast_fp16 = einsum(equation = var_32193_equation_0, values = (var_32015_cast_fp16, var_32171_cast_fp16))[name = tensor<string, []>("op_32193_cast_fp16")];
            tensor<string, []> var_32195_equation_0 = const()[name = tensor<string, []>("op_32195_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32195_cast_fp16 = einsum(equation = var_32195_equation_0, values = (var_32019_cast_fp16, var_32172_cast_fp16))[name = tensor<string, []>("op_32195_cast_fp16")];
            tensor<string, []> var_32197_equation_0 = const()[name = tensor<string, []>("op_32197_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32197_cast_fp16 = einsum(equation = var_32197_equation_0, values = (var_32023_cast_fp16, var_32173_cast_fp16))[name = tensor<string, []>("op_32197_cast_fp16")];
            tensor<string, []> var_32199_equation_0 = const()[name = tensor<string, []>("op_32199_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32199_cast_fp16 = einsum(equation = var_32199_equation_0, values = (var_32027_cast_fp16, var_32174_cast_fp16))[name = tensor<string, []>("op_32199_cast_fp16")];
            tensor<string, []> var_32201_equation_0 = const()[name = tensor<string, []>("op_32201_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32201_cast_fp16 = einsum(equation = var_32201_equation_0, values = (var_32031_cast_fp16, var_32175_cast_fp16))[name = tensor<string, []>("op_32201_cast_fp16")];
            tensor<string, []> var_32203_equation_0 = const()[name = tensor<string, []>("op_32203_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32203_cast_fp16 = einsum(equation = var_32203_equation_0, values = (var_32035_cast_fp16, var_32176_cast_fp16))[name = tensor<string, []>("op_32203_cast_fp16")];
            tensor<string, []> var_32205_equation_0 = const()[name = tensor<string, []>("op_32205_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32205_cast_fp16 = einsum(equation = var_32205_equation_0, values = (var_32039_cast_fp16, var_32177_cast_fp16))[name = tensor<string, []>("op_32205_cast_fp16")];
            tensor<string, []> var_32207_equation_0 = const()[name = tensor<string, []>("op_32207_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32207_cast_fp16 = einsum(equation = var_32207_equation_0, values = (var_32043_cast_fp16, var_32178_cast_fp16))[name = tensor<string, []>("op_32207_cast_fp16")];
            tensor<string, []> var_32209_equation_0 = const()[name = tensor<string, []>("op_32209_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32209_cast_fp16 = einsum(equation = var_32209_equation_0, values = (var_32047_cast_fp16, var_32179_cast_fp16))[name = tensor<string, []>("op_32209_cast_fp16")];
            tensor<string, []> var_32211_equation_0 = const()[name = tensor<string, []>("op_32211_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32211_cast_fp16 = einsum(equation = var_32211_equation_0, values = (var_32051_cast_fp16, var_32180_cast_fp16))[name = tensor<string, []>("op_32211_cast_fp16")];
            tensor<string, []> var_32213_equation_0 = const()[name = tensor<string, []>("op_32213_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32213_cast_fp16 = einsum(equation = var_32213_equation_0, values = (var_32055_cast_fp16, var_32181_cast_fp16))[name = tensor<string, []>("op_32213_cast_fp16")];
            tensor<string, []> var_32215_equation_0 = const()[name = tensor<string, []>("op_32215_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32215_cast_fp16 = einsum(equation = var_32215_equation_0, values = (var_32059_cast_fp16, var_32182_cast_fp16))[name = tensor<string, []>("op_32215_cast_fp16")];
            tensor<string, []> var_32217_equation_0 = const()[name = tensor<string, []>("op_32217_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32217_cast_fp16 = einsum(equation = var_32217_equation_0, values = (var_32063_cast_fp16, var_32183_cast_fp16))[name = tensor<string, []>("op_32217_cast_fp16")];
            tensor<string, []> var_32219_equation_0 = const()[name = tensor<string, []>("op_32219_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32219_cast_fp16 = einsum(equation = var_32219_equation_0, values = (var_32067_cast_fp16, var_32184_cast_fp16))[name = tensor<string, []>("op_32219_cast_fp16")];
            tensor<string, []> var_32221_equation_0 = const()[name = tensor<string, []>("op_32221_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32221_cast_fp16 = einsum(equation = var_32221_equation_0, values = (var_32071_cast_fp16, var_32185_cast_fp16))[name = tensor<string, []>("op_32221_cast_fp16")];
            tensor<string, []> var_32223_equation_0 = const()[name = tensor<string, []>("op_32223_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32223_cast_fp16 = einsum(equation = var_32223_equation_0, values = (var_32075_cast_fp16, var_32186_cast_fp16))[name = tensor<string, []>("op_32223_cast_fp16")];
            tensor<string, []> var_32225_equation_0 = const()[name = tensor<string, []>("op_32225_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32225_cast_fp16 = einsum(equation = var_32225_equation_0, values = (var_32079_cast_fp16, var_32187_cast_fp16))[name = tensor<string, []>("op_32225_cast_fp16")];
            tensor<string, []> var_32227_equation_0 = const()[name = tensor<string, []>("op_32227_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32227_cast_fp16 = einsum(equation = var_32227_equation_0, values = (var_32083_cast_fp16, var_32188_cast_fp16))[name = tensor<string, []>("op_32227_cast_fp16")];
            tensor<string, []> var_32229_equation_0 = const()[name = tensor<string, []>("op_32229_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32229_cast_fp16 = einsum(equation = var_32229_equation_0, values = (var_32087_cast_fp16, var_32189_cast_fp16))[name = tensor<string, []>("op_32229_cast_fp16")];
            tensor<bool, []> input_449_interleave_0 = const()[name = tensor<string, []>("input_449_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_449_cast_fp16 = concat(axis = var_30355, interleave = input_449_interleave_0, values = (var_32191_cast_fp16, var_32193_cast_fp16, var_32195_cast_fp16, var_32197_cast_fp16, var_32199_cast_fp16, var_32201_cast_fp16, var_32203_cast_fp16, var_32205_cast_fp16, var_32207_cast_fp16, var_32209_cast_fp16, var_32211_cast_fp16, var_32213_cast_fp16, var_32215_cast_fp16, var_32217_cast_fp16, var_32219_cast_fp16, var_32221_cast_fp16, var_32223_cast_fp16, var_32225_cast_fp16, var_32227_cast_fp16, var_32229_cast_fp16))[name = tensor<string, []>("input_449_cast_fp16")];
            tensor<string, []> var_32239_pad_type_0 = const()[name = tensor<string, []>("op_32239_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_32239_strides_0 = const()[name = tensor<string, []>("op_32239_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_32239_pad_0 = const()[name = tensor<string, []>("op_32239_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_32239_dilations_0 = const()[name = tensor<string, []>("op_32239_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_32239_groups_0 = const()[name = tensor<string, []>("op_32239_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_1_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39905536))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(41134400))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_1_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_1_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_1_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(41134592)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_32239_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_1_attn2_to_out_0_bias_to_fp16, dilations = var_32239_dilations_0, groups = var_32239_groups_0, pad = var_32239_pad_0, pad_type = var_32239_pad_type_0, strides = var_32239_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_1_attn2_to_out_0_weight_to_fp16_palettized, x = input_449_cast_fp16)[name = tensor<string, []>("op_32239_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_215_cast_fp16 = add(x = var_32239_cast_fp16, y = inputs_213_cast_fp16)[name = tensor<string, []>("inputs_215_cast_fp16")];
            tensor<int32, [1]> input_451_axes_0 = const()[name = tensor<string, []>("input_451_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> input_451_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_451_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(41137216)))];
            tensor<fp16, [1280]> input_451_beta_0_to_fp16 = const()[name = tensor<string, []>("input_451_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(41139840)))];
            tensor<fp16, []> var_32249_to_fp16 = const()[name = tensor<string, []>("op_32249_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> input_451_cast_fp16 = layer_norm(axes = input_451_axes_0, beta = input_451_beta_0_to_fp16, epsilon = var_32249_to_fp16, gamma = input_451_gamma_0_to_fp16, x = inputs_215_cast_fp16)[name = tensor<string, []>("input_451_cast_fp16")];
            tensor<string, []> var_32269_pad_type_0 = const()[name = tensor<string, []>("op_32269_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_32269_strides_0 = const()[name = tensor<string, []>("op_32269_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_32269_pad_0 = const()[name = tensor<string, []>("op_32269_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_32269_dilations_0 = const()[name = tensor<string, []>("op_32269_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_32269_groups_0 = const()[name = tensor<string, []>("op_32269_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_1_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [9830400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(41142464))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(50972928))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_1_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([10240, 1280, 1, 1])];
            tensor<fp16, [10240]> up_blocks_0_attentions_0_transformer_blocks_1_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_1_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(50973120)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_32269_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_1_ff_net_0_proj_bias_to_fp16, dilations = var_32269_dilations_0, groups = var_32269_groups_0, pad = var_32269_pad_0, pad_type = var_32269_pad_type_0, strides = var_32269_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_1_ff_net_0_proj_weight_to_fp16_palettized, x = input_451_cast_fp16)[name = tensor<string, []>("op_32269_cast_fp16")];
            tensor<int32, [2]> var_32270_split_sizes_0 = const()[name = tensor<string, []>("op_32270_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_32270_axis_0 = const()[name = tensor<string, []>("op_32270_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_32270_cast_fp16_0, tensor<fp16, [2, 5120, 1, 1024]> var_32270_cast_fp16_1 = split(axis = var_32270_axis_0, split_sizes = var_32270_split_sizes_0, x = var_32269_cast_fp16)[name = tensor<string, []>("op_32270_cast_fp16")];
            tensor<string, []> var_32272_mode_0 = const()[name = tensor<string, []>("op_32272_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_32272_cast_fp16 = gelu(mode = var_32272_mode_0, x = var_32270_cast_fp16_1)[name = tensor<string, []>("op_32272_cast_fp16")];
            tensor<fp16, [2, 5120, 1, 1024]> input_453_cast_fp16 = mul(x = var_32270_cast_fp16_0, y = var_32272_cast_fp16)[name = tensor<string, []>("input_453_cast_fp16")];
            tensor<string, []> var_32280_pad_type_0 = const()[name = tensor<string, []>("op_32280_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_32280_strides_0 = const()[name = tensor<string, []>("op_32280_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_32280_pad_0 = const()[name = tensor<string, []>("op_32280_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_32280_dilations_0 = const()[name = tensor<string, []>("op_32280_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_32280_groups_0 = const()[name = tensor<string, []>("op_32280_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_1_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [4915200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(50993664))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(55908928))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_1_ff_net_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 5120, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_1_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_1_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(55909120)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_32280_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_1_ff_net_2_bias_to_fp16, dilations = var_32280_dilations_0, groups = var_32280_groups_0, pad = var_32280_pad_0, pad_type = var_32280_pad_type_0, strides = var_32280_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_1_ff_net_2_weight_to_fp16_palettized, x = input_453_cast_fp16)[name = tensor<string, []>("op_32280_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_217_cast_fp16 = add(x = var_32280_cast_fp16, y = inputs_215_cast_fp16)[name = tensor<string, []>("inputs_217_cast_fp16")];
            tensor<int32, [1]> hidden_states_299_axes_0 = const()[name = tensor<string, []>("hidden_states_299_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_299_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_299_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(55911744)))];
            tensor<fp16, [1280]> hidden_states_299_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_299_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(55914368)))];
            tensor<fp16, []> var_32296_to_fp16 = const()[name = tensor<string, []>("op_32296_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_299_cast_fp16 = layer_norm(axes = hidden_states_299_axes_0, beta = hidden_states_299_beta_0_to_fp16, epsilon = var_32296_to_fp16, gamma = hidden_states_299_gamma_0_to_fp16, x = inputs_217_cast_fp16)[name = tensor<string, []>("hidden_states_299_cast_fp16")];
            tensor<string, []> q_145_pad_type_0 = const()[name = tensor<string, []>("q_145_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_145_strides_0 = const()[name = tensor<string, []>("q_145_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_145_pad_0 = const()[name = tensor<string, []>("q_145_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_145_dilations_0 = const()[name = tensor<string, []>("q_145_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_145_groups_0 = const()[name = tensor<string, []>("q_145_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_2_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(55916992))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57145856))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_2_attn1_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_145_cast_fp16 = conv(dilations = q_145_dilations_0, groups = q_145_groups_0, pad = q_145_pad_0, pad_type = q_145_pad_type_0, strides = q_145_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_2_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_299_cast_fp16)[name = tensor<string, []>("q_145_cast_fp16")];
            tensor<string, []> k_289_pad_type_0 = const()[name = tensor<string, []>("k_289_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_289_strides_0 = const()[name = tensor<string, []>("k_289_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_289_pad_0 = const()[name = tensor<string, []>("k_289_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_289_dilations_0 = const()[name = tensor<string, []>("k_289_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_289_groups_0 = const()[name = tensor<string, []>("k_289_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_2_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57146048))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(58374912))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_2_attn1_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> k_289_cast_fp16 = conv(dilations = k_289_dilations_0, groups = k_289_groups_0, pad = k_289_pad_0, pad_type = k_289_pad_type_0, strides = k_289_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_2_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_299_cast_fp16)[name = tensor<string, []>("k_289_cast_fp16")];
            tensor<string, []> v_145_pad_type_0 = const()[name = tensor<string, []>("v_145_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_145_strides_0 = const()[name = tensor<string, []>("v_145_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_145_pad_0 = const()[name = tensor<string, []>("v_145_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_145_dilations_0 = const()[name = tensor<string, []>("v_145_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_145_groups_0 = const()[name = tensor<string, []>("v_145_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_2_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(58375104))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(59603968))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_2_attn1_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> v_145_cast_fp16 = conv(dilations = v_145_dilations_0, groups = v_145_groups_0, pad = v_145_pad_0, pad_type = v_145_pad_type_0, strides = v_145_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_2_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_299_cast_fp16)[name = tensor<string, []>("v_145_cast_fp16")];
            tensor<int32, [4]> var_32329_begin_0 = const()[name = tensor<string, []>("op_32329_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_32329_end_0 = const()[name = tensor<string, []>("op_32329_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_32329_end_mask_0 = const()[name = tensor<string, []>("op_32329_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32329_cast_fp16 = slice_by_index(begin = var_32329_begin_0, end = var_32329_end_0, end_mask = var_32329_end_mask_0, x = q_145_cast_fp16)[name = tensor<string, []>("op_32329_cast_fp16")];
            tensor<int32, [4]> var_32333_begin_0 = const()[name = tensor<string, []>("op_32333_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_32333_end_0 = const()[name = tensor<string, []>("op_32333_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_32333_end_mask_0 = const()[name = tensor<string, []>("op_32333_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32333_cast_fp16 = slice_by_index(begin = var_32333_begin_0, end = var_32333_end_0, end_mask = var_32333_end_mask_0, x = q_145_cast_fp16)[name = tensor<string, []>("op_32333_cast_fp16")];
            tensor<int32, [4]> var_32337_begin_0 = const()[name = tensor<string, []>("op_32337_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_32337_end_0 = const()[name = tensor<string, []>("op_32337_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_32337_end_mask_0 = const()[name = tensor<string, []>("op_32337_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32337_cast_fp16 = slice_by_index(begin = var_32337_begin_0, end = var_32337_end_0, end_mask = var_32337_end_mask_0, x = q_145_cast_fp16)[name = tensor<string, []>("op_32337_cast_fp16")];
            tensor<int32, [4]> var_32341_begin_0 = const()[name = tensor<string, []>("op_32341_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_32341_end_0 = const()[name = tensor<string, []>("op_32341_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_32341_end_mask_0 = const()[name = tensor<string, []>("op_32341_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32341_cast_fp16 = slice_by_index(begin = var_32341_begin_0, end = var_32341_end_0, end_mask = var_32341_end_mask_0, x = q_145_cast_fp16)[name = tensor<string, []>("op_32341_cast_fp16")];
            tensor<int32, [4]> var_32345_begin_0 = const()[name = tensor<string, []>("op_32345_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_32345_end_0 = const()[name = tensor<string, []>("op_32345_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_32345_end_mask_0 = const()[name = tensor<string, []>("op_32345_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32345_cast_fp16 = slice_by_index(begin = var_32345_begin_0, end = var_32345_end_0, end_mask = var_32345_end_mask_0, x = q_145_cast_fp16)[name = tensor<string, []>("op_32345_cast_fp16")];
            tensor<int32, [4]> var_32349_begin_0 = const()[name = tensor<string, []>("op_32349_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_32349_end_0 = const()[name = tensor<string, []>("op_32349_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_32349_end_mask_0 = const()[name = tensor<string, []>("op_32349_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32349_cast_fp16 = slice_by_index(begin = var_32349_begin_0, end = var_32349_end_0, end_mask = var_32349_end_mask_0, x = q_145_cast_fp16)[name = tensor<string, []>("op_32349_cast_fp16")];
            tensor<int32, [4]> var_32353_begin_0 = const()[name = tensor<string, []>("op_32353_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_32353_end_0 = const()[name = tensor<string, []>("op_32353_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_32353_end_mask_0 = const()[name = tensor<string, []>("op_32353_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32353_cast_fp16 = slice_by_index(begin = var_32353_begin_0, end = var_32353_end_0, end_mask = var_32353_end_mask_0, x = q_145_cast_fp16)[name = tensor<string, []>("op_32353_cast_fp16")];
            tensor<int32, [4]> var_32357_begin_0 = const()[name = tensor<string, []>("op_32357_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_32357_end_0 = const()[name = tensor<string, []>("op_32357_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_32357_end_mask_0 = const()[name = tensor<string, []>("op_32357_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32357_cast_fp16 = slice_by_index(begin = var_32357_begin_0, end = var_32357_end_0, end_mask = var_32357_end_mask_0, x = q_145_cast_fp16)[name = tensor<string, []>("op_32357_cast_fp16")];
            tensor<int32, [4]> var_32361_begin_0 = const()[name = tensor<string, []>("op_32361_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_32361_end_0 = const()[name = tensor<string, []>("op_32361_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_32361_end_mask_0 = const()[name = tensor<string, []>("op_32361_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32361_cast_fp16 = slice_by_index(begin = var_32361_begin_0, end = var_32361_end_0, end_mask = var_32361_end_mask_0, x = q_145_cast_fp16)[name = tensor<string, []>("op_32361_cast_fp16")];
            tensor<int32, [4]> var_32365_begin_0 = const()[name = tensor<string, []>("op_32365_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_32365_end_0 = const()[name = tensor<string, []>("op_32365_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_32365_end_mask_0 = const()[name = tensor<string, []>("op_32365_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32365_cast_fp16 = slice_by_index(begin = var_32365_begin_0, end = var_32365_end_0, end_mask = var_32365_end_mask_0, x = q_145_cast_fp16)[name = tensor<string, []>("op_32365_cast_fp16")];
            tensor<int32, [4]> var_32369_begin_0 = const()[name = tensor<string, []>("op_32369_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_32369_end_0 = const()[name = tensor<string, []>("op_32369_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_32369_end_mask_0 = const()[name = tensor<string, []>("op_32369_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32369_cast_fp16 = slice_by_index(begin = var_32369_begin_0, end = var_32369_end_0, end_mask = var_32369_end_mask_0, x = q_145_cast_fp16)[name = tensor<string, []>("op_32369_cast_fp16")];
            tensor<int32, [4]> var_32373_begin_0 = const()[name = tensor<string, []>("op_32373_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_32373_end_0 = const()[name = tensor<string, []>("op_32373_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_32373_end_mask_0 = const()[name = tensor<string, []>("op_32373_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32373_cast_fp16 = slice_by_index(begin = var_32373_begin_0, end = var_32373_end_0, end_mask = var_32373_end_mask_0, x = q_145_cast_fp16)[name = tensor<string, []>("op_32373_cast_fp16")];
            tensor<int32, [4]> var_32377_begin_0 = const()[name = tensor<string, []>("op_32377_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_32377_end_0 = const()[name = tensor<string, []>("op_32377_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_32377_end_mask_0 = const()[name = tensor<string, []>("op_32377_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32377_cast_fp16 = slice_by_index(begin = var_32377_begin_0, end = var_32377_end_0, end_mask = var_32377_end_mask_0, x = q_145_cast_fp16)[name = tensor<string, []>("op_32377_cast_fp16")];
            tensor<int32, [4]> var_32381_begin_0 = const()[name = tensor<string, []>("op_32381_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_32381_end_0 = const()[name = tensor<string, []>("op_32381_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_32381_end_mask_0 = const()[name = tensor<string, []>("op_32381_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32381_cast_fp16 = slice_by_index(begin = var_32381_begin_0, end = var_32381_end_0, end_mask = var_32381_end_mask_0, x = q_145_cast_fp16)[name = tensor<string, []>("op_32381_cast_fp16")];
            tensor<int32, [4]> var_32385_begin_0 = const()[name = tensor<string, []>("op_32385_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_32385_end_0 = const()[name = tensor<string, []>("op_32385_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_32385_end_mask_0 = const()[name = tensor<string, []>("op_32385_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32385_cast_fp16 = slice_by_index(begin = var_32385_begin_0, end = var_32385_end_0, end_mask = var_32385_end_mask_0, x = q_145_cast_fp16)[name = tensor<string, []>("op_32385_cast_fp16")];
            tensor<int32, [4]> var_32389_begin_0 = const()[name = tensor<string, []>("op_32389_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_32389_end_0 = const()[name = tensor<string, []>("op_32389_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_32389_end_mask_0 = const()[name = tensor<string, []>("op_32389_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32389_cast_fp16 = slice_by_index(begin = var_32389_begin_0, end = var_32389_end_0, end_mask = var_32389_end_mask_0, x = q_145_cast_fp16)[name = tensor<string, []>("op_32389_cast_fp16")];
            tensor<int32, [4]> var_32393_begin_0 = const()[name = tensor<string, []>("op_32393_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_32393_end_0 = const()[name = tensor<string, []>("op_32393_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_32393_end_mask_0 = const()[name = tensor<string, []>("op_32393_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32393_cast_fp16 = slice_by_index(begin = var_32393_begin_0, end = var_32393_end_0, end_mask = var_32393_end_mask_0, x = q_145_cast_fp16)[name = tensor<string, []>("op_32393_cast_fp16")];
            tensor<int32, [4]> var_32397_begin_0 = const()[name = tensor<string, []>("op_32397_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_32397_end_0 = const()[name = tensor<string, []>("op_32397_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_32397_end_mask_0 = const()[name = tensor<string, []>("op_32397_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32397_cast_fp16 = slice_by_index(begin = var_32397_begin_0, end = var_32397_end_0, end_mask = var_32397_end_mask_0, x = q_145_cast_fp16)[name = tensor<string, []>("op_32397_cast_fp16")];
            tensor<int32, [4]> var_32401_begin_0 = const()[name = tensor<string, []>("op_32401_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_32401_end_0 = const()[name = tensor<string, []>("op_32401_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_32401_end_mask_0 = const()[name = tensor<string, []>("op_32401_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32401_cast_fp16 = slice_by_index(begin = var_32401_begin_0, end = var_32401_end_0, end_mask = var_32401_end_mask_0, x = q_145_cast_fp16)[name = tensor<string, []>("op_32401_cast_fp16")];
            tensor<int32, [4]> var_32405_begin_0 = const()[name = tensor<string, []>("op_32405_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_32405_end_0 = const()[name = tensor<string, []>("op_32405_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_32405_end_mask_0 = const()[name = tensor<string, []>("op_32405_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32405_cast_fp16 = slice_by_index(begin = var_32405_begin_0, end = var_32405_end_0, end_mask = var_32405_end_mask_0, x = q_145_cast_fp16)[name = tensor<string, []>("op_32405_cast_fp16")];
            tensor<int32, [4]> k_291_perm_0 = const()[name = tensor<string, []>("k_291_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_32412_begin_0 = const()[name = tensor<string, []>("op_32412_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_32412_end_0 = const()[name = tensor<string, []>("op_32412_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_32412_end_mask_0 = const()[name = tensor<string, []>("op_32412_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> k_291_cast_fp16 = transpose(perm = k_291_perm_0, x = k_289_cast_fp16)[name = tensor<string, []>("transpose_67")];
            tensor<fp16, [2, 1024, 1, 64]> var_32412_cast_fp16 = slice_by_index(begin = var_32412_begin_0, end = var_32412_end_0, end_mask = var_32412_end_mask_0, x = k_291_cast_fp16)[name = tensor<string, []>("op_32412_cast_fp16")];
            tensor<int32, [4]> var_32416_begin_0 = const()[name = tensor<string, []>("op_32416_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_32416_end_0 = const()[name = tensor<string, []>("op_32416_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_32416_end_mask_0 = const()[name = tensor<string, []>("op_32416_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_32416_cast_fp16 = slice_by_index(begin = var_32416_begin_0, end = var_32416_end_0, end_mask = var_32416_end_mask_0, x = k_291_cast_fp16)[name = tensor<string, []>("op_32416_cast_fp16")];
            tensor<int32, [4]> var_32420_begin_0 = const()[name = tensor<string, []>("op_32420_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_32420_end_0 = const()[name = tensor<string, []>("op_32420_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_32420_end_mask_0 = const()[name = tensor<string, []>("op_32420_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_32420_cast_fp16 = slice_by_index(begin = var_32420_begin_0, end = var_32420_end_0, end_mask = var_32420_end_mask_0, x = k_291_cast_fp16)[name = tensor<string, []>("op_32420_cast_fp16")];
            tensor<int32, [4]> var_32424_begin_0 = const()[name = tensor<string, []>("op_32424_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_32424_end_0 = const()[name = tensor<string, []>("op_32424_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_32424_end_mask_0 = const()[name = tensor<string, []>("op_32424_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_32424_cast_fp16 = slice_by_index(begin = var_32424_begin_0, end = var_32424_end_0, end_mask = var_32424_end_mask_0, x = k_291_cast_fp16)[name = tensor<string, []>("op_32424_cast_fp16")];
            tensor<int32, [4]> var_32428_begin_0 = const()[name = tensor<string, []>("op_32428_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_32428_end_0 = const()[name = tensor<string, []>("op_32428_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_32428_end_mask_0 = const()[name = tensor<string, []>("op_32428_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_32428_cast_fp16 = slice_by_index(begin = var_32428_begin_0, end = var_32428_end_0, end_mask = var_32428_end_mask_0, x = k_291_cast_fp16)[name = tensor<string, []>("op_32428_cast_fp16")];
            tensor<int32, [4]> var_32432_begin_0 = const()[name = tensor<string, []>("op_32432_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_32432_end_0 = const()[name = tensor<string, []>("op_32432_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_32432_end_mask_0 = const()[name = tensor<string, []>("op_32432_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_32432_cast_fp16 = slice_by_index(begin = var_32432_begin_0, end = var_32432_end_0, end_mask = var_32432_end_mask_0, x = k_291_cast_fp16)[name = tensor<string, []>("op_32432_cast_fp16")];
            tensor<int32, [4]> var_32436_begin_0 = const()[name = tensor<string, []>("op_32436_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_32436_end_0 = const()[name = tensor<string, []>("op_32436_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_32436_end_mask_0 = const()[name = tensor<string, []>("op_32436_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_32436_cast_fp16 = slice_by_index(begin = var_32436_begin_0, end = var_32436_end_0, end_mask = var_32436_end_mask_0, x = k_291_cast_fp16)[name = tensor<string, []>("op_32436_cast_fp16")];
            tensor<int32, [4]> var_32440_begin_0 = const()[name = tensor<string, []>("op_32440_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_32440_end_0 = const()[name = tensor<string, []>("op_32440_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_32440_end_mask_0 = const()[name = tensor<string, []>("op_32440_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_32440_cast_fp16 = slice_by_index(begin = var_32440_begin_0, end = var_32440_end_0, end_mask = var_32440_end_mask_0, x = k_291_cast_fp16)[name = tensor<string, []>("op_32440_cast_fp16")];
            tensor<int32, [4]> var_32444_begin_0 = const()[name = tensor<string, []>("op_32444_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_32444_end_0 = const()[name = tensor<string, []>("op_32444_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_32444_end_mask_0 = const()[name = tensor<string, []>("op_32444_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_32444_cast_fp16 = slice_by_index(begin = var_32444_begin_0, end = var_32444_end_0, end_mask = var_32444_end_mask_0, x = k_291_cast_fp16)[name = tensor<string, []>("op_32444_cast_fp16")];
            tensor<int32, [4]> var_32448_begin_0 = const()[name = tensor<string, []>("op_32448_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_32448_end_0 = const()[name = tensor<string, []>("op_32448_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_32448_end_mask_0 = const()[name = tensor<string, []>("op_32448_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_32448_cast_fp16 = slice_by_index(begin = var_32448_begin_0, end = var_32448_end_0, end_mask = var_32448_end_mask_0, x = k_291_cast_fp16)[name = tensor<string, []>("op_32448_cast_fp16")];
            tensor<int32, [4]> var_32452_begin_0 = const()[name = tensor<string, []>("op_32452_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_32452_end_0 = const()[name = tensor<string, []>("op_32452_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_32452_end_mask_0 = const()[name = tensor<string, []>("op_32452_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_32452_cast_fp16 = slice_by_index(begin = var_32452_begin_0, end = var_32452_end_0, end_mask = var_32452_end_mask_0, x = k_291_cast_fp16)[name = tensor<string, []>("op_32452_cast_fp16")];
            tensor<int32, [4]> var_32456_begin_0 = const()[name = tensor<string, []>("op_32456_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_32456_end_0 = const()[name = tensor<string, []>("op_32456_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_32456_end_mask_0 = const()[name = tensor<string, []>("op_32456_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_32456_cast_fp16 = slice_by_index(begin = var_32456_begin_0, end = var_32456_end_0, end_mask = var_32456_end_mask_0, x = k_291_cast_fp16)[name = tensor<string, []>("op_32456_cast_fp16")];
            tensor<int32, [4]> var_32460_begin_0 = const()[name = tensor<string, []>("op_32460_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_32460_end_0 = const()[name = tensor<string, []>("op_32460_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_32460_end_mask_0 = const()[name = tensor<string, []>("op_32460_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_32460_cast_fp16 = slice_by_index(begin = var_32460_begin_0, end = var_32460_end_0, end_mask = var_32460_end_mask_0, x = k_291_cast_fp16)[name = tensor<string, []>("op_32460_cast_fp16")];
            tensor<int32, [4]> var_32464_begin_0 = const()[name = tensor<string, []>("op_32464_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_32464_end_0 = const()[name = tensor<string, []>("op_32464_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_32464_end_mask_0 = const()[name = tensor<string, []>("op_32464_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_32464_cast_fp16 = slice_by_index(begin = var_32464_begin_0, end = var_32464_end_0, end_mask = var_32464_end_mask_0, x = k_291_cast_fp16)[name = tensor<string, []>("op_32464_cast_fp16")];
            tensor<int32, [4]> var_32468_begin_0 = const()[name = tensor<string, []>("op_32468_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_32468_end_0 = const()[name = tensor<string, []>("op_32468_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_32468_end_mask_0 = const()[name = tensor<string, []>("op_32468_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_32468_cast_fp16 = slice_by_index(begin = var_32468_begin_0, end = var_32468_end_0, end_mask = var_32468_end_mask_0, x = k_291_cast_fp16)[name = tensor<string, []>("op_32468_cast_fp16")];
            tensor<int32, [4]> var_32472_begin_0 = const()[name = tensor<string, []>("op_32472_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_32472_end_0 = const()[name = tensor<string, []>("op_32472_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_32472_end_mask_0 = const()[name = tensor<string, []>("op_32472_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_32472_cast_fp16 = slice_by_index(begin = var_32472_begin_0, end = var_32472_end_0, end_mask = var_32472_end_mask_0, x = k_291_cast_fp16)[name = tensor<string, []>("op_32472_cast_fp16")];
            tensor<int32, [4]> var_32476_begin_0 = const()[name = tensor<string, []>("op_32476_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_32476_end_0 = const()[name = tensor<string, []>("op_32476_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_32476_end_mask_0 = const()[name = tensor<string, []>("op_32476_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_32476_cast_fp16 = slice_by_index(begin = var_32476_begin_0, end = var_32476_end_0, end_mask = var_32476_end_mask_0, x = k_291_cast_fp16)[name = tensor<string, []>("op_32476_cast_fp16")];
            tensor<int32, [4]> var_32480_begin_0 = const()[name = tensor<string, []>("op_32480_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_32480_end_0 = const()[name = tensor<string, []>("op_32480_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_32480_end_mask_0 = const()[name = tensor<string, []>("op_32480_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_32480_cast_fp16 = slice_by_index(begin = var_32480_begin_0, end = var_32480_end_0, end_mask = var_32480_end_mask_0, x = k_291_cast_fp16)[name = tensor<string, []>("op_32480_cast_fp16")];
            tensor<int32, [4]> var_32484_begin_0 = const()[name = tensor<string, []>("op_32484_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_32484_end_0 = const()[name = tensor<string, []>("op_32484_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_32484_end_mask_0 = const()[name = tensor<string, []>("op_32484_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_32484_cast_fp16 = slice_by_index(begin = var_32484_begin_0, end = var_32484_end_0, end_mask = var_32484_end_mask_0, x = k_291_cast_fp16)[name = tensor<string, []>("op_32484_cast_fp16")];
            tensor<int32, [4]> var_32488_begin_0 = const()[name = tensor<string, []>("op_32488_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_32488_end_0 = const()[name = tensor<string, []>("op_32488_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_32488_end_mask_0 = const()[name = tensor<string, []>("op_32488_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_32488_cast_fp16 = slice_by_index(begin = var_32488_begin_0, end = var_32488_end_0, end_mask = var_32488_end_mask_0, x = k_291_cast_fp16)[name = tensor<string, []>("op_32488_cast_fp16")];
            tensor<int32, [4]> var_32490_begin_0 = const()[name = tensor<string, []>("op_32490_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_32490_end_0 = const()[name = tensor<string, []>("op_32490_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_32490_end_mask_0 = const()[name = tensor<string, []>("op_32490_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32490_cast_fp16 = slice_by_index(begin = var_32490_begin_0, end = var_32490_end_0, end_mask = var_32490_end_mask_0, x = v_145_cast_fp16)[name = tensor<string, []>("op_32490_cast_fp16")];
            tensor<int32, [4]> var_32494_begin_0 = const()[name = tensor<string, []>("op_32494_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_32494_end_0 = const()[name = tensor<string, []>("op_32494_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_32494_end_mask_0 = const()[name = tensor<string, []>("op_32494_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32494_cast_fp16 = slice_by_index(begin = var_32494_begin_0, end = var_32494_end_0, end_mask = var_32494_end_mask_0, x = v_145_cast_fp16)[name = tensor<string, []>("op_32494_cast_fp16")];
            tensor<int32, [4]> var_32498_begin_0 = const()[name = tensor<string, []>("op_32498_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_32498_end_0 = const()[name = tensor<string, []>("op_32498_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_32498_end_mask_0 = const()[name = tensor<string, []>("op_32498_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32498_cast_fp16 = slice_by_index(begin = var_32498_begin_0, end = var_32498_end_0, end_mask = var_32498_end_mask_0, x = v_145_cast_fp16)[name = tensor<string, []>("op_32498_cast_fp16")];
            tensor<int32, [4]> var_32502_begin_0 = const()[name = tensor<string, []>("op_32502_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_32502_end_0 = const()[name = tensor<string, []>("op_32502_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_32502_end_mask_0 = const()[name = tensor<string, []>("op_32502_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32502_cast_fp16 = slice_by_index(begin = var_32502_begin_0, end = var_32502_end_0, end_mask = var_32502_end_mask_0, x = v_145_cast_fp16)[name = tensor<string, []>("op_32502_cast_fp16")];
            tensor<int32, [4]> var_32506_begin_0 = const()[name = tensor<string, []>("op_32506_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_32506_end_0 = const()[name = tensor<string, []>("op_32506_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_32506_end_mask_0 = const()[name = tensor<string, []>("op_32506_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32506_cast_fp16 = slice_by_index(begin = var_32506_begin_0, end = var_32506_end_0, end_mask = var_32506_end_mask_0, x = v_145_cast_fp16)[name = tensor<string, []>("op_32506_cast_fp16")];
            tensor<int32, [4]> var_32510_begin_0 = const()[name = tensor<string, []>("op_32510_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_32510_end_0 = const()[name = tensor<string, []>("op_32510_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_32510_end_mask_0 = const()[name = tensor<string, []>("op_32510_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32510_cast_fp16 = slice_by_index(begin = var_32510_begin_0, end = var_32510_end_0, end_mask = var_32510_end_mask_0, x = v_145_cast_fp16)[name = tensor<string, []>("op_32510_cast_fp16")];
            tensor<int32, [4]> var_32514_begin_0 = const()[name = tensor<string, []>("op_32514_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_32514_end_0 = const()[name = tensor<string, []>("op_32514_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_32514_end_mask_0 = const()[name = tensor<string, []>("op_32514_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32514_cast_fp16 = slice_by_index(begin = var_32514_begin_0, end = var_32514_end_0, end_mask = var_32514_end_mask_0, x = v_145_cast_fp16)[name = tensor<string, []>("op_32514_cast_fp16")];
            tensor<int32, [4]> var_32518_begin_0 = const()[name = tensor<string, []>("op_32518_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_32518_end_0 = const()[name = tensor<string, []>("op_32518_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_32518_end_mask_0 = const()[name = tensor<string, []>("op_32518_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32518_cast_fp16 = slice_by_index(begin = var_32518_begin_0, end = var_32518_end_0, end_mask = var_32518_end_mask_0, x = v_145_cast_fp16)[name = tensor<string, []>("op_32518_cast_fp16")];
            tensor<int32, [4]> var_32522_begin_0 = const()[name = tensor<string, []>("op_32522_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_32522_end_0 = const()[name = tensor<string, []>("op_32522_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_32522_end_mask_0 = const()[name = tensor<string, []>("op_32522_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32522_cast_fp16 = slice_by_index(begin = var_32522_begin_0, end = var_32522_end_0, end_mask = var_32522_end_mask_0, x = v_145_cast_fp16)[name = tensor<string, []>("op_32522_cast_fp16")];
            tensor<int32, [4]> var_32526_begin_0 = const()[name = tensor<string, []>("op_32526_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_32526_end_0 = const()[name = tensor<string, []>("op_32526_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_32526_end_mask_0 = const()[name = tensor<string, []>("op_32526_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32526_cast_fp16 = slice_by_index(begin = var_32526_begin_0, end = var_32526_end_0, end_mask = var_32526_end_mask_0, x = v_145_cast_fp16)[name = tensor<string, []>("op_32526_cast_fp16")];
            tensor<int32, [4]> var_32530_begin_0 = const()[name = tensor<string, []>("op_32530_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_32530_end_0 = const()[name = tensor<string, []>("op_32530_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_32530_end_mask_0 = const()[name = tensor<string, []>("op_32530_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32530_cast_fp16 = slice_by_index(begin = var_32530_begin_0, end = var_32530_end_0, end_mask = var_32530_end_mask_0, x = v_145_cast_fp16)[name = tensor<string, []>("op_32530_cast_fp16")];
            tensor<int32, [4]> var_32534_begin_0 = const()[name = tensor<string, []>("op_32534_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_32534_end_0 = const()[name = tensor<string, []>("op_32534_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_32534_end_mask_0 = const()[name = tensor<string, []>("op_32534_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32534_cast_fp16 = slice_by_index(begin = var_32534_begin_0, end = var_32534_end_0, end_mask = var_32534_end_mask_0, x = v_145_cast_fp16)[name = tensor<string, []>("op_32534_cast_fp16")];
            tensor<int32, [4]> var_32538_begin_0 = const()[name = tensor<string, []>("op_32538_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_32538_end_0 = const()[name = tensor<string, []>("op_32538_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_32538_end_mask_0 = const()[name = tensor<string, []>("op_32538_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32538_cast_fp16 = slice_by_index(begin = var_32538_begin_0, end = var_32538_end_0, end_mask = var_32538_end_mask_0, x = v_145_cast_fp16)[name = tensor<string, []>("op_32538_cast_fp16")];
            tensor<int32, [4]> var_32542_begin_0 = const()[name = tensor<string, []>("op_32542_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_32542_end_0 = const()[name = tensor<string, []>("op_32542_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_32542_end_mask_0 = const()[name = tensor<string, []>("op_32542_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32542_cast_fp16 = slice_by_index(begin = var_32542_begin_0, end = var_32542_end_0, end_mask = var_32542_end_mask_0, x = v_145_cast_fp16)[name = tensor<string, []>("op_32542_cast_fp16")];
            tensor<int32, [4]> var_32546_begin_0 = const()[name = tensor<string, []>("op_32546_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_32546_end_0 = const()[name = tensor<string, []>("op_32546_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_32546_end_mask_0 = const()[name = tensor<string, []>("op_32546_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32546_cast_fp16 = slice_by_index(begin = var_32546_begin_0, end = var_32546_end_0, end_mask = var_32546_end_mask_0, x = v_145_cast_fp16)[name = tensor<string, []>("op_32546_cast_fp16")];
            tensor<int32, [4]> var_32550_begin_0 = const()[name = tensor<string, []>("op_32550_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_32550_end_0 = const()[name = tensor<string, []>("op_32550_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_32550_end_mask_0 = const()[name = tensor<string, []>("op_32550_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32550_cast_fp16 = slice_by_index(begin = var_32550_begin_0, end = var_32550_end_0, end_mask = var_32550_end_mask_0, x = v_145_cast_fp16)[name = tensor<string, []>("op_32550_cast_fp16")];
            tensor<int32, [4]> var_32554_begin_0 = const()[name = tensor<string, []>("op_32554_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_32554_end_0 = const()[name = tensor<string, []>("op_32554_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_32554_end_mask_0 = const()[name = tensor<string, []>("op_32554_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32554_cast_fp16 = slice_by_index(begin = var_32554_begin_0, end = var_32554_end_0, end_mask = var_32554_end_mask_0, x = v_145_cast_fp16)[name = tensor<string, []>("op_32554_cast_fp16")];
            tensor<int32, [4]> var_32558_begin_0 = const()[name = tensor<string, []>("op_32558_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_32558_end_0 = const()[name = tensor<string, []>("op_32558_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_32558_end_mask_0 = const()[name = tensor<string, []>("op_32558_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32558_cast_fp16 = slice_by_index(begin = var_32558_begin_0, end = var_32558_end_0, end_mask = var_32558_end_mask_0, x = v_145_cast_fp16)[name = tensor<string, []>("op_32558_cast_fp16")];
            tensor<int32, [4]> var_32562_begin_0 = const()[name = tensor<string, []>("op_32562_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_32562_end_0 = const()[name = tensor<string, []>("op_32562_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_32562_end_mask_0 = const()[name = tensor<string, []>("op_32562_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32562_cast_fp16 = slice_by_index(begin = var_32562_begin_0, end = var_32562_end_0, end_mask = var_32562_end_mask_0, x = v_145_cast_fp16)[name = tensor<string, []>("op_32562_cast_fp16")];
            tensor<int32, [4]> var_32566_begin_0 = const()[name = tensor<string, []>("op_32566_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_32566_end_0 = const()[name = tensor<string, []>("op_32566_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_32566_end_mask_0 = const()[name = tensor<string, []>("op_32566_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32566_cast_fp16 = slice_by_index(begin = var_32566_begin_0, end = var_32566_end_0, end_mask = var_32566_end_mask_0, x = v_145_cast_fp16)[name = tensor<string, []>("op_32566_cast_fp16")];
            tensor<string, []> var_32570_equation_0 = const()[name = tensor<string, []>("op_32570_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32570_cast_fp16 = einsum(equation = var_32570_equation_0, values = (var_32412_cast_fp16, var_32329_cast_fp16))[name = tensor<string, []>("op_32570_cast_fp16")];
            tensor<fp16, []> var_32571_to_fp16 = const()[name = tensor<string, []>("op_32571_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2721_cast_fp16 = mul(x = var_32570_cast_fp16, y = var_32571_to_fp16)[name = tensor<string, []>("aw_2721_cast_fp16")];
            tensor<string, []> var_32574_equation_0 = const()[name = tensor<string, []>("op_32574_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32574_cast_fp16 = einsum(equation = var_32574_equation_0, values = (var_32416_cast_fp16, var_32333_cast_fp16))[name = tensor<string, []>("op_32574_cast_fp16")];
            tensor<fp16, []> var_32575_to_fp16 = const()[name = tensor<string, []>("op_32575_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2723_cast_fp16 = mul(x = var_32574_cast_fp16, y = var_32575_to_fp16)[name = tensor<string, []>("aw_2723_cast_fp16")];
            tensor<string, []> var_32578_equation_0 = const()[name = tensor<string, []>("op_32578_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32578_cast_fp16 = einsum(equation = var_32578_equation_0, values = (var_32420_cast_fp16, var_32337_cast_fp16))[name = tensor<string, []>("op_32578_cast_fp16")];
            tensor<fp16, []> var_32579_to_fp16 = const()[name = tensor<string, []>("op_32579_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2725_cast_fp16 = mul(x = var_32578_cast_fp16, y = var_32579_to_fp16)[name = tensor<string, []>("aw_2725_cast_fp16")];
            tensor<string, []> var_32582_equation_0 = const()[name = tensor<string, []>("op_32582_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32582_cast_fp16 = einsum(equation = var_32582_equation_0, values = (var_32424_cast_fp16, var_32341_cast_fp16))[name = tensor<string, []>("op_32582_cast_fp16")];
            tensor<fp16, []> var_32583_to_fp16 = const()[name = tensor<string, []>("op_32583_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2727_cast_fp16 = mul(x = var_32582_cast_fp16, y = var_32583_to_fp16)[name = tensor<string, []>("aw_2727_cast_fp16")];
            tensor<string, []> var_32586_equation_0 = const()[name = tensor<string, []>("op_32586_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32586_cast_fp16 = einsum(equation = var_32586_equation_0, values = (var_32428_cast_fp16, var_32345_cast_fp16))[name = tensor<string, []>("op_32586_cast_fp16")];
            tensor<fp16, []> var_32587_to_fp16 = const()[name = tensor<string, []>("op_32587_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2729_cast_fp16 = mul(x = var_32586_cast_fp16, y = var_32587_to_fp16)[name = tensor<string, []>("aw_2729_cast_fp16")];
            tensor<string, []> var_32590_equation_0 = const()[name = tensor<string, []>("op_32590_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32590_cast_fp16 = einsum(equation = var_32590_equation_0, values = (var_32432_cast_fp16, var_32349_cast_fp16))[name = tensor<string, []>("op_32590_cast_fp16")];
            tensor<fp16, []> var_32591_to_fp16 = const()[name = tensor<string, []>("op_32591_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2731_cast_fp16 = mul(x = var_32590_cast_fp16, y = var_32591_to_fp16)[name = tensor<string, []>("aw_2731_cast_fp16")];
            tensor<string, []> var_32594_equation_0 = const()[name = tensor<string, []>("op_32594_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32594_cast_fp16 = einsum(equation = var_32594_equation_0, values = (var_32436_cast_fp16, var_32353_cast_fp16))[name = tensor<string, []>("op_32594_cast_fp16")];
            tensor<fp16, []> var_32595_to_fp16 = const()[name = tensor<string, []>("op_32595_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2733_cast_fp16 = mul(x = var_32594_cast_fp16, y = var_32595_to_fp16)[name = tensor<string, []>("aw_2733_cast_fp16")];
            tensor<string, []> var_32598_equation_0 = const()[name = tensor<string, []>("op_32598_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32598_cast_fp16 = einsum(equation = var_32598_equation_0, values = (var_32440_cast_fp16, var_32357_cast_fp16))[name = tensor<string, []>("op_32598_cast_fp16")];
            tensor<fp16, []> var_32599_to_fp16 = const()[name = tensor<string, []>("op_32599_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2735_cast_fp16 = mul(x = var_32598_cast_fp16, y = var_32599_to_fp16)[name = tensor<string, []>("aw_2735_cast_fp16")];
            tensor<string, []> var_32602_equation_0 = const()[name = tensor<string, []>("op_32602_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32602_cast_fp16 = einsum(equation = var_32602_equation_0, values = (var_32444_cast_fp16, var_32361_cast_fp16))[name = tensor<string, []>("op_32602_cast_fp16")];
            tensor<fp16, []> var_32603_to_fp16 = const()[name = tensor<string, []>("op_32603_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2737_cast_fp16 = mul(x = var_32602_cast_fp16, y = var_32603_to_fp16)[name = tensor<string, []>("aw_2737_cast_fp16")];
            tensor<string, []> var_32606_equation_0 = const()[name = tensor<string, []>("op_32606_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32606_cast_fp16 = einsum(equation = var_32606_equation_0, values = (var_32448_cast_fp16, var_32365_cast_fp16))[name = tensor<string, []>("op_32606_cast_fp16")];
            tensor<fp16, []> var_32607_to_fp16 = const()[name = tensor<string, []>("op_32607_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2739_cast_fp16 = mul(x = var_32606_cast_fp16, y = var_32607_to_fp16)[name = tensor<string, []>("aw_2739_cast_fp16")];
            tensor<string, []> var_32610_equation_0 = const()[name = tensor<string, []>("op_32610_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32610_cast_fp16 = einsum(equation = var_32610_equation_0, values = (var_32452_cast_fp16, var_32369_cast_fp16))[name = tensor<string, []>("op_32610_cast_fp16")];
            tensor<fp16, []> var_32611_to_fp16 = const()[name = tensor<string, []>("op_32611_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2741_cast_fp16 = mul(x = var_32610_cast_fp16, y = var_32611_to_fp16)[name = tensor<string, []>("aw_2741_cast_fp16")];
            tensor<string, []> var_32614_equation_0 = const()[name = tensor<string, []>("op_32614_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32614_cast_fp16 = einsum(equation = var_32614_equation_0, values = (var_32456_cast_fp16, var_32373_cast_fp16))[name = tensor<string, []>("op_32614_cast_fp16")];
            tensor<fp16, []> var_32615_to_fp16 = const()[name = tensor<string, []>("op_32615_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2743_cast_fp16 = mul(x = var_32614_cast_fp16, y = var_32615_to_fp16)[name = tensor<string, []>("aw_2743_cast_fp16")];
            tensor<string, []> var_32618_equation_0 = const()[name = tensor<string, []>("op_32618_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32618_cast_fp16 = einsum(equation = var_32618_equation_0, values = (var_32460_cast_fp16, var_32377_cast_fp16))[name = tensor<string, []>("op_32618_cast_fp16")];
            tensor<fp16, []> var_32619_to_fp16 = const()[name = tensor<string, []>("op_32619_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2745_cast_fp16 = mul(x = var_32618_cast_fp16, y = var_32619_to_fp16)[name = tensor<string, []>("aw_2745_cast_fp16")];
            tensor<string, []> var_32622_equation_0 = const()[name = tensor<string, []>("op_32622_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32622_cast_fp16 = einsum(equation = var_32622_equation_0, values = (var_32464_cast_fp16, var_32381_cast_fp16))[name = tensor<string, []>("op_32622_cast_fp16")];
            tensor<fp16, []> var_32623_to_fp16 = const()[name = tensor<string, []>("op_32623_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2747_cast_fp16 = mul(x = var_32622_cast_fp16, y = var_32623_to_fp16)[name = tensor<string, []>("aw_2747_cast_fp16")];
            tensor<string, []> var_32626_equation_0 = const()[name = tensor<string, []>("op_32626_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32626_cast_fp16 = einsum(equation = var_32626_equation_0, values = (var_32468_cast_fp16, var_32385_cast_fp16))[name = tensor<string, []>("op_32626_cast_fp16")];
            tensor<fp16, []> var_32627_to_fp16 = const()[name = tensor<string, []>("op_32627_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2749_cast_fp16 = mul(x = var_32626_cast_fp16, y = var_32627_to_fp16)[name = tensor<string, []>("aw_2749_cast_fp16")];
            tensor<string, []> var_32630_equation_0 = const()[name = tensor<string, []>("op_32630_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32630_cast_fp16 = einsum(equation = var_32630_equation_0, values = (var_32472_cast_fp16, var_32389_cast_fp16))[name = tensor<string, []>("op_32630_cast_fp16")];
            tensor<fp16, []> var_32631_to_fp16 = const()[name = tensor<string, []>("op_32631_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2751_cast_fp16 = mul(x = var_32630_cast_fp16, y = var_32631_to_fp16)[name = tensor<string, []>("aw_2751_cast_fp16")];
            tensor<string, []> var_32634_equation_0 = const()[name = tensor<string, []>("op_32634_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32634_cast_fp16 = einsum(equation = var_32634_equation_0, values = (var_32476_cast_fp16, var_32393_cast_fp16))[name = tensor<string, []>("op_32634_cast_fp16")];
            tensor<fp16, []> var_32635_to_fp16 = const()[name = tensor<string, []>("op_32635_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2753_cast_fp16 = mul(x = var_32634_cast_fp16, y = var_32635_to_fp16)[name = tensor<string, []>("aw_2753_cast_fp16")];
            tensor<string, []> var_32638_equation_0 = const()[name = tensor<string, []>("op_32638_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32638_cast_fp16 = einsum(equation = var_32638_equation_0, values = (var_32480_cast_fp16, var_32397_cast_fp16))[name = tensor<string, []>("op_32638_cast_fp16")];
            tensor<fp16, []> var_32639_to_fp16 = const()[name = tensor<string, []>("op_32639_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2755_cast_fp16 = mul(x = var_32638_cast_fp16, y = var_32639_to_fp16)[name = tensor<string, []>("aw_2755_cast_fp16")];
            tensor<string, []> var_32642_equation_0 = const()[name = tensor<string, []>("op_32642_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32642_cast_fp16 = einsum(equation = var_32642_equation_0, values = (var_32484_cast_fp16, var_32401_cast_fp16))[name = tensor<string, []>("op_32642_cast_fp16")];
            tensor<fp16, []> var_32643_to_fp16 = const()[name = tensor<string, []>("op_32643_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2757_cast_fp16 = mul(x = var_32642_cast_fp16, y = var_32643_to_fp16)[name = tensor<string, []>("aw_2757_cast_fp16")];
            tensor<string, []> var_32646_equation_0 = const()[name = tensor<string, []>("op_32646_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32646_cast_fp16 = einsum(equation = var_32646_equation_0, values = (var_32488_cast_fp16, var_32405_cast_fp16))[name = tensor<string, []>("op_32646_cast_fp16")];
            tensor<fp16, []> var_32647_to_fp16 = const()[name = tensor<string, []>("op_32647_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2759_cast_fp16 = mul(x = var_32646_cast_fp16, y = var_32647_to_fp16)[name = tensor<string, []>("aw_2759_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32649_cast_fp16 = softmax(axis = var_30355, x = aw_2721_cast_fp16)[name = tensor<string, []>("op_32649_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32650_cast_fp16 = softmax(axis = var_30355, x = aw_2723_cast_fp16)[name = tensor<string, []>("op_32650_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32651_cast_fp16 = softmax(axis = var_30355, x = aw_2725_cast_fp16)[name = tensor<string, []>("op_32651_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32652_cast_fp16 = softmax(axis = var_30355, x = aw_2727_cast_fp16)[name = tensor<string, []>("op_32652_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32653_cast_fp16 = softmax(axis = var_30355, x = aw_2729_cast_fp16)[name = tensor<string, []>("op_32653_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32654_cast_fp16 = softmax(axis = var_30355, x = aw_2731_cast_fp16)[name = tensor<string, []>("op_32654_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32655_cast_fp16 = softmax(axis = var_30355, x = aw_2733_cast_fp16)[name = tensor<string, []>("op_32655_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32656_cast_fp16 = softmax(axis = var_30355, x = aw_2735_cast_fp16)[name = tensor<string, []>("op_32656_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32657_cast_fp16 = softmax(axis = var_30355, x = aw_2737_cast_fp16)[name = tensor<string, []>("op_32657_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32658_cast_fp16 = softmax(axis = var_30355, x = aw_2739_cast_fp16)[name = tensor<string, []>("op_32658_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32659_cast_fp16 = softmax(axis = var_30355, x = aw_2741_cast_fp16)[name = tensor<string, []>("op_32659_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32660_cast_fp16 = softmax(axis = var_30355, x = aw_2743_cast_fp16)[name = tensor<string, []>("op_32660_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32661_cast_fp16 = softmax(axis = var_30355, x = aw_2745_cast_fp16)[name = tensor<string, []>("op_32661_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32662_cast_fp16 = softmax(axis = var_30355, x = aw_2747_cast_fp16)[name = tensor<string, []>("op_32662_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32663_cast_fp16 = softmax(axis = var_30355, x = aw_2749_cast_fp16)[name = tensor<string, []>("op_32663_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32664_cast_fp16 = softmax(axis = var_30355, x = aw_2751_cast_fp16)[name = tensor<string, []>("op_32664_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32665_cast_fp16 = softmax(axis = var_30355, x = aw_2753_cast_fp16)[name = tensor<string, []>("op_32665_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32666_cast_fp16 = softmax(axis = var_30355, x = aw_2755_cast_fp16)[name = tensor<string, []>("op_32666_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32667_cast_fp16 = softmax(axis = var_30355, x = aw_2757_cast_fp16)[name = tensor<string, []>("op_32667_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32668_cast_fp16 = softmax(axis = var_30355, x = aw_2759_cast_fp16)[name = tensor<string, []>("op_32668_cast_fp16")];
            tensor<string, []> var_32670_equation_0 = const()[name = tensor<string, []>("op_32670_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32670_cast_fp16 = einsum(equation = var_32670_equation_0, values = (var_32490_cast_fp16, var_32649_cast_fp16))[name = tensor<string, []>("op_32670_cast_fp16")];
            tensor<string, []> var_32672_equation_0 = const()[name = tensor<string, []>("op_32672_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32672_cast_fp16 = einsum(equation = var_32672_equation_0, values = (var_32494_cast_fp16, var_32650_cast_fp16))[name = tensor<string, []>("op_32672_cast_fp16")];
            tensor<string, []> var_32674_equation_0 = const()[name = tensor<string, []>("op_32674_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32674_cast_fp16 = einsum(equation = var_32674_equation_0, values = (var_32498_cast_fp16, var_32651_cast_fp16))[name = tensor<string, []>("op_32674_cast_fp16")];
            tensor<string, []> var_32676_equation_0 = const()[name = tensor<string, []>("op_32676_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32676_cast_fp16 = einsum(equation = var_32676_equation_0, values = (var_32502_cast_fp16, var_32652_cast_fp16))[name = tensor<string, []>("op_32676_cast_fp16")];
            tensor<string, []> var_32678_equation_0 = const()[name = tensor<string, []>("op_32678_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32678_cast_fp16 = einsum(equation = var_32678_equation_0, values = (var_32506_cast_fp16, var_32653_cast_fp16))[name = tensor<string, []>("op_32678_cast_fp16")];
            tensor<string, []> var_32680_equation_0 = const()[name = tensor<string, []>("op_32680_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32680_cast_fp16 = einsum(equation = var_32680_equation_0, values = (var_32510_cast_fp16, var_32654_cast_fp16))[name = tensor<string, []>("op_32680_cast_fp16")];
            tensor<string, []> var_32682_equation_0 = const()[name = tensor<string, []>("op_32682_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32682_cast_fp16 = einsum(equation = var_32682_equation_0, values = (var_32514_cast_fp16, var_32655_cast_fp16))[name = tensor<string, []>("op_32682_cast_fp16")];
            tensor<string, []> var_32684_equation_0 = const()[name = tensor<string, []>("op_32684_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32684_cast_fp16 = einsum(equation = var_32684_equation_0, values = (var_32518_cast_fp16, var_32656_cast_fp16))[name = tensor<string, []>("op_32684_cast_fp16")];
            tensor<string, []> var_32686_equation_0 = const()[name = tensor<string, []>("op_32686_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32686_cast_fp16 = einsum(equation = var_32686_equation_0, values = (var_32522_cast_fp16, var_32657_cast_fp16))[name = tensor<string, []>("op_32686_cast_fp16")];
            tensor<string, []> var_32688_equation_0 = const()[name = tensor<string, []>("op_32688_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32688_cast_fp16 = einsum(equation = var_32688_equation_0, values = (var_32526_cast_fp16, var_32658_cast_fp16))[name = tensor<string, []>("op_32688_cast_fp16")];
            tensor<string, []> var_32690_equation_0 = const()[name = tensor<string, []>("op_32690_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32690_cast_fp16 = einsum(equation = var_32690_equation_0, values = (var_32530_cast_fp16, var_32659_cast_fp16))[name = tensor<string, []>("op_32690_cast_fp16")];
            tensor<string, []> var_32692_equation_0 = const()[name = tensor<string, []>("op_32692_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32692_cast_fp16 = einsum(equation = var_32692_equation_0, values = (var_32534_cast_fp16, var_32660_cast_fp16))[name = tensor<string, []>("op_32692_cast_fp16")];
            tensor<string, []> var_32694_equation_0 = const()[name = tensor<string, []>("op_32694_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32694_cast_fp16 = einsum(equation = var_32694_equation_0, values = (var_32538_cast_fp16, var_32661_cast_fp16))[name = tensor<string, []>("op_32694_cast_fp16")];
            tensor<string, []> var_32696_equation_0 = const()[name = tensor<string, []>("op_32696_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32696_cast_fp16 = einsum(equation = var_32696_equation_0, values = (var_32542_cast_fp16, var_32662_cast_fp16))[name = tensor<string, []>("op_32696_cast_fp16")];
            tensor<string, []> var_32698_equation_0 = const()[name = tensor<string, []>("op_32698_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32698_cast_fp16 = einsum(equation = var_32698_equation_0, values = (var_32546_cast_fp16, var_32663_cast_fp16))[name = tensor<string, []>("op_32698_cast_fp16")];
            tensor<string, []> var_32700_equation_0 = const()[name = tensor<string, []>("op_32700_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32700_cast_fp16 = einsum(equation = var_32700_equation_0, values = (var_32550_cast_fp16, var_32664_cast_fp16))[name = tensor<string, []>("op_32700_cast_fp16")];
            tensor<string, []> var_32702_equation_0 = const()[name = tensor<string, []>("op_32702_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32702_cast_fp16 = einsum(equation = var_32702_equation_0, values = (var_32554_cast_fp16, var_32665_cast_fp16))[name = tensor<string, []>("op_32702_cast_fp16")];
            tensor<string, []> var_32704_equation_0 = const()[name = tensor<string, []>("op_32704_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32704_cast_fp16 = einsum(equation = var_32704_equation_0, values = (var_32558_cast_fp16, var_32666_cast_fp16))[name = tensor<string, []>("op_32704_cast_fp16")];
            tensor<string, []> var_32706_equation_0 = const()[name = tensor<string, []>("op_32706_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32706_cast_fp16 = einsum(equation = var_32706_equation_0, values = (var_32562_cast_fp16, var_32667_cast_fp16))[name = tensor<string, []>("op_32706_cast_fp16")];
            tensor<string, []> var_32708_equation_0 = const()[name = tensor<string, []>("op_32708_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32708_cast_fp16 = einsum(equation = var_32708_equation_0, values = (var_32566_cast_fp16, var_32668_cast_fp16))[name = tensor<string, []>("op_32708_cast_fp16")];
            tensor<bool, []> input_455_interleave_0 = const()[name = tensor<string, []>("input_455_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_455_cast_fp16 = concat(axis = var_30355, interleave = input_455_interleave_0, values = (var_32670_cast_fp16, var_32672_cast_fp16, var_32674_cast_fp16, var_32676_cast_fp16, var_32678_cast_fp16, var_32680_cast_fp16, var_32682_cast_fp16, var_32684_cast_fp16, var_32686_cast_fp16, var_32688_cast_fp16, var_32690_cast_fp16, var_32692_cast_fp16, var_32694_cast_fp16, var_32696_cast_fp16, var_32698_cast_fp16, var_32700_cast_fp16, var_32702_cast_fp16, var_32704_cast_fp16, var_32706_cast_fp16, var_32708_cast_fp16))[name = tensor<string, []>("input_455_cast_fp16")];
            tensor<string, []> var_32718_pad_type_0 = const()[name = tensor<string, []>("op_32718_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_32718_strides_0 = const()[name = tensor<string, []>("op_32718_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_32718_pad_0 = const()[name = tensor<string, []>("op_32718_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_32718_dilations_0 = const()[name = tensor<string, []>("op_32718_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_32718_groups_0 = const()[name = tensor<string, []>("op_32718_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_2_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(59604160))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60833024))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_2_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_2_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_2_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60833216)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_32718_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_2_attn1_to_out_0_bias_to_fp16, dilations = var_32718_dilations_0, groups = var_32718_groups_0, pad = var_32718_pad_0, pad_type = var_32718_pad_type_0, strides = var_32718_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_2_attn1_to_out_0_weight_to_fp16_palettized, x = input_455_cast_fp16)[name = tensor<string, []>("op_32718_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_219_cast_fp16 = add(x = var_32718_cast_fp16, y = inputs_217_cast_fp16)[name = tensor<string, []>("inputs_219_cast_fp16")];
            tensor<int32, [1]> hidden_states_301_axes_0 = const()[name = tensor<string, []>("hidden_states_301_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_301_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_301_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60835840)))];
            tensor<fp16, [1280]> hidden_states_301_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_301_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60838464)))];
            tensor<fp16, []> var_32728_to_fp16 = const()[name = tensor<string, []>("op_32728_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_301_cast_fp16 = layer_norm(axes = hidden_states_301_axes_0, beta = hidden_states_301_beta_0_to_fp16, epsilon = var_32728_to_fp16, gamma = hidden_states_301_gamma_0_to_fp16, x = inputs_219_cast_fp16)[name = tensor<string, []>("hidden_states_301_cast_fp16")];
            tensor<string, []> q_147_pad_type_0 = const()[name = tensor<string, []>("q_147_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_147_strides_0 = const()[name = tensor<string, []>("q_147_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_147_pad_0 = const()[name = tensor<string, []>("q_147_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_147_dilations_0 = const()[name = tensor<string, []>("q_147_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_147_groups_0 = const()[name = tensor<string, []>("q_147_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_2_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60841088))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(62069952))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_2_attn2_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_147_cast_fp16 = conv(dilations = q_147_dilations_0, groups = q_147_groups_0, pad = q_147_pad_0, pad_type = q_147_pad_type_0, strides = q_147_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_2_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_301_cast_fp16)[name = tensor<string, []>("q_147_cast_fp16")];
            tensor<string, []> k_293_pad_type_0 = const()[name = tensor<string, []>("k_293_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_293_strides_0 = const()[name = tensor<string, []>("k_293_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_293_pad_0 = const()[name = tensor<string, []>("k_293_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_293_dilations_0 = const()[name = tensor<string, []>("k_293_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_293_groups_0 = const()[name = tensor<string, []>("k_293_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_2_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(62070144))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64036288))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_2_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> k_293_cast_fp16 = conv(dilations = k_293_dilations_0, groups = k_293_groups_0, pad = k_293_pad_0, pad_type = k_293_pad_type_0, strides = k_293_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_2_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("k_293_cast_fp16")];
            tensor<string, []> v_147_pad_type_0 = const()[name = tensor<string, []>("v_147_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_147_strides_0 = const()[name = tensor<string, []>("v_147_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_147_pad_0 = const()[name = tensor<string, []>("v_147_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_147_dilations_0 = const()[name = tensor<string, []>("v_147_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_147_groups_0 = const()[name = tensor<string, []>("v_147_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_2_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64036480))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66002624))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_2_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> v_147_cast_fp16 = conv(dilations = v_147_dilations_0, groups = v_147_groups_0, pad = v_147_pad_0, pad_type = v_147_pad_type_0, strides = v_147_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_2_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("v_147_cast_fp16")];
            tensor<int32, [4]> var_32761_begin_0 = const()[name = tensor<string, []>("op_32761_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_32761_end_0 = const()[name = tensor<string, []>("op_32761_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_32761_end_mask_0 = const()[name = tensor<string, []>("op_32761_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32761_cast_fp16 = slice_by_index(begin = var_32761_begin_0, end = var_32761_end_0, end_mask = var_32761_end_mask_0, x = q_147_cast_fp16)[name = tensor<string, []>("op_32761_cast_fp16")];
            tensor<int32, [4]> var_32765_begin_0 = const()[name = tensor<string, []>("op_32765_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_32765_end_0 = const()[name = tensor<string, []>("op_32765_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_32765_end_mask_0 = const()[name = tensor<string, []>("op_32765_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32765_cast_fp16 = slice_by_index(begin = var_32765_begin_0, end = var_32765_end_0, end_mask = var_32765_end_mask_0, x = q_147_cast_fp16)[name = tensor<string, []>("op_32765_cast_fp16")];
            tensor<int32, [4]> var_32769_begin_0 = const()[name = tensor<string, []>("op_32769_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_32769_end_0 = const()[name = tensor<string, []>("op_32769_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_32769_end_mask_0 = const()[name = tensor<string, []>("op_32769_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32769_cast_fp16 = slice_by_index(begin = var_32769_begin_0, end = var_32769_end_0, end_mask = var_32769_end_mask_0, x = q_147_cast_fp16)[name = tensor<string, []>("op_32769_cast_fp16")];
            tensor<int32, [4]> var_32773_begin_0 = const()[name = tensor<string, []>("op_32773_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_32773_end_0 = const()[name = tensor<string, []>("op_32773_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_32773_end_mask_0 = const()[name = tensor<string, []>("op_32773_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32773_cast_fp16 = slice_by_index(begin = var_32773_begin_0, end = var_32773_end_0, end_mask = var_32773_end_mask_0, x = q_147_cast_fp16)[name = tensor<string, []>("op_32773_cast_fp16")];
            tensor<int32, [4]> var_32777_begin_0 = const()[name = tensor<string, []>("op_32777_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_32777_end_0 = const()[name = tensor<string, []>("op_32777_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_32777_end_mask_0 = const()[name = tensor<string, []>("op_32777_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32777_cast_fp16 = slice_by_index(begin = var_32777_begin_0, end = var_32777_end_0, end_mask = var_32777_end_mask_0, x = q_147_cast_fp16)[name = tensor<string, []>("op_32777_cast_fp16")];
            tensor<int32, [4]> var_32781_begin_0 = const()[name = tensor<string, []>("op_32781_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_32781_end_0 = const()[name = tensor<string, []>("op_32781_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_32781_end_mask_0 = const()[name = tensor<string, []>("op_32781_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32781_cast_fp16 = slice_by_index(begin = var_32781_begin_0, end = var_32781_end_0, end_mask = var_32781_end_mask_0, x = q_147_cast_fp16)[name = tensor<string, []>("op_32781_cast_fp16")];
            tensor<int32, [4]> var_32785_begin_0 = const()[name = tensor<string, []>("op_32785_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_32785_end_0 = const()[name = tensor<string, []>("op_32785_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_32785_end_mask_0 = const()[name = tensor<string, []>("op_32785_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32785_cast_fp16 = slice_by_index(begin = var_32785_begin_0, end = var_32785_end_0, end_mask = var_32785_end_mask_0, x = q_147_cast_fp16)[name = tensor<string, []>("op_32785_cast_fp16")];
            tensor<int32, [4]> var_32789_begin_0 = const()[name = tensor<string, []>("op_32789_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_32789_end_0 = const()[name = tensor<string, []>("op_32789_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_32789_end_mask_0 = const()[name = tensor<string, []>("op_32789_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32789_cast_fp16 = slice_by_index(begin = var_32789_begin_0, end = var_32789_end_0, end_mask = var_32789_end_mask_0, x = q_147_cast_fp16)[name = tensor<string, []>("op_32789_cast_fp16")];
            tensor<int32, [4]> var_32793_begin_0 = const()[name = tensor<string, []>("op_32793_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_32793_end_0 = const()[name = tensor<string, []>("op_32793_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_32793_end_mask_0 = const()[name = tensor<string, []>("op_32793_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32793_cast_fp16 = slice_by_index(begin = var_32793_begin_0, end = var_32793_end_0, end_mask = var_32793_end_mask_0, x = q_147_cast_fp16)[name = tensor<string, []>("op_32793_cast_fp16")];
            tensor<int32, [4]> var_32797_begin_0 = const()[name = tensor<string, []>("op_32797_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_32797_end_0 = const()[name = tensor<string, []>("op_32797_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_32797_end_mask_0 = const()[name = tensor<string, []>("op_32797_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32797_cast_fp16 = slice_by_index(begin = var_32797_begin_0, end = var_32797_end_0, end_mask = var_32797_end_mask_0, x = q_147_cast_fp16)[name = tensor<string, []>("op_32797_cast_fp16")];
            tensor<int32, [4]> var_32801_begin_0 = const()[name = tensor<string, []>("op_32801_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_32801_end_0 = const()[name = tensor<string, []>("op_32801_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_32801_end_mask_0 = const()[name = tensor<string, []>("op_32801_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32801_cast_fp16 = slice_by_index(begin = var_32801_begin_0, end = var_32801_end_0, end_mask = var_32801_end_mask_0, x = q_147_cast_fp16)[name = tensor<string, []>("op_32801_cast_fp16")];
            tensor<int32, [4]> var_32805_begin_0 = const()[name = tensor<string, []>("op_32805_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_32805_end_0 = const()[name = tensor<string, []>("op_32805_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_32805_end_mask_0 = const()[name = tensor<string, []>("op_32805_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32805_cast_fp16 = slice_by_index(begin = var_32805_begin_0, end = var_32805_end_0, end_mask = var_32805_end_mask_0, x = q_147_cast_fp16)[name = tensor<string, []>("op_32805_cast_fp16")];
            tensor<int32, [4]> var_32809_begin_0 = const()[name = tensor<string, []>("op_32809_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_32809_end_0 = const()[name = tensor<string, []>("op_32809_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_32809_end_mask_0 = const()[name = tensor<string, []>("op_32809_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32809_cast_fp16 = slice_by_index(begin = var_32809_begin_0, end = var_32809_end_0, end_mask = var_32809_end_mask_0, x = q_147_cast_fp16)[name = tensor<string, []>("op_32809_cast_fp16")];
            tensor<int32, [4]> var_32813_begin_0 = const()[name = tensor<string, []>("op_32813_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_32813_end_0 = const()[name = tensor<string, []>("op_32813_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_32813_end_mask_0 = const()[name = tensor<string, []>("op_32813_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32813_cast_fp16 = slice_by_index(begin = var_32813_begin_0, end = var_32813_end_0, end_mask = var_32813_end_mask_0, x = q_147_cast_fp16)[name = tensor<string, []>("op_32813_cast_fp16")];
            tensor<int32, [4]> var_32817_begin_0 = const()[name = tensor<string, []>("op_32817_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_32817_end_0 = const()[name = tensor<string, []>("op_32817_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_32817_end_mask_0 = const()[name = tensor<string, []>("op_32817_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32817_cast_fp16 = slice_by_index(begin = var_32817_begin_0, end = var_32817_end_0, end_mask = var_32817_end_mask_0, x = q_147_cast_fp16)[name = tensor<string, []>("op_32817_cast_fp16")];
            tensor<int32, [4]> var_32821_begin_0 = const()[name = tensor<string, []>("op_32821_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_32821_end_0 = const()[name = tensor<string, []>("op_32821_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_32821_end_mask_0 = const()[name = tensor<string, []>("op_32821_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32821_cast_fp16 = slice_by_index(begin = var_32821_begin_0, end = var_32821_end_0, end_mask = var_32821_end_mask_0, x = q_147_cast_fp16)[name = tensor<string, []>("op_32821_cast_fp16")];
            tensor<int32, [4]> var_32825_begin_0 = const()[name = tensor<string, []>("op_32825_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_32825_end_0 = const()[name = tensor<string, []>("op_32825_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_32825_end_mask_0 = const()[name = tensor<string, []>("op_32825_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32825_cast_fp16 = slice_by_index(begin = var_32825_begin_0, end = var_32825_end_0, end_mask = var_32825_end_mask_0, x = q_147_cast_fp16)[name = tensor<string, []>("op_32825_cast_fp16")];
            tensor<int32, [4]> var_32829_begin_0 = const()[name = tensor<string, []>("op_32829_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_32829_end_0 = const()[name = tensor<string, []>("op_32829_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_32829_end_mask_0 = const()[name = tensor<string, []>("op_32829_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32829_cast_fp16 = slice_by_index(begin = var_32829_begin_0, end = var_32829_end_0, end_mask = var_32829_end_mask_0, x = q_147_cast_fp16)[name = tensor<string, []>("op_32829_cast_fp16")];
            tensor<int32, [4]> var_32833_begin_0 = const()[name = tensor<string, []>("op_32833_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_32833_end_0 = const()[name = tensor<string, []>("op_32833_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_32833_end_mask_0 = const()[name = tensor<string, []>("op_32833_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32833_cast_fp16 = slice_by_index(begin = var_32833_begin_0, end = var_32833_end_0, end_mask = var_32833_end_mask_0, x = q_147_cast_fp16)[name = tensor<string, []>("op_32833_cast_fp16")];
            tensor<int32, [4]> var_32837_begin_0 = const()[name = tensor<string, []>("op_32837_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_32837_end_0 = const()[name = tensor<string, []>("op_32837_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_32837_end_mask_0 = const()[name = tensor<string, []>("op_32837_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32837_cast_fp16 = slice_by_index(begin = var_32837_begin_0, end = var_32837_end_0, end_mask = var_32837_end_mask_0, x = q_147_cast_fp16)[name = tensor<string, []>("op_32837_cast_fp16")];
            tensor<int32, [4]> k_295_perm_0 = const()[name = tensor<string, []>("k_295_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_32844_begin_0 = const()[name = tensor<string, []>("op_32844_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_32844_end_0 = const()[name = tensor<string, []>("op_32844_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_32844_end_mask_0 = const()[name = tensor<string, []>("op_32844_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> k_295_cast_fp16 = transpose(perm = k_295_perm_0, x = k_293_cast_fp16)[name = tensor<string, []>("transpose_66")];
            tensor<fp16, [2, 77, 1, 64]> var_32844_cast_fp16 = slice_by_index(begin = var_32844_begin_0, end = var_32844_end_0, end_mask = var_32844_end_mask_0, x = k_295_cast_fp16)[name = tensor<string, []>("op_32844_cast_fp16")];
            tensor<int32, [4]> var_32848_begin_0 = const()[name = tensor<string, []>("op_32848_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_32848_end_0 = const()[name = tensor<string, []>("op_32848_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_32848_end_mask_0 = const()[name = tensor<string, []>("op_32848_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_32848_cast_fp16 = slice_by_index(begin = var_32848_begin_0, end = var_32848_end_0, end_mask = var_32848_end_mask_0, x = k_295_cast_fp16)[name = tensor<string, []>("op_32848_cast_fp16")];
            tensor<int32, [4]> var_32852_begin_0 = const()[name = tensor<string, []>("op_32852_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_32852_end_0 = const()[name = tensor<string, []>("op_32852_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_32852_end_mask_0 = const()[name = tensor<string, []>("op_32852_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_32852_cast_fp16 = slice_by_index(begin = var_32852_begin_0, end = var_32852_end_0, end_mask = var_32852_end_mask_0, x = k_295_cast_fp16)[name = tensor<string, []>("op_32852_cast_fp16")];
            tensor<int32, [4]> var_32856_begin_0 = const()[name = tensor<string, []>("op_32856_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_32856_end_0 = const()[name = tensor<string, []>("op_32856_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_32856_end_mask_0 = const()[name = tensor<string, []>("op_32856_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_32856_cast_fp16 = slice_by_index(begin = var_32856_begin_0, end = var_32856_end_0, end_mask = var_32856_end_mask_0, x = k_295_cast_fp16)[name = tensor<string, []>("op_32856_cast_fp16")];
            tensor<int32, [4]> var_32860_begin_0 = const()[name = tensor<string, []>("op_32860_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_32860_end_0 = const()[name = tensor<string, []>("op_32860_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_32860_end_mask_0 = const()[name = tensor<string, []>("op_32860_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_32860_cast_fp16 = slice_by_index(begin = var_32860_begin_0, end = var_32860_end_0, end_mask = var_32860_end_mask_0, x = k_295_cast_fp16)[name = tensor<string, []>("op_32860_cast_fp16")];
            tensor<int32, [4]> var_32864_begin_0 = const()[name = tensor<string, []>("op_32864_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_32864_end_0 = const()[name = tensor<string, []>("op_32864_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_32864_end_mask_0 = const()[name = tensor<string, []>("op_32864_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_32864_cast_fp16 = slice_by_index(begin = var_32864_begin_0, end = var_32864_end_0, end_mask = var_32864_end_mask_0, x = k_295_cast_fp16)[name = tensor<string, []>("op_32864_cast_fp16")];
            tensor<int32, [4]> var_32868_begin_0 = const()[name = tensor<string, []>("op_32868_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_32868_end_0 = const()[name = tensor<string, []>("op_32868_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_32868_end_mask_0 = const()[name = tensor<string, []>("op_32868_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_32868_cast_fp16 = slice_by_index(begin = var_32868_begin_0, end = var_32868_end_0, end_mask = var_32868_end_mask_0, x = k_295_cast_fp16)[name = tensor<string, []>("op_32868_cast_fp16")];
            tensor<int32, [4]> var_32872_begin_0 = const()[name = tensor<string, []>("op_32872_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_32872_end_0 = const()[name = tensor<string, []>("op_32872_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_32872_end_mask_0 = const()[name = tensor<string, []>("op_32872_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_32872_cast_fp16 = slice_by_index(begin = var_32872_begin_0, end = var_32872_end_0, end_mask = var_32872_end_mask_0, x = k_295_cast_fp16)[name = tensor<string, []>("op_32872_cast_fp16")];
            tensor<int32, [4]> var_32876_begin_0 = const()[name = tensor<string, []>("op_32876_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_32876_end_0 = const()[name = tensor<string, []>("op_32876_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_32876_end_mask_0 = const()[name = tensor<string, []>("op_32876_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_32876_cast_fp16 = slice_by_index(begin = var_32876_begin_0, end = var_32876_end_0, end_mask = var_32876_end_mask_0, x = k_295_cast_fp16)[name = tensor<string, []>("op_32876_cast_fp16")];
            tensor<int32, [4]> var_32880_begin_0 = const()[name = tensor<string, []>("op_32880_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_32880_end_0 = const()[name = tensor<string, []>("op_32880_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_32880_end_mask_0 = const()[name = tensor<string, []>("op_32880_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_32880_cast_fp16 = slice_by_index(begin = var_32880_begin_0, end = var_32880_end_0, end_mask = var_32880_end_mask_0, x = k_295_cast_fp16)[name = tensor<string, []>("op_32880_cast_fp16")];
            tensor<int32, [4]> var_32884_begin_0 = const()[name = tensor<string, []>("op_32884_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_32884_end_0 = const()[name = tensor<string, []>("op_32884_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_32884_end_mask_0 = const()[name = tensor<string, []>("op_32884_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_32884_cast_fp16 = slice_by_index(begin = var_32884_begin_0, end = var_32884_end_0, end_mask = var_32884_end_mask_0, x = k_295_cast_fp16)[name = tensor<string, []>("op_32884_cast_fp16")];
            tensor<int32, [4]> var_32888_begin_0 = const()[name = tensor<string, []>("op_32888_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_32888_end_0 = const()[name = tensor<string, []>("op_32888_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_32888_end_mask_0 = const()[name = tensor<string, []>("op_32888_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_32888_cast_fp16 = slice_by_index(begin = var_32888_begin_0, end = var_32888_end_0, end_mask = var_32888_end_mask_0, x = k_295_cast_fp16)[name = tensor<string, []>("op_32888_cast_fp16")];
            tensor<int32, [4]> var_32892_begin_0 = const()[name = tensor<string, []>("op_32892_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_32892_end_0 = const()[name = tensor<string, []>("op_32892_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_32892_end_mask_0 = const()[name = tensor<string, []>("op_32892_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_32892_cast_fp16 = slice_by_index(begin = var_32892_begin_0, end = var_32892_end_0, end_mask = var_32892_end_mask_0, x = k_295_cast_fp16)[name = tensor<string, []>("op_32892_cast_fp16")];
            tensor<int32, [4]> var_32896_begin_0 = const()[name = tensor<string, []>("op_32896_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_32896_end_0 = const()[name = tensor<string, []>("op_32896_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_32896_end_mask_0 = const()[name = tensor<string, []>("op_32896_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_32896_cast_fp16 = slice_by_index(begin = var_32896_begin_0, end = var_32896_end_0, end_mask = var_32896_end_mask_0, x = k_295_cast_fp16)[name = tensor<string, []>("op_32896_cast_fp16")];
            tensor<int32, [4]> var_32900_begin_0 = const()[name = tensor<string, []>("op_32900_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_32900_end_0 = const()[name = tensor<string, []>("op_32900_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_32900_end_mask_0 = const()[name = tensor<string, []>("op_32900_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_32900_cast_fp16 = slice_by_index(begin = var_32900_begin_0, end = var_32900_end_0, end_mask = var_32900_end_mask_0, x = k_295_cast_fp16)[name = tensor<string, []>("op_32900_cast_fp16")];
            tensor<int32, [4]> var_32904_begin_0 = const()[name = tensor<string, []>("op_32904_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_32904_end_0 = const()[name = tensor<string, []>("op_32904_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_32904_end_mask_0 = const()[name = tensor<string, []>("op_32904_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_32904_cast_fp16 = slice_by_index(begin = var_32904_begin_0, end = var_32904_end_0, end_mask = var_32904_end_mask_0, x = k_295_cast_fp16)[name = tensor<string, []>("op_32904_cast_fp16")];
            tensor<int32, [4]> var_32908_begin_0 = const()[name = tensor<string, []>("op_32908_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_32908_end_0 = const()[name = tensor<string, []>("op_32908_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_32908_end_mask_0 = const()[name = tensor<string, []>("op_32908_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_32908_cast_fp16 = slice_by_index(begin = var_32908_begin_0, end = var_32908_end_0, end_mask = var_32908_end_mask_0, x = k_295_cast_fp16)[name = tensor<string, []>("op_32908_cast_fp16")];
            tensor<int32, [4]> var_32912_begin_0 = const()[name = tensor<string, []>("op_32912_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_32912_end_0 = const()[name = tensor<string, []>("op_32912_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_32912_end_mask_0 = const()[name = tensor<string, []>("op_32912_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_32912_cast_fp16 = slice_by_index(begin = var_32912_begin_0, end = var_32912_end_0, end_mask = var_32912_end_mask_0, x = k_295_cast_fp16)[name = tensor<string, []>("op_32912_cast_fp16")];
            tensor<int32, [4]> var_32916_begin_0 = const()[name = tensor<string, []>("op_32916_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_32916_end_0 = const()[name = tensor<string, []>("op_32916_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_32916_end_mask_0 = const()[name = tensor<string, []>("op_32916_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_32916_cast_fp16 = slice_by_index(begin = var_32916_begin_0, end = var_32916_end_0, end_mask = var_32916_end_mask_0, x = k_295_cast_fp16)[name = tensor<string, []>("op_32916_cast_fp16")];
            tensor<int32, [4]> var_32920_begin_0 = const()[name = tensor<string, []>("op_32920_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_32920_end_0 = const()[name = tensor<string, []>("op_32920_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_32920_end_mask_0 = const()[name = tensor<string, []>("op_32920_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_32920_cast_fp16 = slice_by_index(begin = var_32920_begin_0, end = var_32920_end_0, end_mask = var_32920_end_mask_0, x = k_295_cast_fp16)[name = tensor<string, []>("op_32920_cast_fp16")];
            tensor<int32, [4]> var_32922_begin_0 = const()[name = tensor<string, []>("op_32922_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_32922_end_0 = const()[name = tensor<string, []>("op_32922_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_32922_end_mask_0 = const()[name = tensor<string, []>("op_32922_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32922_cast_fp16 = slice_by_index(begin = var_32922_begin_0, end = var_32922_end_0, end_mask = var_32922_end_mask_0, x = v_147_cast_fp16)[name = tensor<string, []>("op_32922_cast_fp16")];
            tensor<int32, [4]> var_32926_begin_0 = const()[name = tensor<string, []>("op_32926_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_32926_end_0 = const()[name = tensor<string, []>("op_32926_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_32926_end_mask_0 = const()[name = tensor<string, []>("op_32926_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32926_cast_fp16 = slice_by_index(begin = var_32926_begin_0, end = var_32926_end_0, end_mask = var_32926_end_mask_0, x = v_147_cast_fp16)[name = tensor<string, []>("op_32926_cast_fp16")];
            tensor<int32, [4]> var_32930_begin_0 = const()[name = tensor<string, []>("op_32930_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_32930_end_0 = const()[name = tensor<string, []>("op_32930_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_32930_end_mask_0 = const()[name = tensor<string, []>("op_32930_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32930_cast_fp16 = slice_by_index(begin = var_32930_begin_0, end = var_32930_end_0, end_mask = var_32930_end_mask_0, x = v_147_cast_fp16)[name = tensor<string, []>("op_32930_cast_fp16")];
            tensor<int32, [4]> var_32934_begin_0 = const()[name = tensor<string, []>("op_32934_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_32934_end_0 = const()[name = tensor<string, []>("op_32934_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_32934_end_mask_0 = const()[name = tensor<string, []>("op_32934_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32934_cast_fp16 = slice_by_index(begin = var_32934_begin_0, end = var_32934_end_0, end_mask = var_32934_end_mask_0, x = v_147_cast_fp16)[name = tensor<string, []>("op_32934_cast_fp16")];
            tensor<int32, [4]> var_32938_begin_0 = const()[name = tensor<string, []>("op_32938_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_32938_end_0 = const()[name = tensor<string, []>("op_32938_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_32938_end_mask_0 = const()[name = tensor<string, []>("op_32938_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32938_cast_fp16 = slice_by_index(begin = var_32938_begin_0, end = var_32938_end_0, end_mask = var_32938_end_mask_0, x = v_147_cast_fp16)[name = tensor<string, []>("op_32938_cast_fp16")];
            tensor<int32, [4]> var_32942_begin_0 = const()[name = tensor<string, []>("op_32942_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_32942_end_0 = const()[name = tensor<string, []>("op_32942_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_32942_end_mask_0 = const()[name = tensor<string, []>("op_32942_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32942_cast_fp16 = slice_by_index(begin = var_32942_begin_0, end = var_32942_end_0, end_mask = var_32942_end_mask_0, x = v_147_cast_fp16)[name = tensor<string, []>("op_32942_cast_fp16")];
            tensor<int32, [4]> var_32946_begin_0 = const()[name = tensor<string, []>("op_32946_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_32946_end_0 = const()[name = tensor<string, []>("op_32946_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_32946_end_mask_0 = const()[name = tensor<string, []>("op_32946_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32946_cast_fp16 = slice_by_index(begin = var_32946_begin_0, end = var_32946_end_0, end_mask = var_32946_end_mask_0, x = v_147_cast_fp16)[name = tensor<string, []>("op_32946_cast_fp16")];
            tensor<int32, [4]> var_32950_begin_0 = const()[name = tensor<string, []>("op_32950_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_32950_end_0 = const()[name = tensor<string, []>("op_32950_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_32950_end_mask_0 = const()[name = tensor<string, []>("op_32950_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32950_cast_fp16 = slice_by_index(begin = var_32950_begin_0, end = var_32950_end_0, end_mask = var_32950_end_mask_0, x = v_147_cast_fp16)[name = tensor<string, []>("op_32950_cast_fp16")];
            tensor<int32, [4]> var_32954_begin_0 = const()[name = tensor<string, []>("op_32954_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_32954_end_0 = const()[name = tensor<string, []>("op_32954_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_32954_end_mask_0 = const()[name = tensor<string, []>("op_32954_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32954_cast_fp16 = slice_by_index(begin = var_32954_begin_0, end = var_32954_end_0, end_mask = var_32954_end_mask_0, x = v_147_cast_fp16)[name = tensor<string, []>("op_32954_cast_fp16")];
            tensor<int32, [4]> var_32958_begin_0 = const()[name = tensor<string, []>("op_32958_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_32958_end_0 = const()[name = tensor<string, []>("op_32958_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_32958_end_mask_0 = const()[name = tensor<string, []>("op_32958_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32958_cast_fp16 = slice_by_index(begin = var_32958_begin_0, end = var_32958_end_0, end_mask = var_32958_end_mask_0, x = v_147_cast_fp16)[name = tensor<string, []>("op_32958_cast_fp16")];
            tensor<int32, [4]> var_32962_begin_0 = const()[name = tensor<string, []>("op_32962_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_32962_end_0 = const()[name = tensor<string, []>("op_32962_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_32962_end_mask_0 = const()[name = tensor<string, []>("op_32962_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32962_cast_fp16 = slice_by_index(begin = var_32962_begin_0, end = var_32962_end_0, end_mask = var_32962_end_mask_0, x = v_147_cast_fp16)[name = tensor<string, []>("op_32962_cast_fp16")];
            tensor<int32, [4]> var_32966_begin_0 = const()[name = tensor<string, []>("op_32966_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_32966_end_0 = const()[name = tensor<string, []>("op_32966_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_32966_end_mask_0 = const()[name = tensor<string, []>("op_32966_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32966_cast_fp16 = slice_by_index(begin = var_32966_begin_0, end = var_32966_end_0, end_mask = var_32966_end_mask_0, x = v_147_cast_fp16)[name = tensor<string, []>("op_32966_cast_fp16")];
            tensor<int32, [4]> var_32970_begin_0 = const()[name = tensor<string, []>("op_32970_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_32970_end_0 = const()[name = tensor<string, []>("op_32970_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_32970_end_mask_0 = const()[name = tensor<string, []>("op_32970_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32970_cast_fp16 = slice_by_index(begin = var_32970_begin_0, end = var_32970_end_0, end_mask = var_32970_end_mask_0, x = v_147_cast_fp16)[name = tensor<string, []>("op_32970_cast_fp16")];
            tensor<int32, [4]> var_32974_begin_0 = const()[name = tensor<string, []>("op_32974_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_32974_end_0 = const()[name = tensor<string, []>("op_32974_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_32974_end_mask_0 = const()[name = tensor<string, []>("op_32974_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32974_cast_fp16 = slice_by_index(begin = var_32974_begin_0, end = var_32974_end_0, end_mask = var_32974_end_mask_0, x = v_147_cast_fp16)[name = tensor<string, []>("op_32974_cast_fp16")];
            tensor<int32, [4]> var_32978_begin_0 = const()[name = tensor<string, []>("op_32978_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_32978_end_0 = const()[name = tensor<string, []>("op_32978_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_32978_end_mask_0 = const()[name = tensor<string, []>("op_32978_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32978_cast_fp16 = slice_by_index(begin = var_32978_begin_0, end = var_32978_end_0, end_mask = var_32978_end_mask_0, x = v_147_cast_fp16)[name = tensor<string, []>("op_32978_cast_fp16")];
            tensor<int32, [4]> var_32982_begin_0 = const()[name = tensor<string, []>("op_32982_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_32982_end_0 = const()[name = tensor<string, []>("op_32982_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_32982_end_mask_0 = const()[name = tensor<string, []>("op_32982_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32982_cast_fp16 = slice_by_index(begin = var_32982_begin_0, end = var_32982_end_0, end_mask = var_32982_end_mask_0, x = v_147_cast_fp16)[name = tensor<string, []>("op_32982_cast_fp16")];
            tensor<int32, [4]> var_32986_begin_0 = const()[name = tensor<string, []>("op_32986_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_32986_end_0 = const()[name = tensor<string, []>("op_32986_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_32986_end_mask_0 = const()[name = tensor<string, []>("op_32986_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32986_cast_fp16 = slice_by_index(begin = var_32986_begin_0, end = var_32986_end_0, end_mask = var_32986_end_mask_0, x = v_147_cast_fp16)[name = tensor<string, []>("op_32986_cast_fp16")];
            tensor<int32, [4]> var_32990_begin_0 = const()[name = tensor<string, []>("op_32990_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_32990_end_0 = const()[name = tensor<string, []>("op_32990_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_32990_end_mask_0 = const()[name = tensor<string, []>("op_32990_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32990_cast_fp16 = slice_by_index(begin = var_32990_begin_0, end = var_32990_end_0, end_mask = var_32990_end_mask_0, x = v_147_cast_fp16)[name = tensor<string, []>("op_32990_cast_fp16")];
            tensor<int32, [4]> var_32994_begin_0 = const()[name = tensor<string, []>("op_32994_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_32994_end_0 = const()[name = tensor<string, []>("op_32994_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_32994_end_mask_0 = const()[name = tensor<string, []>("op_32994_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32994_cast_fp16 = slice_by_index(begin = var_32994_begin_0, end = var_32994_end_0, end_mask = var_32994_end_mask_0, x = v_147_cast_fp16)[name = tensor<string, []>("op_32994_cast_fp16")];
            tensor<int32, [4]> var_32998_begin_0 = const()[name = tensor<string, []>("op_32998_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_32998_end_0 = const()[name = tensor<string, []>("op_32998_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_32998_end_mask_0 = const()[name = tensor<string, []>("op_32998_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32998_cast_fp16 = slice_by_index(begin = var_32998_begin_0, end = var_32998_end_0, end_mask = var_32998_end_mask_0, x = v_147_cast_fp16)[name = tensor<string, []>("op_32998_cast_fp16")];
            tensor<string, []> var_33002_equation_0 = const()[name = tensor<string, []>("op_33002_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33002_cast_fp16 = einsum(equation = var_33002_equation_0, values = (var_32844_cast_fp16, var_32761_cast_fp16))[name = tensor<string, []>("op_33002_cast_fp16")];
            tensor<fp16, []> var_33003_to_fp16 = const()[name = tensor<string, []>("op_33003_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2761_cast_fp16 = mul(x = var_33002_cast_fp16, y = var_33003_to_fp16)[name = tensor<string, []>("aw_2761_cast_fp16")];
            tensor<string, []> var_33006_equation_0 = const()[name = tensor<string, []>("op_33006_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33006_cast_fp16 = einsum(equation = var_33006_equation_0, values = (var_32848_cast_fp16, var_32765_cast_fp16))[name = tensor<string, []>("op_33006_cast_fp16")];
            tensor<fp16, []> var_33007_to_fp16 = const()[name = tensor<string, []>("op_33007_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2763_cast_fp16 = mul(x = var_33006_cast_fp16, y = var_33007_to_fp16)[name = tensor<string, []>("aw_2763_cast_fp16")];
            tensor<string, []> var_33010_equation_0 = const()[name = tensor<string, []>("op_33010_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33010_cast_fp16 = einsum(equation = var_33010_equation_0, values = (var_32852_cast_fp16, var_32769_cast_fp16))[name = tensor<string, []>("op_33010_cast_fp16")];
            tensor<fp16, []> var_33011_to_fp16 = const()[name = tensor<string, []>("op_33011_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2765_cast_fp16 = mul(x = var_33010_cast_fp16, y = var_33011_to_fp16)[name = tensor<string, []>("aw_2765_cast_fp16")];
            tensor<string, []> var_33014_equation_0 = const()[name = tensor<string, []>("op_33014_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33014_cast_fp16 = einsum(equation = var_33014_equation_0, values = (var_32856_cast_fp16, var_32773_cast_fp16))[name = tensor<string, []>("op_33014_cast_fp16")];
            tensor<fp16, []> var_33015_to_fp16 = const()[name = tensor<string, []>("op_33015_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2767_cast_fp16 = mul(x = var_33014_cast_fp16, y = var_33015_to_fp16)[name = tensor<string, []>("aw_2767_cast_fp16")];
            tensor<string, []> var_33018_equation_0 = const()[name = tensor<string, []>("op_33018_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33018_cast_fp16 = einsum(equation = var_33018_equation_0, values = (var_32860_cast_fp16, var_32777_cast_fp16))[name = tensor<string, []>("op_33018_cast_fp16")];
            tensor<fp16, []> var_33019_to_fp16 = const()[name = tensor<string, []>("op_33019_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2769_cast_fp16 = mul(x = var_33018_cast_fp16, y = var_33019_to_fp16)[name = tensor<string, []>("aw_2769_cast_fp16")];
            tensor<string, []> var_33022_equation_0 = const()[name = tensor<string, []>("op_33022_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33022_cast_fp16 = einsum(equation = var_33022_equation_0, values = (var_32864_cast_fp16, var_32781_cast_fp16))[name = tensor<string, []>("op_33022_cast_fp16")];
            tensor<fp16, []> var_33023_to_fp16 = const()[name = tensor<string, []>("op_33023_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2771_cast_fp16 = mul(x = var_33022_cast_fp16, y = var_33023_to_fp16)[name = tensor<string, []>("aw_2771_cast_fp16")];
            tensor<string, []> var_33026_equation_0 = const()[name = tensor<string, []>("op_33026_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33026_cast_fp16 = einsum(equation = var_33026_equation_0, values = (var_32868_cast_fp16, var_32785_cast_fp16))[name = tensor<string, []>("op_33026_cast_fp16")];
            tensor<fp16, []> var_33027_to_fp16 = const()[name = tensor<string, []>("op_33027_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2773_cast_fp16 = mul(x = var_33026_cast_fp16, y = var_33027_to_fp16)[name = tensor<string, []>("aw_2773_cast_fp16")];
            tensor<string, []> var_33030_equation_0 = const()[name = tensor<string, []>("op_33030_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33030_cast_fp16 = einsum(equation = var_33030_equation_0, values = (var_32872_cast_fp16, var_32789_cast_fp16))[name = tensor<string, []>("op_33030_cast_fp16")];
            tensor<fp16, []> var_33031_to_fp16 = const()[name = tensor<string, []>("op_33031_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2775_cast_fp16 = mul(x = var_33030_cast_fp16, y = var_33031_to_fp16)[name = tensor<string, []>("aw_2775_cast_fp16")];
            tensor<string, []> var_33034_equation_0 = const()[name = tensor<string, []>("op_33034_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33034_cast_fp16 = einsum(equation = var_33034_equation_0, values = (var_32876_cast_fp16, var_32793_cast_fp16))[name = tensor<string, []>("op_33034_cast_fp16")];
            tensor<fp16, []> var_33035_to_fp16 = const()[name = tensor<string, []>("op_33035_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2777_cast_fp16 = mul(x = var_33034_cast_fp16, y = var_33035_to_fp16)[name = tensor<string, []>("aw_2777_cast_fp16")];
            tensor<string, []> var_33038_equation_0 = const()[name = tensor<string, []>("op_33038_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33038_cast_fp16 = einsum(equation = var_33038_equation_0, values = (var_32880_cast_fp16, var_32797_cast_fp16))[name = tensor<string, []>("op_33038_cast_fp16")];
            tensor<fp16, []> var_33039_to_fp16 = const()[name = tensor<string, []>("op_33039_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2779_cast_fp16 = mul(x = var_33038_cast_fp16, y = var_33039_to_fp16)[name = tensor<string, []>("aw_2779_cast_fp16")];
            tensor<string, []> var_33042_equation_0 = const()[name = tensor<string, []>("op_33042_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33042_cast_fp16 = einsum(equation = var_33042_equation_0, values = (var_32884_cast_fp16, var_32801_cast_fp16))[name = tensor<string, []>("op_33042_cast_fp16")];
            tensor<fp16, []> var_33043_to_fp16 = const()[name = tensor<string, []>("op_33043_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2781_cast_fp16 = mul(x = var_33042_cast_fp16, y = var_33043_to_fp16)[name = tensor<string, []>("aw_2781_cast_fp16")];
            tensor<string, []> var_33046_equation_0 = const()[name = tensor<string, []>("op_33046_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33046_cast_fp16 = einsum(equation = var_33046_equation_0, values = (var_32888_cast_fp16, var_32805_cast_fp16))[name = tensor<string, []>("op_33046_cast_fp16")];
            tensor<fp16, []> var_33047_to_fp16 = const()[name = tensor<string, []>("op_33047_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2783_cast_fp16 = mul(x = var_33046_cast_fp16, y = var_33047_to_fp16)[name = tensor<string, []>("aw_2783_cast_fp16")];
            tensor<string, []> var_33050_equation_0 = const()[name = tensor<string, []>("op_33050_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33050_cast_fp16 = einsum(equation = var_33050_equation_0, values = (var_32892_cast_fp16, var_32809_cast_fp16))[name = tensor<string, []>("op_33050_cast_fp16")];
            tensor<fp16, []> var_33051_to_fp16 = const()[name = tensor<string, []>("op_33051_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2785_cast_fp16 = mul(x = var_33050_cast_fp16, y = var_33051_to_fp16)[name = tensor<string, []>("aw_2785_cast_fp16")];
            tensor<string, []> var_33054_equation_0 = const()[name = tensor<string, []>("op_33054_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33054_cast_fp16 = einsum(equation = var_33054_equation_0, values = (var_32896_cast_fp16, var_32813_cast_fp16))[name = tensor<string, []>("op_33054_cast_fp16")];
            tensor<fp16, []> var_33055_to_fp16 = const()[name = tensor<string, []>("op_33055_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2787_cast_fp16 = mul(x = var_33054_cast_fp16, y = var_33055_to_fp16)[name = tensor<string, []>("aw_2787_cast_fp16")];
            tensor<string, []> var_33058_equation_0 = const()[name = tensor<string, []>("op_33058_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33058_cast_fp16 = einsum(equation = var_33058_equation_0, values = (var_32900_cast_fp16, var_32817_cast_fp16))[name = tensor<string, []>("op_33058_cast_fp16")];
            tensor<fp16, []> var_33059_to_fp16 = const()[name = tensor<string, []>("op_33059_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2789_cast_fp16 = mul(x = var_33058_cast_fp16, y = var_33059_to_fp16)[name = tensor<string, []>("aw_2789_cast_fp16")];
            tensor<string, []> var_33062_equation_0 = const()[name = tensor<string, []>("op_33062_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33062_cast_fp16 = einsum(equation = var_33062_equation_0, values = (var_32904_cast_fp16, var_32821_cast_fp16))[name = tensor<string, []>("op_33062_cast_fp16")];
            tensor<fp16, []> var_33063_to_fp16 = const()[name = tensor<string, []>("op_33063_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2791_cast_fp16 = mul(x = var_33062_cast_fp16, y = var_33063_to_fp16)[name = tensor<string, []>("aw_2791_cast_fp16")];
            tensor<string, []> var_33066_equation_0 = const()[name = tensor<string, []>("op_33066_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33066_cast_fp16 = einsum(equation = var_33066_equation_0, values = (var_32908_cast_fp16, var_32825_cast_fp16))[name = tensor<string, []>("op_33066_cast_fp16")];
            tensor<fp16, []> var_33067_to_fp16 = const()[name = tensor<string, []>("op_33067_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2793_cast_fp16 = mul(x = var_33066_cast_fp16, y = var_33067_to_fp16)[name = tensor<string, []>("aw_2793_cast_fp16")];
            tensor<string, []> var_33070_equation_0 = const()[name = tensor<string, []>("op_33070_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33070_cast_fp16 = einsum(equation = var_33070_equation_0, values = (var_32912_cast_fp16, var_32829_cast_fp16))[name = tensor<string, []>("op_33070_cast_fp16")];
            tensor<fp16, []> var_33071_to_fp16 = const()[name = tensor<string, []>("op_33071_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2795_cast_fp16 = mul(x = var_33070_cast_fp16, y = var_33071_to_fp16)[name = tensor<string, []>("aw_2795_cast_fp16")];
            tensor<string, []> var_33074_equation_0 = const()[name = tensor<string, []>("op_33074_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33074_cast_fp16 = einsum(equation = var_33074_equation_0, values = (var_32916_cast_fp16, var_32833_cast_fp16))[name = tensor<string, []>("op_33074_cast_fp16")];
            tensor<fp16, []> var_33075_to_fp16 = const()[name = tensor<string, []>("op_33075_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2797_cast_fp16 = mul(x = var_33074_cast_fp16, y = var_33075_to_fp16)[name = tensor<string, []>("aw_2797_cast_fp16")];
            tensor<string, []> var_33078_equation_0 = const()[name = tensor<string, []>("op_33078_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33078_cast_fp16 = einsum(equation = var_33078_equation_0, values = (var_32920_cast_fp16, var_32837_cast_fp16))[name = tensor<string, []>("op_33078_cast_fp16")];
            tensor<fp16, []> var_33079_to_fp16 = const()[name = tensor<string, []>("op_33079_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2799_cast_fp16 = mul(x = var_33078_cast_fp16, y = var_33079_to_fp16)[name = tensor<string, []>("aw_2799_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_33081_cast_fp16 = softmax(axis = var_30355, x = aw_2761_cast_fp16)[name = tensor<string, []>("op_33081_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_33082_cast_fp16 = softmax(axis = var_30355, x = aw_2763_cast_fp16)[name = tensor<string, []>("op_33082_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_33083_cast_fp16 = softmax(axis = var_30355, x = aw_2765_cast_fp16)[name = tensor<string, []>("op_33083_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_33084_cast_fp16 = softmax(axis = var_30355, x = aw_2767_cast_fp16)[name = tensor<string, []>("op_33084_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_33085_cast_fp16 = softmax(axis = var_30355, x = aw_2769_cast_fp16)[name = tensor<string, []>("op_33085_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_33086_cast_fp16 = softmax(axis = var_30355, x = aw_2771_cast_fp16)[name = tensor<string, []>("op_33086_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_33087_cast_fp16 = softmax(axis = var_30355, x = aw_2773_cast_fp16)[name = tensor<string, []>("op_33087_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_33088_cast_fp16 = softmax(axis = var_30355, x = aw_2775_cast_fp16)[name = tensor<string, []>("op_33088_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_33089_cast_fp16 = softmax(axis = var_30355, x = aw_2777_cast_fp16)[name = tensor<string, []>("op_33089_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_33090_cast_fp16 = softmax(axis = var_30355, x = aw_2779_cast_fp16)[name = tensor<string, []>("op_33090_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_33091_cast_fp16 = softmax(axis = var_30355, x = aw_2781_cast_fp16)[name = tensor<string, []>("op_33091_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_33092_cast_fp16 = softmax(axis = var_30355, x = aw_2783_cast_fp16)[name = tensor<string, []>("op_33092_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_33093_cast_fp16 = softmax(axis = var_30355, x = aw_2785_cast_fp16)[name = tensor<string, []>("op_33093_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_33094_cast_fp16 = softmax(axis = var_30355, x = aw_2787_cast_fp16)[name = tensor<string, []>("op_33094_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_33095_cast_fp16 = softmax(axis = var_30355, x = aw_2789_cast_fp16)[name = tensor<string, []>("op_33095_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_33096_cast_fp16 = softmax(axis = var_30355, x = aw_2791_cast_fp16)[name = tensor<string, []>("op_33096_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_33097_cast_fp16 = softmax(axis = var_30355, x = aw_2793_cast_fp16)[name = tensor<string, []>("op_33097_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_33098_cast_fp16 = softmax(axis = var_30355, x = aw_2795_cast_fp16)[name = tensor<string, []>("op_33098_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_33099_cast_fp16 = softmax(axis = var_30355, x = aw_2797_cast_fp16)[name = tensor<string, []>("op_33099_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_33100_cast_fp16 = softmax(axis = var_30355, x = aw_2799_cast_fp16)[name = tensor<string, []>("op_33100_cast_fp16")];
            tensor<string, []> var_33102_equation_0 = const()[name = tensor<string, []>("op_33102_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33102_cast_fp16 = einsum(equation = var_33102_equation_0, values = (var_32922_cast_fp16, var_33081_cast_fp16))[name = tensor<string, []>("op_33102_cast_fp16")];
            tensor<string, []> var_33104_equation_0 = const()[name = tensor<string, []>("op_33104_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33104_cast_fp16 = einsum(equation = var_33104_equation_0, values = (var_32926_cast_fp16, var_33082_cast_fp16))[name = tensor<string, []>("op_33104_cast_fp16")];
            tensor<string, []> var_33106_equation_0 = const()[name = tensor<string, []>("op_33106_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33106_cast_fp16 = einsum(equation = var_33106_equation_0, values = (var_32930_cast_fp16, var_33083_cast_fp16))[name = tensor<string, []>("op_33106_cast_fp16")];
            tensor<string, []> var_33108_equation_0 = const()[name = tensor<string, []>("op_33108_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33108_cast_fp16 = einsum(equation = var_33108_equation_0, values = (var_32934_cast_fp16, var_33084_cast_fp16))[name = tensor<string, []>("op_33108_cast_fp16")];
            tensor<string, []> var_33110_equation_0 = const()[name = tensor<string, []>("op_33110_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33110_cast_fp16 = einsum(equation = var_33110_equation_0, values = (var_32938_cast_fp16, var_33085_cast_fp16))[name = tensor<string, []>("op_33110_cast_fp16")];
            tensor<string, []> var_33112_equation_0 = const()[name = tensor<string, []>("op_33112_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33112_cast_fp16 = einsum(equation = var_33112_equation_0, values = (var_32942_cast_fp16, var_33086_cast_fp16))[name = tensor<string, []>("op_33112_cast_fp16")];
            tensor<string, []> var_33114_equation_0 = const()[name = tensor<string, []>("op_33114_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33114_cast_fp16 = einsum(equation = var_33114_equation_0, values = (var_32946_cast_fp16, var_33087_cast_fp16))[name = tensor<string, []>("op_33114_cast_fp16")];
            tensor<string, []> var_33116_equation_0 = const()[name = tensor<string, []>("op_33116_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33116_cast_fp16 = einsum(equation = var_33116_equation_0, values = (var_32950_cast_fp16, var_33088_cast_fp16))[name = tensor<string, []>("op_33116_cast_fp16")];
            tensor<string, []> var_33118_equation_0 = const()[name = tensor<string, []>("op_33118_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33118_cast_fp16 = einsum(equation = var_33118_equation_0, values = (var_32954_cast_fp16, var_33089_cast_fp16))[name = tensor<string, []>("op_33118_cast_fp16")];
            tensor<string, []> var_33120_equation_0 = const()[name = tensor<string, []>("op_33120_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33120_cast_fp16 = einsum(equation = var_33120_equation_0, values = (var_32958_cast_fp16, var_33090_cast_fp16))[name = tensor<string, []>("op_33120_cast_fp16")];
            tensor<string, []> var_33122_equation_0 = const()[name = tensor<string, []>("op_33122_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33122_cast_fp16 = einsum(equation = var_33122_equation_0, values = (var_32962_cast_fp16, var_33091_cast_fp16))[name = tensor<string, []>("op_33122_cast_fp16")];
            tensor<string, []> var_33124_equation_0 = const()[name = tensor<string, []>("op_33124_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33124_cast_fp16 = einsum(equation = var_33124_equation_0, values = (var_32966_cast_fp16, var_33092_cast_fp16))[name = tensor<string, []>("op_33124_cast_fp16")];
            tensor<string, []> var_33126_equation_0 = const()[name = tensor<string, []>("op_33126_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33126_cast_fp16 = einsum(equation = var_33126_equation_0, values = (var_32970_cast_fp16, var_33093_cast_fp16))[name = tensor<string, []>("op_33126_cast_fp16")];
            tensor<string, []> var_33128_equation_0 = const()[name = tensor<string, []>("op_33128_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33128_cast_fp16 = einsum(equation = var_33128_equation_0, values = (var_32974_cast_fp16, var_33094_cast_fp16))[name = tensor<string, []>("op_33128_cast_fp16")];
            tensor<string, []> var_33130_equation_0 = const()[name = tensor<string, []>("op_33130_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33130_cast_fp16 = einsum(equation = var_33130_equation_0, values = (var_32978_cast_fp16, var_33095_cast_fp16))[name = tensor<string, []>("op_33130_cast_fp16")];
            tensor<string, []> var_33132_equation_0 = const()[name = tensor<string, []>("op_33132_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33132_cast_fp16 = einsum(equation = var_33132_equation_0, values = (var_32982_cast_fp16, var_33096_cast_fp16))[name = tensor<string, []>("op_33132_cast_fp16")];
            tensor<string, []> var_33134_equation_0 = const()[name = tensor<string, []>("op_33134_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33134_cast_fp16 = einsum(equation = var_33134_equation_0, values = (var_32986_cast_fp16, var_33097_cast_fp16))[name = tensor<string, []>("op_33134_cast_fp16")];
            tensor<string, []> var_33136_equation_0 = const()[name = tensor<string, []>("op_33136_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33136_cast_fp16 = einsum(equation = var_33136_equation_0, values = (var_32990_cast_fp16, var_33098_cast_fp16))[name = tensor<string, []>("op_33136_cast_fp16")];
            tensor<string, []> var_33138_equation_0 = const()[name = tensor<string, []>("op_33138_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33138_cast_fp16 = einsum(equation = var_33138_equation_0, values = (var_32994_cast_fp16, var_33099_cast_fp16))[name = tensor<string, []>("op_33138_cast_fp16")];
            tensor<string, []> var_33140_equation_0 = const()[name = tensor<string, []>("op_33140_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33140_cast_fp16 = einsum(equation = var_33140_equation_0, values = (var_32998_cast_fp16, var_33100_cast_fp16))[name = tensor<string, []>("op_33140_cast_fp16")];
            tensor<bool, []> input_457_interleave_0 = const()[name = tensor<string, []>("input_457_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_457_cast_fp16 = concat(axis = var_30355, interleave = input_457_interleave_0, values = (var_33102_cast_fp16, var_33104_cast_fp16, var_33106_cast_fp16, var_33108_cast_fp16, var_33110_cast_fp16, var_33112_cast_fp16, var_33114_cast_fp16, var_33116_cast_fp16, var_33118_cast_fp16, var_33120_cast_fp16, var_33122_cast_fp16, var_33124_cast_fp16, var_33126_cast_fp16, var_33128_cast_fp16, var_33130_cast_fp16, var_33132_cast_fp16, var_33134_cast_fp16, var_33136_cast_fp16, var_33138_cast_fp16, var_33140_cast_fp16))[name = tensor<string, []>("input_457_cast_fp16")];
            tensor<string, []> var_33150_pad_type_0 = const()[name = tensor<string, []>("op_33150_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_33150_strides_0 = const()[name = tensor<string, []>("op_33150_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_33150_pad_0 = const()[name = tensor<string, []>("op_33150_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_33150_dilations_0 = const()[name = tensor<string, []>("op_33150_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_33150_groups_0 = const()[name = tensor<string, []>("op_33150_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_2_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66002816))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67231680))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_2_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_2_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_2_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67231872)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_33150_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_2_attn2_to_out_0_bias_to_fp16, dilations = var_33150_dilations_0, groups = var_33150_groups_0, pad = var_33150_pad_0, pad_type = var_33150_pad_type_0, strides = var_33150_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_2_attn2_to_out_0_weight_to_fp16_palettized, x = input_457_cast_fp16)[name = tensor<string, []>("op_33150_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_221_cast_fp16 = add(x = var_33150_cast_fp16, y = inputs_219_cast_fp16)[name = tensor<string, []>("inputs_221_cast_fp16")];
            tensor<int32, [1]> input_459_axes_0 = const()[name = tensor<string, []>("input_459_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> input_459_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_459_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67234496)))];
            tensor<fp16, [1280]> input_459_beta_0_to_fp16 = const()[name = tensor<string, []>("input_459_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67237120)))];
            tensor<fp16, []> var_33160_to_fp16 = const()[name = tensor<string, []>("op_33160_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> input_459_cast_fp16 = layer_norm(axes = input_459_axes_0, beta = input_459_beta_0_to_fp16, epsilon = var_33160_to_fp16, gamma = input_459_gamma_0_to_fp16, x = inputs_221_cast_fp16)[name = tensor<string, []>("input_459_cast_fp16")];
            tensor<string, []> var_33180_pad_type_0 = const()[name = tensor<string, []>("op_33180_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_33180_strides_0 = const()[name = tensor<string, []>("op_33180_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_33180_pad_0 = const()[name = tensor<string, []>("op_33180_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_33180_dilations_0 = const()[name = tensor<string, []>("op_33180_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_33180_groups_0 = const()[name = tensor<string, []>("op_33180_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_2_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [9830400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67239744))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77070208))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_2_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([10240, 1280, 1, 1])];
            tensor<fp16, [10240]> up_blocks_0_attentions_0_transformer_blocks_2_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_2_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77070400)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_33180_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_2_ff_net_0_proj_bias_to_fp16, dilations = var_33180_dilations_0, groups = var_33180_groups_0, pad = var_33180_pad_0, pad_type = var_33180_pad_type_0, strides = var_33180_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_2_ff_net_0_proj_weight_to_fp16_palettized, x = input_459_cast_fp16)[name = tensor<string, []>("op_33180_cast_fp16")];
            tensor<int32, [2]> var_33181_split_sizes_0 = const()[name = tensor<string, []>("op_33181_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_33181_axis_0 = const()[name = tensor<string, []>("op_33181_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_33181_cast_fp16_0, tensor<fp16, [2, 5120, 1, 1024]> var_33181_cast_fp16_1 = split(axis = var_33181_axis_0, split_sizes = var_33181_split_sizes_0, x = var_33180_cast_fp16)[name = tensor<string, []>("op_33181_cast_fp16")];
            tensor<string, []> var_33183_mode_0 = const()[name = tensor<string, []>("op_33183_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_33183_cast_fp16 = gelu(mode = var_33183_mode_0, x = var_33181_cast_fp16_1)[name = tensor<string, []>("op_33183_cast_fp16")];
            tensor<fp16, [2, 5120, 1, 1024]> input_461_cast_fp16 = mul(x = var_33181_cast_fp16_0, y = var_33183_cast_fp16)[name = tensor<string, []>("input_461_cast_fp16")];
            tensor<string, []> var_33191_pad_type_0 = const()[name = tensor<string, []>("op_33191_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_33191_strides_0 = const()[name = tensor<string, []>("op_33191_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_33191_pad_0 = const()[name = tensor<string, []>("op_33191_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_33191_dilations_0 = const()[name = tensor<string, []>("op_33191_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_33191_groups_0 = const()[name = tensor<string, []>("op_33191_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_2_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [4915200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77090944))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82006208))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_2_ff_net_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 5120, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_2_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_2_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82006400)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_33191_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_2_ff_net_2_bias_to_fp16, dilations = var_33191_dilations_0, groups = var_33191_groups_0, pad = var_33191_pad_0, pad_type = var_33191_pad_type_0, strides = var_33191_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_2_ff_net_2_weight_to_fp16_palettized, x = input_461_cast_fp16)[name = tensor<string, []>("op_33191_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_223_cast_fp16 = add(x = var_33191_cast_fp16, y = inputs_221_cast_fp16)[name = tensor<string, []>("inputs_223_cast_fp16")];
            tensor<int32, [1]> hidden_states_305_axes_0 = const()[name = tensor<string, []>("hidden_states_305_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_305_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_305_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82009024)))];
            tensor<fp16, [1280]> hidden_states_305_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_305_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82011648)))];
            tensor<fp16, []> var_33207_to_fp16 = const()[name = tensor<string, []>("op_33207_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_305_cast_fp16 = layer_norm(axes = hidden_states_305_axes_0, beta = hidden_states_305_beta_0_to_fp16, epsilon = var_33207_to_fp16, gamma = hidden_states_305_gamma_0_to_fp16, x = inputs_223_cast_fp16)[name = tensor<string, []>("hidden_states_305_cast_fp16")];
            tensor<string, []> q_149_pad_type_0 = const()[name = tensor<string, []>("q_149_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_149_strides_0 = const()[name = tensor<string, []>("q_149_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_149_pad_0 = const()[name = tensor<string, []>("q_149_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_149_dilations_0 = const()[name = tensor<string, []>("q_149_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_149_groups_0 = const()[name = tensor<string, []>("q_149_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_3_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82014272))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83243136))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_3_attn1_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_149_cast_fp16 = conv(dilations = q_149_dilations_0, groups = q_149_groups_0, pad = q_149_pad_0, pad_type = q_149_pad_type_0, strides = q_149_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_3_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_305_cast_fp16)[name = tensor<string, []>("q_149_cast_fp16")];
            tensor<string, []> k_297_pad_type_0 = const()[name = tensor<string, []>("k_297_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_297_strides_0 = const()[name = tensor<string, []>("k_297_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_297_pad_0 = const()[name = tensor<string, []>("k_297_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_297_dilations_0 = const()[name = tensor<string, []>("k_297_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_297_groups_0 = const()[name = tensor<string, []>("k_297_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_3_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83243328))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(84472192))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_3_attn1_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> k_297_cast_fp16 = conv(dilations = k_297_dilations_0, groups = k_297_groups_0, pad = k_297_pad_0, pad_type = k_297_pad_type_0, strides = k_297_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_3_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_305_cast_fp16)[name = tensor<string, []>("k_297_cast_fp16")];
            tensor<string, []> v_149_pad_type_0 = const()[name = tensor<string, []>("v_149_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_149_strides_0 = const()[name = tensor<string, []>("v_149_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_149_pad_0 = const()[name = tensor<string, []>("v_149_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_149_dilations_0 = const()[name = tensor<string, []>("v_149_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_149_groups_0 = const()[name = tensor<string, []>("v_149_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_3_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(84472384))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(85701248))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_3_attn1_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> v_149_cast_fp16 = conv(dilations = v_149_dilations_0, groups = v_149_groups_0, pad = v_149_pad_0, pad_type = v_149_pad_type_0, strides = v_149_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_3_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_305_cast_fp16)[name = tensor<string, []>("v_149_cast_fp16")];
            tensor<int32, [4]> var_33240_begin_0 = const()[name = tensor<string, []>("op_33240_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_33240_end_0 = const()[name = tensor<string, []>("op_33240_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_33240_end_mask_0 = const()[name = tensor<string, []>("op_33240_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33240_cast_fp16 = slice_by_index(begin = var_33240_begin_0, end = var_33240_end_0, end_mask = var_33240_end_mask_0, x = q_149_cast_fp16)[name = tensor<string, []>("op_33240_cast_fp16")];
            tensor<int32, [4]> var_33244_begin_0 = const()[name = tensor<string, []>("op_33244_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_33244_end_0 = const()[name = tensor<string, []>("op_33244_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_33244_end_mask_0 = const()[name = tensor<string, []>("op_33244_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33244_cast_fp16 = slice_by_index(begin = var_33244_begin_0, end = var_33244_end_0, end_mask = var_33244_end_mask_0, x = q_149_cast_fp16)[name = tensor<string, []>("op_33244_cast_fp16")];
            tensor<int32, [4]> var_33248_begin_0 = const()[name = tensor<string, []>("op_33248_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_33248_end_0 = const()[name = tensor<string, []>("op_33248_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_33248_end_mask_0 = const()[name = tensor<string, []>("op_33248_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33248_cast_fp16 = slice_by_index(begin = var_33248_begin_0, end = var_33248_end_0, end_mask = var_33248_end_mask_0, x = q_149_cast_fp16)[name = tensor<string, []>("op_33248_cast_fp16")];
            tensor<int32, [4]> var_33252_begin_0 = const()[name = tensor<string, []>("op_33252_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_33252_end_0 = const()[name = tensor<string, []>("op_33252_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_33252_end_mask_0 = const()[name = tensor<string, []>("op_33252_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33252_cast_fp16 = slice_by_index(begin = var_33252_begin_0, end = var_33252_end_0, end_mask = var_33252_end_mask_0, x = q_149_cast_fp16)[name = tensor<string, []>("op_33252_cast_fp16")];
            tensor<int32, [4]> var_33256_begin_0 = const()[name = tensor<string, []>("op_33256_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_33256_end_0 = const()[name = tensor<string, []>("op_33256_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_33256_end_mask_0 = const()[name = tensor<string, []>("op_33256_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33256_cast_fp16 = slice_by_index(begin = var_33256_begin_0, end = var_33256_end_0, end_mask = var_33256_end_mask_0, x = q_149_cast_fp16)[name = tensor<string, []>("op_33256_cast_fp16")];
            tensor<int32, [4]> var_33260_begin_0 = const()[name = tensor<string, []>("op_33260_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_33260_end_0 = const()[name = tensor<string, []>("op_33260_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_33260_end_mask_0 = const()[name = tensor<string, []>("op_33260_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33260_cast_fp16 = slice_by_index(begin = var_33260_begin_0, end = var_33260_end_0, end_mask = var_33260_end_mask_0, x = q_149_cast_fp16)[name = tensor<string, []>("op_33260_cast_fp16")];
            tensor<int32, [4]> var_33264_begin_0 = const()[name = tensor<string, []>("op_33264_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_33264_end_0 = const()[name = tensor<string, []>("op_33264_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_33264_end_mask_0 = const()[name = tensor<string, []>("op_33264_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33264_cast_fp16 = slice_by_index(begin = var_33264_begin_0, end = var_33264_end_0, end_mask = var_33264_end_mask_0, x = q_149_cast_fp16)[name = tensor<string, []>("op_33264_cast_fp16")];
            tensor<int32, [4]> var_33268_begin_0 = const()[name = tensor<string, []>("op_33268_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_33268_end_0 = const()[name = tensor<string, []>("op_33268_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_33268_end_mask_0 = const()[name = tensor<string, []>("op_33268_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33268_cast_fp16 = slice_by_index(begin = var_33268_begin_0, end = var_33268_end_0, end_mask = var_33268_end_mask_0, x = q_149_cast_fp16)[name = tensor<string, []>("op_33268_cast_fp16")];
            tensor<int32, [4]> var_33272_begin_0 = const()[name = tensor<string, []>("op_33272_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_33272_end_0 = const()[name = tensor<string, []>("op_33272_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_33272_end_mask_0 = const()[name = tensor<string, []>("op_33272_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33272_cast_fp16 = slice_by_index(begin = var_33272_begin_0, end = var_33272_end_0, end_mask = var_33272_end_mask_0, x = q_149_cast_fp16)[name = tensor<string, []>("op_33272_cast_fp16")];
            tensor<int32, [4]> var_33276_begin_0 = const()[name = tensor<string, []>("op_33276_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_33276_end_0 = const()[name = tensor<string, []>("op_33276_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_33276_end_mask_0 = const()[name = tensor<string, []>("op_33276_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33276_cast_fp16 = slice_by_index(begin = var_33276_begin_0, end = var_33276_end_0, end_mask = var_33276_end_mask_0, x = q_149_cast_fp16)[name = tensor<string, []>("op_33276_cast_fp16")];
            tensor<int32, [4]> var_33280_begin_0 = const()[name = tensor<string, []>("op_33280_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_33280_end_0 = const()[name = tensor<string, []>("op_33280_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_33280_end_mask_0 = const()[name = tensor<string, []>("op_33280_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33280_cast_fp16 = slice_by_index(begin = var_33280_begin_0, end = var_33280_end_0, end_mask = var_33280_end_mask_0, x = q_149_cast_fp16)[name = tensor<string, []>("op_33280_cast_fp16")];
            tensor<int32, [4]> var_33284_begin_0 = const()[name = tensor<string, []>("op_33284_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_33284_end_0 = const()[name = tensor<string, []>("op_33284_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_33284_end_mask_0 = const()[name = tensor<string, []>("op_33284_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33284_cast_fp16 = slice_by_index(begin = var_33284_begin_0, end = var_33284_end_0, end_mask = var_33284_end_mask_0, x = q_149_cast_fp16)[name = tensor<string, []>("op_33284_cast_fp16")];
            tensor<int32, [4]> var_33288_begin_0 = const()[name = tensor<string, []>("op_33288_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_33288_end_0 = const()[name = tensor<string, []>("op_33288_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_33288_end_mask_0 = const()[name = tensor<string, []>("op_33288_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33288_cast_fp16 = slice_by_index(begin = var_33288_begin_0, end = var_33288_end_0, end_mask = var_33288_end_mask_0, x = q_149_cast_fp16)[name = tensor<string, []>("op_33288_cast_fp16")];
            tensor<int32, [4]> var_33292_begin_0 = const()[name = tensor<string, []>("op_33292_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_33292_end_0 = const()[name = tensor<string, []>("op_33292_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_33292_end_mask_0 = const()[name = tensor<string, []>("op_33292_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33292_cast_fp16 = slice_by_index(begin = var_33292_begin_0, end = var_33292_end_0, end_mask = var_33292_end_mask_0, x = q_149_cast_fp16)[name = tensor<string, []>("op_33292_cast_fp16")];
            tensor<int32, [4]> var_33296_begin_0 = const()[name = tensor<string, []>("op_33296_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_33296_end_0 = const()[name = tensor<string, []>("op_33296_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_33296_end_mask_0 = const()[name = tensor<string, []>("op_33296_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33296_cast_fp16 = slice_by_index(begin = var_33296_begin_0, end = var_33296_end_0, end_mask = var_33296_end_mask_0, x = q_149_cast_fp16)[name = tensor<string, []>("op_33296_cast_fp16")];
            tensor<int32, [4]> var_33300_begin_0 = const()[name = tensor<string, []>("op_33300_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_33300_end_0 = const()[name = tensor<string, []>("op_33300_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_33300_end_mask_0 = const()[name = tensor<string, []>("op_33300_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33300_cast_fp16 = slice_by_index(begin = var_33300_begin_0, end = var_33300_end_0, end_mask = var_33300_end_mask_0, x = q_149_cast_fp16)[name = tensor<string, []>("op_33300_cast_fp16")];
            tensor<int32, [4]> var_33304_begin_0 = const()[name = tensor<string, []>("op_33304_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_33304_end_0 = const()[name = tensor<string, []>("op_33304_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_33304_end_mask_0 = const()[name = tensor<string, []>("op_33304_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33304_cast_fp16 = slice_by_index(begin = var_33304_begin_0, end = var_33304_end_0, end_mask = var_33304_end_mask_0, x = q_149_cast_fp16)[name = tensor<string, []>("op_33304_cast_fp16")];
            tensor<int32, [4]> var_33308_begin_0 = const()[name = tensor<string, []>("op_33308_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_33308_end_0 = const()[name = tensor<string, []>("op_33308_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_33308_end_mask_0 = const()[name = tensor<string, []>("op_33308_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33308_cast_fp16 = slice_by_index(begin = var_33308_begin_0, end = var_33308_end_0, end_mask = var_33308_end_mask_0, x = q_149_cast_fp16)[name = tensor<string, []>("op_33308_cast_fp16")];
            tensor<int32, [4]> var_33312_begin_0 = const()[name = tensor<string, []>("op_33312_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_33312_end_0 = const()[name = tensor<string, []>("op_33312_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_33312_end_mask_0 = const()[name = tensor<string, []>("op_33312_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33312_cast_fp16 = slice_by_index(begin = var_33312_begin_0, end = var_33312_end_0, end_mask = var_33312_end_mask_0, x = q_149_cast_fp16)[name = tensor<string, []>("op_33312_cast_fp16")];
            tensor<int32, [4]> var_33316_begin_0 = const()[name = tensor<string, []>("op_33316_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_33316_end_0 = const()[name = tensor<string, []>("op_33316_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_33316_end_mask_0 = const()[name = tensor<string, []>("op_33316_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33316_cast_fp16 = slice_by_index(begin = var_33316_begin_0, end = var_33316_end_0, end_mask = var_33316_end_mask_0, x = q_149_cast_fp16)[name = tensor<string, []>("op_33316_cast_fp16")];
            tensor<int32, [4]> k_299_perm_0 = const()[name = tensor<string, []>("k_299_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_33323_begin_0 = const()[name = tensor<string, []>("op_33323_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_33323_end_0 = const()[name = tensor<string, []>("op_33323_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_33323_end_mask_0 = const()[name = tensor<string, []>("op_33323_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> k_299_cast_fp16 = transpose(perm = k_299_perm_0, x = k_297_cast_fp16)[name = tensor<string, []>("transpose_65")];
            tensor<fp16, [2, 1024, 1, 64]> var_33323_cast_fp16 = slice_by_index(begin = var_33323_begin_0, end = var_33323_end_0, end_mask = var_33323_end_mask_0, x = k_299_cast_fp16)[name = tensor<string, []>("op_33323_cast_fp16")];
            tensor<int32, [4]> var_33327_begin_0 = const()[name = tensor<string, []>("op_33327_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_33327_end_0 = const()[name = tensor<string, []>("op_33327_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_33327_end_mask_0 = const()[name = tensor<string, []>("op_33327_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_33327_cast_fp16 = slice_by_index(begin = var_33327_begin_0, end = var_33327_end_0, end_mask = var_33327_end_mask_0, x = k_299_cast_fp16)[name = tensor<string, []>("op_33327_cast_fp16")];
            tensor<int32, [4]> var_33331_begin_0 = const()[name = tensor<string, []>("op_33331_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_33331_end_0 = const()[name = tensor<string, []>("op_33331_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_33331_end_mask_0 = const()[name = tensor<string, []>("op_33331_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_33331_cast_fp16 = slice_by_index(begin = var_33331_begin_0, end = var_33331_end_0, end_mask = var_33331_end_mask_0, x = k_299_cast_fp16)[name = tensor<string, []>("op_33331_cast_fp16")];
            tensor<int32, [4]> var_33335_begin_0 = const()[name = tensor<string, []>("op_33335_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_33335_end_0 = const()[name = tensor<string, []>("op_33335_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_33335_end_mask_0 = const()[name = tensor<string, []>("op_33335_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_33335_cast_fp16 = slice_by_index(begin = var_33335_begin_0, end = var_33335_end_0, end_mask = var_33335_end_mask_0, x = k_299_cast_fp16)[name = tensor<string, []>("op_33335_cast_fp16")];
            tensor<int32, [4]> var_33339_begin_0 = const()[name = tensor<string, []>("op_33339_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_33339_end_0 = const()[name = tensor<string, []>("op_33339_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_33339_end_mask_0 = const()[name = tensor<string, []>("op_33339_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_33339_cast_fp16 = slice_by_index(begin = var_33339_begin_0, end = var_33339_end_0, end_mask = var_33339_end_mask_0, x = k_299_cast_fp16)[name = tensor<string, []>("op_33339_cast_fp16")];
            tensor<int32, [4]> var_33343_begin_0 = const()[name = tensor<string, []>("op_33343_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_33343_end_0 = const()[name = tensor<string, []>("op_33343_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_33343_end_mask_0 = const()[name = tensor<string, []>("op_33343_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_33343_cast_fp16 = slice_by_index(begin = var_33343_begin_0, end = var_33343_end_0, end_mask = var_33343_end_mask_0, x = k_299_cast_fp16)[name = tensor<string, []>("op_33343_cast_fp16")];
            tensor<int32, [4]> var_33347_begin_0 = const()[name = tensor<string, []>("op_33347_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_33347_end_0 = const()[name = tensor<string, []>("op_33347_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_33347_end_mask_0 = const()[name = tensor<string, []>("op_33347_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_33347_cast_fp16 = slice_by_index(begin = var_33347_begin_0, end = var_33347_end_0, end_mask = var_33347_end_mask_0, x = k_299_cast_fp16)[name = tensor<string, []>("op_33347_cast_fp16")];
            tensor<int32, [4]> var_33351_begin_0 = const()[name = tensor<string, []>("op_33351_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_33351_end_0 = const()[name = tensor<string, []>("op_33351_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_33351_end_mask_0 = const()[name = tensor<string, []>("op_33351_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_33351_cast_fp16 = slice_by_index(begin = var_33351_begin_0, end = var_33351_end_0, end_mask = var_33351_end_mask_0, x = k_299_cast_fp16)[name = tensor<string, []>("op_33351_cast_fp16")];
            tensor<int32, [4]> var_33355_begin_0 = const()[name = tensor<string, []>("op_33355_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_33355_end_0 = const()[name = tensor<string, []>("op_33355_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_33355_end_mask_0 = const()[name = tensor<string, []>("op_33355_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_33355_cast_fp16 = slice_by_index(begin = var_33355_begin_0, end = var_33355_end_0, end_mask = var_33355_end_mask_0, x = k_299_cast_fp16)[name = tensor<string, []>("op_33355_cast_fp16")];
            tensor<int32, [4]> var_33359_begin_0 = const()[name = tensor<string, []>("op_33359_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_33359_end_0 = const()[name = tensor<string, []>("op_33359_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_33359_end_mask_0 = const()[name = tensor<string, []>("op_33359_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_33359_cast_fp16 = slice_by_index(begin = var_33359_begin_0, end = var_33359_end_0, end_mask = var_33359_end_mask_0, x = k_299_cast_fp16)[name = tensor<string, []>("op_33359_cast_fp16")];
            tensor<int32, [4]> var_33363_begin_0 = const()[name = tensor<string, []>("op_33363_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_33363_end_0 = const()[name = tensor<string, []>("op_33363_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_33363_end_mask_0 = const()[name = tensor<string, []>("op_33363_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_33363_cast_fp16 = slice_by_index(begin = var_33363_begin_0, end = var_33363_end_0, end_mask = var_33363_end_mask_0, x = k_299_cast_fp16)[name = tensor<string, []>("op_33363_cast_fp16")];
            tensor<int32, [4]> var_33367_begin_0 = const()[name = tensor<string, []>("op_33367_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_33367_end_0 = const()[name = tensor<string, []>("op_33367_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_33367_end_mask_0 = const()[name = tensor<string, []>("op_33367_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_33367_cast_fp16 = slice_by_index(begin = var_33367_begin_0, end = var_33367_end_0, end_mask = var_33367_end_mask_0, x = k_299_cast_fp16)[name = tensor<string, []>("op_33367_cast_fp16")];
            tensor<int32, [4]> var_33371_begin_0 = const()[name = tensor<string, []>("op_33371_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_33371_end_0 = const()[name = tensor<string, []>("op_33371_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_33371_end_mask_0 = const()[name = tensor<string, []>("op_33371_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_33371_cast_fp16 = slice_by_index(begin = var_33371_begin_0, end = var_33371_end_0, end_mask = var_33371_end_mask_0, x = k_299_cast_fp16)[name = tensor<string, []>("op_33371_cast_fp16")];
            tensor<int32, [4]> var_33375_begin_0 = const()[name = tensor<string, []>("op_33375_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_33375_end_0 = const()[name = tensor<string, []>("op_33375_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_33375_end_mask_0 = const()[name = tensor<string, []>("op_33375_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_33375_cast_fp16 = slice_by_index(begin = var_33375_begin_0, end = var_33375_end_0, end_mask = var_33375_end_mask_0, x = k_299_cast_fp16)[name = tensor<string, []>("op_33375_cast_fp16")];
            tensor<int32, [4]> var_33379_begin_0 = const()[name = tensor<string, []>("op_33379_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_33379_end_0 = const()[name = tensor<string, []>("op_33379_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_33379_end_mask_0 = const()[name = tensor<string, []>("op_33379_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_33379_cast_fp16 = slice_by_index(begin = var_33379_begin_0, end = var_33379_end_0, end_mask = var_33379_end_mask_0, x = k_299_cast_fp16)[name = tensor<string, []>("op_33379_cast_fp16")];
            tensor<int32, [4]> var_33383_begin_0 = const()[name = tensor<string, []>("op_33383_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_33383_end_0 = const()[name = tensor<string, []>("op_33383_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_33383_end_mask_0 = const()[name = tensor<string, []>("op_33383_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_33383_cast_fp16 = slice_by_index(begin = var_33383_begin_0, end = var_33383_end_0, end_mask = var_33383_end_mask_0, x = k_299_cast_fp16)[name = tensor<string, []>("op_33383_cast_fp16")];
            tensor<int32, [4]> var_33387_begin_0 = const()[name = tensor<string, []>("op_33387_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_33387_end_0 = const()[name = tensor<string, []>("op_33387_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_33387_end_mask_0 = const()[name = tensor<string, []>("op_33387_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_33387_cast_fp16 = slice_by_index(begin = var_33387_begin_0, end = var_33387_end_0, end_mask = var_33387_end_mask_0, x = k_299_cast_fp16)[name = tensor<string, []>("op_33387_cast_fp16")];
            tensor<int32, [4]> var_33391_begin_0 = const()[name = tensor<string, []>("op_33391_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_33391_end_0 = const()[name = tensor<string, []>("op_33391_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_33391_end_mask_0 = const()[name = tensor<string, []>("op_33391_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_33391_cast_fp16 = slice_by_index(begin = var_33391_begin_0, end = var_33391_end_0, end_mask = var_33391_end_mask_0, x = k_299_cast_fp16)[name = tensor<string, []>("op_33391_cast_fp16")];
            tensor<int32, [4]> var_33395_begin_0 = const()[name = tensor<string, []>("op_33395_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_33395_end_0 = const()[name = tensor<string, []>("op_33395_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_33395_end_mask_0 = const()[name = tensor<string, []>("op_33395_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_33395_cast_fp16 = slice_by_index(begin = var_33395_begin_0, end = var_33395_end_0, end_mask = var_33395_end_mask_0, x = k_299_cast_fp16)[name = tensor<string, []>("op_33395_cast_fp16")];
            tensor<int32, [4]> var_33399_begin_0 = const()[name = tensor<string, []>("op_33399_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_33399_end_0 = const()[name = tensor<string, []>("op_33399_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_33399_end_mask_0 = const()[name = tensor<string, []>("op_33399_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_33399_cast_fp16 = slice_by_index(begin = var_33399_begin_0, end = var_33399_end_0, end_mask = var_33399_end_mask_0, x = k_299_cast_fp16)[name = tensor<string, []>("op_33399_cast_fp16")];
            tensor<int32, [4]> var_33401_begin_0 = const()[name = tensor<string, []>("op_33401_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_33401_end_0 = const()[name = tensor<string, []>("op_33401_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_33401_end_mask_0 = const()[name = tensor<string, []>("op_33401_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33401_cast_fp16 = slice_by_index(begin = var_33401_begin_0, end = var_33401_end_0, end_mask = var_33401_end_mask_0, x = v_149_cast_fp16)[name = tensor<string, []>("op_33401_cast_fp16")];
            tensor<int32, [4]> var_33405_begin_0 = const()[name = tensor<string, []>("op_33405_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_33405_end_0 = const()[name = tensor<string, []>("op_33405_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_33405_end_mask_0 = const()[name = tensor<string, []>("op_33405_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33405_cast_fp16 = slice_by_index(begin = var_33405_begin_0, end = var_33405_end_0, end_mask = var_33405_end_mask_0, x = v_149_cast_fp16)[name = tensor<string, []>("op_33405_cast_fp16")];
            tensor<int32, [4]> var_33409_begin_0 = const()[name = tensor<string, []>("op_33409_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_33409_end_0 = const()[name = tensor<string, []>("op_33409_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_33409_end_mask_0 = const()[name = tensor<string, []>("op_33409_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33409_cast_fp16 = slice_by_index(begin = var_33409_begin_0, end = var_33409_end_0, end_mask = var_33409_end_mask_0, x = v_149_cast_fp16)[name = tensor<string, []>("op_33409_cast_fp16")];
            tensor<int32, [4]> var_33413_begin_0 = const()[name = tensor<string, []>("op_33413_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_33413_end_0 = const()[name = tensor<string, []>("op_33413_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_33413_end_mask_0 = const()[name = tensor<string, []>("op_33413_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33413_cast_fp16 = slice_by_index(begin = var_33413_begin_0, end = var_33413_end_0, end_mask = var_33413_end_mask_0, x = v_149_cast_fp16)[name = tensor<string, []>("op_33413_cast_fp16")];
            tensor<int32, [4]> var_33417_begin_0 = const()[name = tensor<string, []>("op_33417_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_33417_end_0 = const()[name = tensor<string, []>("op_33417_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_33417_end_mask_0 = const()[name = tensor<string, []>("op_33417_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33417_cast_fp16 = slice_by_index(begin = var_33417_begin_0, end = var_33417_end_0, end_mask = var_33417_end_mask_0, x = v_149_cast_fp16)[name = tensor<string, []>("op_33417_cast_fp16")];
            tensor<int32, [4]> var_33421_begin_0 = const()[name = tensor<string, []>("op_33421_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_33421_end_0 = const()[name = tensor<string, []>("op_33421_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_33421_end_mask_0 = const()[name = tensor<string, []>("op_33421_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33421_cast_fp16 = slice_by_index(begin = var_33421_begin_0, end = var_33421_end_0, end_mask = var_33421_end_mask_0, x = v_149_cast_fp16)[name = tensor<string, []>("op_33421_cast_fp16")];
            tensor<int32, [4]> var_33425_begin_0 = const()[name = tensor<string, []>("op_33425_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_33425_end_0 = const()[name = tensor<string, []>("op_33425_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_33425_end_mask_0 = const()[name = tensor<string, []>("op_33425_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33425_cast_fp16 = slice_by_index(begin = var_33425_begin_0, end = var_33425_end_0, end_mask = var_33425_end_mask_0, x = v_149_cast_fp16)[name = tensor<string, []>("op_33425_cast_fp16")];
            tensor<int32, [4]> var_33429_begin_0 = const()[name = tensor<string, []>("op_33429_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_33429_end_0 = const()[name = tensor<string, []>("op_33429_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_33429_end_mask_0 = const()[name = tensor<string, []>("op_33429_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33429_cast_fp16 = slice_by_index(begin = var_33429_begin_0, end = var_33429_end_0, end_mask = var_33429_end_mask_0, x = v_149_cast_fp16)[name = tensor<string, []>("op_33429_cast_fp16")];
            tensor<int32, [4]> var_33433_begin_0 = const()[name = tensor<string, []>("op_33433_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_33433_end_0 = const()[name = tensor<string, []>("op_33433_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_33433_end_mask_0 = const()[name = tensor<string, []>("op_33433_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33433_cast_fp16 = slice_by_index(begin = var_33433_begin_0, end = var_33433_end_0, end_mask = var_33433_end_mask_0, x = v_149_cast_fp16)[name = tensor<string, []>("op_33433_cast_fp16")];
            tensor<int32, [4]> var_33437_begin_0 = const()[name = tensor<string, []>("op_33437_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_33437_end_0 = const()[name = tensor<string, []>("op_33437_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_33437_end_mask_0 = const()[name = tensor<string, []>("op_33437_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33437_cast_fp16 = slice_by_index(begin = var_33437_begin_0, end = var_33437_end_0, end_mask = var_33437_end_mask_0, x = v_149_cast_fp16)[name = tensor<string, []>("op_33437_cast_fp16")];
            tensor<int32, [4]> var_33441_begin_0 = const()[name = tensor<string, []>("op_33441_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_33441_end_0 = const()[name = tensor<string, []>("op_33441_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_33441_end_mask_0 = const()[name = tensor<string, []>("op_33441_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33441_cast_fp16 = slice_by_index(begin = var_33441_begin_0, end = var_33441_end_0, end_mask = var_33441_end_mask_0, x = v_149_cast_fp16)[name = tensor<string, []>("op_33441_cast_fp16")];
            tensor<int32, [4]> var_33445_begin_0 = const()[name = tensor<string, []>("op_33445_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_33445_end_0 = const()[name = tensor<string, []>("op_33445_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_33445_end_mask_0 = const()[name = tensor<string, []>("op_33445_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33445_cast_fp16 = slice_by_index(begin = var_33445_begin_0, end = var_33445_end_0, end_mask = var_33445_end_mask_0, x = v_149_cast_fp16)[name = tensor<string, []>("op_33445_cast_fp16")];
            tensor<int32, [4]> var_33449_begin_0 = const()[name = tensor<string, []>("op_33449_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_33449_end_0 = const()[name = tensor<string, []>("op_33449_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_33449_end_mask_0 = const()[name = tensor<string, []>("op_33449_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33449_cast_fp16 = slice_by_index(begin = var_33449_begin_0, end = var_33449_end_0, end_mask = var_33449_end_mask_0, x = v_149_cast_fp16)[name = tensor<string, []>("op_33449_cast_fp16")];
            tensor<int32, [4]> var_33453_begin_0 = const()[name = tensor<string, []>("op_33453_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_33453_end_0 = const()[name = tensor<string, []>("op_33453_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_33453_end_mask_0 = const()[name = tensor<string, []>("op_33453_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33453_cast_fp16 = slice_by_index(begin = var_33453_begin_0, end = var_33453_end_0, end_mask = var_33453_end_mask_0, x = v_149_cast_fp16)[name = tensor<string, []>("op_33453_cast_fp16")];
            tensor<int32, [4]> var_33457_begin_0 = const()[name = tensor<string, []>("op_33457_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_33457_end_0 = const()[name = tensor<string, []>("op_33457_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_33457_end_mask_0 = const()[name = tensor<string, []>("op_33457_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33457_cast_fp16 = slice_by_index(begin = var_33457_begin_0, end = var_33457_end_0, end_mask = var_33457_end_mask_0, x = v_149_cast_fp16)[name = tensor<string, []>("op_33457_cast_fp16")];
            tensor<int32, [4]> var_33461_begin_0 = const()[name = tensor<string, []>("op_33461_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_33461_end_0 = const()[name = tensor<string, []>("op_33461_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_33461_end_mask_0 = const()[name = tensor<string, []>("op_33461_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33461_cast_fp16 = slice_by_index(begin = var_33461_begin_0, end = var_33461_end_0, end_mask = var_33461_end_mask_0, x = v_149_cast_fp16)[name = tensor<string, []>("op_33461_cast_fp16")];
            tensor<int32, [4]> var_33465_begin_0 = const()[name = tensor<string, []>("op_33465_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_33465_end_0 = const()[name = tensor<string, []>("op_33465_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_33465_end_mask_0 = const()[name = tensor<string, []>("op_33465_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33465_cast_fp16 = slice_by_index(begin = var_33465_begin_0, end = var_33465_end_0, end_mask = var_33465_end_mask_0, x = v_149_cast_fp16)[name = tensor<string, []>("op_33465_cast_fp16")];
            tensor<int32, [4]> var_33469_begin_0 = const()[name = tensor<string, []>("op_33469_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_33469_end_0 = const()[name = tensor<string, []>("op_33469_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_33469_end_mask_0 = const()[name = tensor<string, []>("op_33469_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33469_cast_fp16 = slice_by_index(begin = var_33469_begin_0, end = var_33469_end_0, end_mask = var_33469_end_mask_0, x = v_149_cast_fp16)[name = tensor<string, []>("op_33469_cast_fp16")];
            tensor<int32, [4]> var_33473_begin_0 = const()[name = tensor<string, []>("op_33473_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_33473_end_0 = const()[name = tensor<string, []>("op_33473_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_33473_end_mask_0 = const()[name = tensor<string, []>("op_33473_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33473_cast_fp16 = slice_by_index(begin = var_33473_begin_0, end = var_33473_end_0, end_mask = var_33473_end_mask_0, x = v_149_cast_fp16)[name = tensor<string, []>("op_33473_cast_fp16")];
            tensor<int32, [4]> var_33477_begin_0 = const()[name = tensor<string, []>("op_33477_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_33477_end_0 = const()[name = tensor<string, []>("op_33477_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_33477_end_mask_0 = const()[name = tensor<string, []>("op_33477_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33477_cast_fp16 = slice_by_index(begin = var_33477_begin_0, end = var_33477_end_0, end_mask = var_33477_end_mask_0, x = v_149_cast_fp16)[name = tensor<string, []>("op_33477_cast_fp16")];
            tensor<string, []> var_33481_equation_0 = const()[name = tensor<string, []>("op_33481_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33481_cast_fp16 = einsum(equation = var_33481_equation_0, values = (var_33323_cast_fp16, var_33240_cast_fp16))[name = tensor<string, []>("op_33481_cast_fp16")];
            tensor<fp16, []> var_33482_to_fp16 = const()[name = tensor<string, []>("op_33482_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2801_cast_fp16 = mul(x = var_33481_cast_fp16, y = var_33482_to_fp16)[name = tensor<string, []>("aw_2801_cast_fp16")];
            tensor<string, []> var_33485_equation_0 = const()[name = tensor<string, []>("op_33485_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33485_cast_fp16 = einsum(equation = var_33485_equation_0, values = (var_33327_cast_fp16, var_33244_cast_fp16))[name = tensor<string, []>("op_33485_cast_fp16")];
            tensor<fp16, []> var_33486_to_fp16 = const()[name = tensor<string, []>("op_33486_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2803_cast_fp16 = mul(x = var_33485_cast_fp16, y = var_33486_to_fp16)[name = tensor<string, []>("aw_2803_cast_fp16")];
            tensor<string, []> var_33489_equation_0 = const()[name = tensor<string, []>("op_33489_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33489_cast_fp16 = einsum(equation = var_33489_equation_0, values = (var_33331_cast_fp16, var_33248_cast_fp16))[name = tensor<string, []>("op_33489_cast_fp16")];
            tensor<fp16, []> var_33490_to_fp16 = const()[name = tensor<string, []>("op_33490_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2805_cast_fp16 = mul(x = var_33489_cast_fp16, y = var_33490_to_fp16)[name = tensor<string, []>("aw_2805_cast_fp16")];
            tensor<string, []> var_33493_equation_0 = const()[name = tensor<string, []>("op_33493_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33493_cast_fp16 = einsum(equation = var_33493_equation_0, values = (var_33335_cast_fp16, var_33252_cast_fp16))[name = tensor<string, []>("op_33493_cast_fp16")];
            tensor<fp16, []> var_33494_to_fp16 = const()[name = tensor<string, []>("op_33494_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2807_cast_fp16 = mul(x = var_33493_cast_fp16, y = var_33494_to_fp16)[name = tensor<string, []>("aw_2807_cast_fp16")];
            tensor<string, []> var_33497_equation_0 = const()[name = tensor<string, []>("op_33497_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33497_cast_fp16 = einsum(equation = var_33497_equation_0, values = (var_33339_cast_fp16, var_33256_cast_fp16))[name = tensor<string, []>("op_33497_cast_fp16")];
            tensor<fp16, []> var_33498_to_fp16 = const()[name = tensor<string, []>("op_33498_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2809_cast_fp16 = mul(x = var_33497_cast_fp16, y = var_33498_to_fp16)[name = tensor<string, []>("aw_2809_cast_fp16")];
            tensor<string, []> var_33501_equation_0 = const()[name = tensor<string, []>("op_33501_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33501_cast_fp16 = einsum(equation = var_33501_equation_0, values = (var_33343_cast_fp16, var_33260_cast_fp16))[name = tensor<string, []>("op_33501_cast_fp16")];
            tensor<fp16, []> var_33502_to_fp16 = const()[name = tensor<string, []>("op_33502_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2811_cast_fp16 = mul(x = var_33501_cast_fp16, y = var_33502_to_fp16)[name = tensor<string, []>("aw_2811_cast_fp16")];
            tensor<string, []> var_33505_equation_0 = const()[name = tensor<string, []>("op_33505_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33505_cast_fp16 = einsum(equation = var_33505_equation_0, values = (var_33347_cast_fp16, var_33264_cast_fp16))[name = tensor<string, []>("op_33505_cast_fp16")];
            tensor<fp16, []> var_33506_to_fp16 = const()[name = tensor<string, []>("op_33506_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2813_cast_fp16 = mul(x = var_33505_cast_fp16, y = var_33506_to_fp16)[name = tensor<string, []>("aw_2813_cast_fp16")];
            tensor<string, []> var_33509_equation_0 = const()[name = tensor<string, []>("op_33509_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33509_cast_fp16 = einsum(equation = var_33509_equation_0, values = (var_33351_cast_fp16, var_33268_cast_fp16))[name = tensor<string, []>("op_33509_cast_fp16")];
            tensor<fp16, []> var_33510_to_fp16 = const()[name = tensor<string, []>("op_33510_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2815_cast_fp16 = mul(x = var_33509_cast_fp16, y = var_33510_to_fp16)[name = tensor<string, []>("aw_2815_cast_fp16")];
            tensor<string, []> var_33513_equation_0 = const()[name = tensor<string, []>("op_33513_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33513_cast_fp16 = einsum(equation = var_33513_equation_0, values = (var_33355_cast_fp16, var_33272_cast_fp16))[name = tensor<string, []>("op_33513_cast_fp16")];
            tensor<fp16, []> var_33514_to_fp16 = const()[name = tensor<string, []>("op_33514_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2817_cast_fp16 = mul(x = var_33513_cast_fp16, y = var_33514_to_fp16)[name = tensor<string, []>("aw_2817_cast_fp16")];
            tensor<string, []> var_33517_equation_0 = const()[name = tensor<string, []>("op_33517_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33517_cast_fp16 = einsum(equation = var_33517_equation_0, values = (var_33359_cast_fp16, var_33276_cast_fp16))[name = tensor<string, []>("op_33517_cast_fp16")];
            tensor<fp16, []> var_33518_to_fp16 = const()[name = tensor<string, []>("op_33518_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2819_cast_fp16 = mul(x = var_33517_cast_fp16, y = var_33518_to_fp16)[name = tensor<string, []>("aw_2819_cast_fp16")];
            tensor<string, []> var_33521_equation_0 = const()[name = tensor<string, []>("op_33521_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33521_cast_fp16 = einsum(equation = var_33521_equation_0, values = (var_33363_cast_fp16, var_33280_cast_fp16))[name = tensor<string, []>("op_33521_cast_fp16")];
            tensor<fp16, []> var_33522_to_fp16 = const()[name = tensor<string, []>("op_33522_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2821_cast_fp16 = mul(x = var_33521_cast_fp16, y = var_33522_to_fp16)[name = tensor<string, []>("aw_2821_cast_fp16")];
            tensor<string, []> var_33525_equation_0 = const()[name = tensor<string, []>("op_33525_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33525_cast_fp16 = einsum(equation = var_33525_equation_0, values = (var_33367_cast_fp16, var_33284_cast_fp16))[name = tensor<string, []>("op_33525_cast_fp16")];
            tensor<fp16, []> var_33526_to_fp16 = const()[name = tensor<string, []>("op_33526_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2823_cast_fp16 = mul(x = var_33525_cast_fp16, y = var_33526_to_fp16)[name = tensor<string, []>("aw_2823_cast_fp16")];
            tensor<string, []> var_33529_equation_0 = const()[name = tensor<string, []>("op_33529_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33529_cast_fp16 = einsum(equation = var_33529_equation_0, values = (var_33371_cast_fp16, var_33288_cast_fp16))[name = tensor<string, []>("op_33529_cast_fp16")];
            tensor<fp16, []> var_33530_to_fp16 = const()[name = tensor<string, []>("op_33530_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2825_cast_fp16 = mul(x = var_33529_cast_fp16, y = var_33530_to_fp16)[name = tensor<string, []>("aw_2825_cast_fp16")];
            tensor<string, []> var_33533_equation_0 = const()[name = tensor<string, []>("op_33533_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33533_cast_fp16 = einsum(equation = var_33533_equation_0, values = (var_33375_cast_fp16, var_33292_cast_fp16))[name = tensor<string, []>("op_33533_cast_fp16")];
            tensor<fp16, []> var_33534_to_fp16 = const()[name = tensor<string, []>("op_33534_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2827_cast_fp16 = mul(x = var_33533_cast_fp16, y = var_33534_to_fp16)[name = tensor<string, []>("aw_2827_cast_fp16")];
            tensor<string, []> var_33537_equation_0 = const()[name = tensor<string, []>("op_33537_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33537_cast_fp16 = einsum(equation = var_33537_equation_0, values = (var_33379_cast_fp16, var_33296_cast_fp16))[name = tensor<string, []>("op_33537_cast_fp16")];
            tensor<fp16, []> var_33538_to_fp16 = const()[name = tensor<string, []>("op_33538_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2829_cast_fp16 = mul(x = var_33537_cast_fp16, y = var_33538_to_fp16)[name = tensor<string, []>("aw_2829_cast_fp16")];
            tensor<string, []> var_33541_equation_0 = const()[name = tensor<string, []>("op_33541_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33541_cast_fp16 = einsum(equation = var_33541_equation_0, values = (var_33383_cast_fp16, var_33300_cast_fp16))[name = tensor<string, []>("op_33541_cast_fp16")];
            tensor<fp16, []> var_33542_to_fp16 = const()[name = tensor<string, []>("op_33542_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2831_cast_fp16 = mul(x = var_33541_cast_fp16, y = var_33542_to_fp16)[name = tensor<string, []>("aw_2831_cast_fp16")];
            tensor<string, []> var_33545_equation_0 = const()[name = tensor<string, []>("op_33545_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33545_cast_fp16 = einsum(equation = var_33545_equation_0, values = (var_33387_cast_fp16, var_33304_cast_fp16))[name = tensor<string, []>("op_33545_cast_fp16")];
            tensor<fp16, []> var_33546_to_fp16 = const()[name = tensor<string, []>("op_33546_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2833_cast_fp16 = mul(x = var_33545_cast_fp16, y = var_33546_to_fp16)[name = tensor<string, []>("aw_2833_cast_fp16")];
            tensor<string, []> var_33549_equation_0 = const()[name = tensor<string, []>("op_33549_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33549_cast_fp16 = einsum(equation = var_33549_equation_0, values = (var_33391_cast_fp16, var_33308_cast_fp16))[name = tensor<string, []>("op_33549_cast_fp16")];
            tensor<fp16, []> var_33550_to_fp16 = const()[name = tensor<string, []>("op_33550_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2835_cast_fp16 = mul(x = var_33549_cast_fp16, y = var_33550_to_fp16)[name = tensor<string, []>("aw_2835_cast_fp16")];
            tensor<string, []> var_33553_equation_0 = const()[name = tensor<string, []>("op_33553_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33553_cast_fp16 = einsum(equation = var_33553_equation_0, values = (var_33395_cast_fp16, var_33312_cast_fp16))[name = tensor<string, []>("op_33553_cast_fp16")];
            tensor<fp16, []> var_33554_to_fp16 = const()[name = tensor<string, []>("op_33554_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2837_cast_fp16 = mul(x = var_33553_cast_fp16, y = var_33554_to_fp16)[name = tensor<string, []>("aw_2837_cast_fp16")];
            tensor<string, []> var_33557_equation_0 = const()[name = tensor<string, []>("op_33557_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33557_cast_fp16 = einsum(equation = var_33557_equation_0, values = (var_33399_cast_fp16, var_33316_cast_fp16))[name = tensor<string, []>("op_33557_cast_fp16")];
            tensor<fp16, []> var_33558_to_fp16 = const()[name = tensor<string, []>("op_33558_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2839_cast_fp16 = mul(x = var_33557_cast_fp16, y = var_33558_to_fp16)[name = tensor<string, []>("aw_2839_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33560_cast_fp16 = softmax(axis = var_30355, x = aw_2801_cast_fp16)[name = tensor<string, []>("op_33560_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33561_cast_fp16 = softmax(axis = var_30355, x = aw_2803_cast_fp16)[name = tensor<string, []>("op_33561_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33562_cast_fp16 = softmax(axis = var_30355, x = aw_2805_cast_fp16)[name = tensor<string, []>("op_33562_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33563_cast_fp16 = softmax(axis = var_30355, x = aw_2807_cast_fp16)[name = tensor<string, []>("op_33563_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33564_cast_fp16 = softmax(axis = var_30355, x = aw_2809_cast_fp16)[name = tensor<string, []>("op_33564_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33565_cast_fp16 = softmax(axis = var_30355, x = aw_2811_cast_fp16)[name = tensor<string, []>("op_33565_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33566_cast_fp16 = softmax(axis = var_30355, x = aw_2813_cast_fp16)[name = tensor<string, []>("op_33566_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33567_cast_fp16 = softmax(axis = var_30355, x = aw_2815_cast_fp16)[name = tensor<string, []>("op_33567_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33568_cast_fp16 = softmax(axis = var_30355, x = aw_2817_cast_fp16)[name = tensor<string, []>("op_33568_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33569_cast_fp16 = softmax(axis = var_30355, x = aw_2819_cast_fp16)[name = tensor<string, []>("op_33569_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33570_cast_fp16 = softmax(axis = var_30355, x = aw_2821_cast_fp16)[name = tensor<string, []>("op_33570_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33571_cast_fp16 = softmax(axis = var_30355, x = aw_2823_cast_fp16)[name = tensor<string, []>("op_33571_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33572_cast_fp16 = softmax(axis = var_30355, x = aw_2825_cast_fp16)[name = tensor<string, []>("op_33572_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33573_cast_fp16 = softmax(axis = var_30355, x = aw_2827_cast_fp16)[name = tensor<string, []>("op_33573_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33574_cast_fp16 = softmax(axis = var_30355, x = aw_2829_cast_fp16)[name = tensor<string, []>("op_33574_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33575_cast_fp16 = softmax(axis = var_30355, x = aw_2831_cast_fp16)[name = tensor<string, []>("op_33575_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33576_cast_fp16 = softmax(axis = var_30355, x = aw_2833_cast_fp16)[name = tensor<string, []>("op_33576_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33577_cast_fp16 = softmax(axis = var_30355, x = aw_2835_cast_fp16)[name = tensor<string, []>("op_33577_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33578_cast_fp16 = softmax(axis = var_30355, x = aw_2837_cast_fp16)[name = tensor<string, []>("op_33578_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33579_cast_fp16 = softmax(axis = var_30355, x = aw_2839_cast_fp16)[name = tensor<string, []>("op_33579_cast_fp16")];
            tensor<string, []> var_33581_equation_0 = const()[name = tensor<string, []>("op_33581_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33581_cast_fp16 = einsum(equation = var_33581_equation_0, values = (var_33401_cast_fp16, var_33560_cast_fp16))[name = tensor<string, []>("op_33581_cast_fp16")];
            tensor<string, []> var_33583_equation_0 = const()[name = tensor<string, []>("op_33583_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33583_cast_fp16 = einsum(equation = var_33583_equation_0, values = (var_33405_cast_fp16, var_33561_cast_fp16))[name = tensor<string, []>("op_33583_cast_fp16")];
            tensor<string, []> var_33585_equation_0 = const()[name = tensor<string, []>("op_33585_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33585_cast_fp16 = einsum(equation = var_33585_equation_0, values = (var_33409_cast_fp16, var_33562_cast_fp16))[name = tensor<string, []>("op_33585_cast_fp16")];
            tensor<string, []> var_33587_equation_0 = const()[name = tensor<string, []>("op_33587_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33587_cast_fp16 = einsum(equation = var_33587_equation_0, values = (var_33413_cast_fp16, var_33563_cast_fp16))[name = tensor<string, []>("op_33587_cast_fp16")];
            tensor<string, []> var_33589_equation_0 = const()[name = tensor<string, []>("op_33589_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33589_cast_fp16 = einsum(equation = var_33589_equation_0, values = (var_33417_cast_fp16, var_33564_cast_fp16))[name = tensor<string, []>("op_33589_cast_fp16")];
            tensor<string, []> var_33591_equation_0 = const()[name = tensor<string, []>("op_33591_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33591_cast_fp16 = einsum(equation = var_33591_equation_0, values = (var_33421_cast_fp16, var_33565_cast_fp16))[name = tensor<string, []>("op_33591_cast_fp16")];
            tensor<string, []> var_33593_equation_0 = const()[name = tensor<string, []>("op_33593_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33593_cast_fp16 = einsum(equation = var_33593_equation_0, values = (var_33425_cast_fp16, var_33566_cast_fp16))[name = tensor<string, []>("op_33593_cast_fp16")];
            tensor<string, []> var_33595_equation_0 = const()[name = tensor<string, []>("op_33595_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33595_cast_fp16 = einsum(equation = var_33595_equation_0, values = (var_33429_cast_fp16, var_33567_cast_fp16))[name = tensor<string, []>("op_33595_cast_fp16")];
            tensor<string, []> var_33597_equation_0 = const()[name = tensor<string, []>("op_33597_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33597_cast_fp16 = einsum(equation = var_33597_equation_0, values = (var_33433_cast_fp16, var_33568_cast_fp16))[name = tensor<string, []>("op_33597_cast_fp16")];
            tensor<string, []> var_33599_equation_0 = const()[name = tensor<string, []>("op_33599_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33599_cast_fp16 = einsum(equation = var_33599_equation_0, values = (var_33437_cast_fp16, var_33569_cast_fp16))[name = tensor<string, []>("op_33599_cast_fp16")];
            tensor<string, []> var_33601_equation_0 = const()[name = tensor<string, []>("op_33601_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33601_cast_fp16 = einsum(equation = var_33601_equation_0, values = (var_33441_cast_fp16, var_33570_cast_fp16))[name = tensor<string, []>("op_33601_cast_fp16")];
            tensor<string, []> var_33603_equation_0 = const()[name = tensor<string, []>("op_33603_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33603_cast_fp16 = einsum(equation = var_33603_equation_0, values = (var_33445_cast_fp16, var_33571_cast_fp16))[name = tensor<string, []>("op_33603_cast_fp16")];
            tensor<string, []> var_33605_equation_0 = const()[name = tensor<string, []>("op_33605_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33605_cast_fp16 = einsum(equation = var_33605_equation_0, values = (var_33449_cast_fp16, var_33572_cast_fp16))[name = tensor<string, []>("op_33605_cast_fp16")];
            tensor<string, []> var_33607_equation_0 = const()[name = tensor<string, []>("op_33607_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33607_cast_fp16 = einsum(equation = var_33607_equation_0, values = (var_33453_cast_fp16, var_33573_cast_fp16))[name = tensor<string, []>("op_33607_cast_fp16")];
            tensor<string, []> var_33609_equation_0 = const()[name = tensor<string, []>("op_33609_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33609_cast_fp16 = einsum(equation = var_33609_equation_0, values = (var_33457_cast_fp16, var_33574_cast_fp16))[name = tensor<string, []>("op_33609_cast_fp16")];
            tensor<string, []> var_33611_equation_0 = const()[name = tensor<string, []>("op_33611_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33611_cast_fp16 = einsum(equation = var_33611_equation_0, values = (var_33461_cast_fp16, var_33575_cast_fp16))[name = tensor<string, []>("op_33611_cast_fp16")];
            tensor<string, []> var_33613_equation_0 = const()[name = tensor<string, []>("op_33613_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33613_cast_fp16 = einsum(equation = var_33613_equation_0, values = (var_33465_cast_fp16, var_33576_cast_fp16))[name = tensor<string, []>("op_33613_cast_fp16")];
            tensor<string, []> var_33615_equation_0 = const()[name = tensor<string, []>("op_33615_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33615_cast_fp16 = einsum(equation = var_33615_equation_0, values = (var_33469_cast_fp16, var_33577_cast_fp16))[name = tensor<string, []>("op_33615_cast_fp16")];
            tensor<string, []> var_33617_equation_0 = const()[name = tensor<string, []>("op_33617_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33617_cast_fp16 = einsum(equation = var_33617_equation_0, values = (var_33473_cast_fp16, var_33578_cast_fp16))[name = tensor<string, []>("op_33617_cast_fp16")];
            tensor<string, []> var_33619_equation_0 = const()[name = tensor<string, []>("op_33619_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33619_cast_fp16 = einsum(equation = var_33619_equation_0, values = (var_33477_cast_fp16, var_33579_cast_fp16))[name = tensor<string, []>("op_33619_cast_fp16")];
            tensor<bool, []> input_463_interleave_0 = const()[name = tensor<string, []>("input_463_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_463_cast_fp16 = concat(axis = var_30355, interleave = input_463_interleave_0, values = (var_33581_cast_fp16, var_33583_cast_fp16, var_33585_cast_fp16, var_33587_cast_fp16, var_33589_cast_fp16, var_33591_cast_fp16, var_33593_cast_fp16, var_33595_cast_fp16, var_33597_cast_fp16, var_33599_cast_fp16, var_33601_cast_fp16, var_33603_cast_fp16, var_33605_cast_fp16, var_33607_cast_fp16, var_33609_cast_fp16, var_33611_cast_fp16, var_33613_cast_fp16, var_33615_cast_fp16, var_33617_cast_fp16, var_33619_cast_fp16))[name = tensor<string, []>("input_463_cast_fp16")];
            tensor<string, []> var_33629_pad_type_0 = const()[name = tensor<string, []>("op_33629_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_33629_strides_0 = const()[name = tensor<string, []>("op_33629_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_33629_pad_0 = const()[name = tensor<string, []>("op_33629_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_33629_dilations_0 = const()[name = tensor<string, []>("op_33629_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_33629_groups_0 = const()[name = tensor<string, []>("op_33629_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_3_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(85701440))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(86930304))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_3_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_3_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_3_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(86930496)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_33629_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_3_attn1_to_out_0_bias_to_fp16, dilations = var_33629_dilations_0, groups = var_33629_groups_0, pad = var_33629_pad_0, pad_type = var_33629_pad_type_0, strides = var_33629_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_3_attn1_to_out_0_weight_to_fp16_palettized, x = input_463_cast_fp16)[name = tensor<string, []>("op_33629_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_225_cast_fp16 = add(x = var_33629_cast_fp16, y = inputs_223_cast_fp16)[name = tensor<string, []>("inputs_225_cast_fp16")];
            tensor<int32, [1]> hidden_states_307_axes_0 = const()[name = tensor<string, []>("hidden_states_307_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_307_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_307_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(86933120)))];
            tensor<fp16, [1280]> hidden_states_307_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_307_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(86935744)))];
            tensor<fp16, []> var_33639_to_fp16 = const()[name = tensor<string, []>("op_33639_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_307_cast_fp16 = layer_norm(axes = hidden_states_307_axes_0, beta = hidden_states_307_beta_0_to_fp16, epsilon = var_33639_to_fp16, gamma = hidden_states_307_gamma_0_to_fp16, x = inputs_225_cast_fp16)[name = tensor<string, []>("hidden_states_307_cast_fp16")];
            tensor<string, []> q_151_pad_type_0 = const()[name = tensor<string, []>("q_151_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_151_strides_0 = const()[name = tensor<string, []>("q_151_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_151_pad_0 = const()[name = tensor<string, []>("q_151_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_151_dilations_0 = const()[name = tensor<string, []>("q_151_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_151_groups_0 = const()[name = tensor<string, []>("q_151_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_3_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(86938368))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(88167232))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_3_attn2_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_151_cast_fp16 = conv(dilations = q_151_dilations_0, groups = q_151_groups_0, pad = q_151_pad_0, pad_type = q_151_pad_type_0, strides = q_151_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_3_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_307_cast_fp16)[name = tensor<string, []>("q_151_cast_fp16")];
            tensor<string, []> k_301_pad_type_0 = const()[name = tensor<string, []>("k_301_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_301_strides_0 = const()[name = tensor<string, []>("k_301_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_301_pad_0 = const()[name = tensor<string, []>("k_301_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_301_dilations_0 = const()[name = tensor<string, []>("k_301_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_301_groups_0 = const()[name = tensor<string, []>("k_301_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_3_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(88167424))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(90133568))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_3_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> k_301_cast_fp16 = conv(dilations = k_301_dilations_0, groups = k_301_groups_0, pad = k_301_pad_0, pad_type = k_301_pad_type_0, strides = k_301_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_3_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("k_301_cast_fp16")];
            tensor<string, []> v_151_pad_type_0 = const()[name = tensor<string, []>("v_151_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_151_strides_0 = const()[name = tensor<string, []>("v_151_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_151_pad_0 = const()[name = tensor<string, []>("v_151_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_151_dilations_0 = const()[name = tensor<string, []>("v_151_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_151_groups_0 = const()[name = tensor<string, []>("v_151_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_3_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(90133760))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92099904))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_3_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> v_151_cast_fp16 = conv(dilations = v_151_dilations_0, groups = v_151_groups_0, pad = v_151_pad_0, pad_type = v_151_pad_type_0, strides = v_151_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_3_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("v_151_cast_fp16")];
            tensor<int32, [4]> var_33672_begin_0 = const()[name = tensor<string, []>("op_33672_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_33672_end_0 = const()[name = tensor<string, []>("op_33672_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_33672_end_mask_0 = const()[name = tensor<string, []>("op_33672_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33672_cast_fp16 = slice_by_index(begin = var_33672_begin_0, end = var_33672_end_0, end_mask = var_33672_end_mask_0, x = q_151_cast_fp16)[name = tensor<string, []>("op_33672_cast_fp16")];
            tensor<int32, [4]> var_33676_begin_0 = const()[name = tensor<string, []>("op_33676_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_33676_end_0 = const()[name = tensor<string, []>("op_33676_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_33676_end_mask_0 = const()[name = tensor<string, []>("op_33676_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33676_cast_fp16 = slice_by_index(begin = var_33676_begin_0, end = var_33676_end_0, end_mask = var_33676_end_mask_0, x = q_151_cast_fp16)[name = tensor<string, []>("op_33676_cast_fp16")];
            tensor<int32, [4]> var_33680_begin_0 = const()[name = tensor<string, []>("op_33680_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_33680_end_0 = const()[name = tensor<string, []>("op_33680_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_33680_end_mask_0 = const()[name = tensor<string, []>("op_33680_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33680_cast_fp16 = slice_by_index(begin = var_33680_begin_0, end = var_33680_end_0, end_mask = var_33680_end_mask_0, x = q_151_cast_fp16)[name = tensor<string, []>("op_33680_cast_fp16")];
            tensor<int32, [4]> var_33684_begin_0 = const()[name = tensor<string, []>("op_33684_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_33684_end_0 = const()[name = tensor<string, []>("op_33684_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_33684_end_mask_0 = const()[name = tensor<string, []>("op_33684_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33684_cast_fp16 = slice_by_index(begin = var_33684_begin_0, end = var_33684_end_0, end_mask = var_33684_end_mask_0, x = q_151_cast_fp16)[name = tensor<string, []>("op_33684_cast_fp16")];
            tensor<int32, [4]> var_33688_begin_0 = const()[name = tensor<string, []>("op_33688_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_33688_end_0 = const()[name = tensor<string, []>("op_33688_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_33688_end_mask_0 = const()[name = tensor<string, []>("op_33688_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33688_cast_fp16 = slice_by_index(begin = var_33688_begin_0, end = var_33688_end_0, end_mask = var_33688_end_mask_0, x = q_151_cast_fp16)[name = tensor<string, []>("op_33688_cast_fp16")];
            tensor<int32, [4]> var_33692_begin_0 = const()[name = tensor<string, []>("op_33692_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_33692_end_0 = const()[name = tensor<string, []>("op_33692_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_33692_end_mask_0 = const()[name = tensor<string, []>("op_33692_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33692_cast_fp16 = slice_by_index(begin = var_33692_begin_0, end = var_33692_end_0, end_mask = var_33692_end_mask_0, x = q_151_cast_fp16)[name = tensor<string, []>("op_33692_cast_fp16")];
            tensor<int32, [4]> var_33696_begin_0 = const()[name = tensor<string, []>("op_33696_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_33696_end_0 = const()[name = tensor<string, []>("op_33696_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_33696_end_mask_0 = const()[name = tensor<string, []>("op_33696_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33696_cast_fp16 = slice_by_index(begin = var_33696_begin_0, end = var_33696_end_0, end_mask = var_33696_end_mask_0, x = q_151_cast_fp16)[name = tensor<string, []>("op_33696_cast_fp16")];
            tensor<int32, [4]> var_33700_begin_0 = const()[name = tensor<string, []>("op_33700_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_33700_end_0 = const()[name = tensor<string, []>("op_33700_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_33700_end_mask_0 = const()[name = tensor<string, []>("op_33700_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33700_cast_fp16 = slice_by_index(begin = var_33700_begin_0, end = var_33700_end_0, end_mask = var_33700_end_mask_0, x = q_151_cast_fp16)[name = tensor<string, []>("op_33700_cast_fp16")];
            tensor<int32, [4]> var_33704_begin_0 = const()[name = tensor<string, []>("op_33704_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_33704_end_0 = const()[name = tensor<string, []>("op_33704_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_33704_end_mask_0 = const()[name = tensor<string, []>("op_33704_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33704_cast_fp16 = slice_by_index(begin = var_33704_begin_0, end = var_33704_end_0, end_mask = var_33704_end_mask_0, x = q_151_cast_fp16)[name = tensor<string, []>("op_33704_cast_fp16")];
            tensor<int32, [4]> var_33708_begin_0 = const()[name = tensor<string, []>("op_33708_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_33708_end_0 = const()[name = tensor<string, []>("op_33708_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_33708_end_mask_0 = const()[name = tensor<string, []>("op_33708_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33708_cast_fp16 = slice_by_index(begin = var_33708_begin_0, end = var_33708_end_0, end_mask = var_33708_end_mask_0, x = q_151_cast_fp16)[name = tensor<string, []>("op_33708_cast_fp16")];
            tensor<int32, [4]> var_33712_begin_0 = const()[name = tensor<string, []>("op_33712_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_33712_end_0 = const()[name = tensor<string, []>("op_33712_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_33712_end_mask_0 = const()[name = tensor<string, []>("op_33712_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33712_cast_fp16 = slice_by_index(begin = var_33712_begin_0, end = var_33712_end_0, end_mask = var_33712_end_mask_0, x = q_151_cast_fp16)[name = tensor<string, []>("op_33712_cast_fp16")];
            tensor<int32, [4]> var_33716_begin_0 = const()[name = tensor<string, []>("op_33716_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_33716_end_0 = const()[name = tensor<string, []>("op_33716_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_33716_end_mask_0 = const()[name = tensor<string, []>("op_33716_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33716_cast_fp16 = slice_by_index(begin = var_33716_begin_0, end = var_33716_end_0, end_mask = var_33716_end_mask_0, x = q_151_cast_fp16)[name = tensor<string, []>("op_33716_cast_fp16")];
            tensor<int32, [4]> var_33720_begin_0 = const()[name = tensor<string, []>("op_33720_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_33720_end_0 = const()[name = tensor<string, []>("op_33720_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_33720_end_mask_0 = const()[name = tensor<string, []>("op_33720_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33720_cast_fp16 = slice_by_index(begin = var_33720_begin_0, end = var_33720_end_0, end_mask = var_33720_end_mask_0, x = q_151_cast_fp16)[name = tensor<string, []>("op_33720_cast_fp16")];
            tensor<int32, [4]> var_33724_begin_0 = const()[name = tensor<string, []>("op_33724_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_33724_end_0 = const()[name = tensor<string, []>("op_33724_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_33724_end_mask_0 = const()[name = tensor<string, []>("op_33724_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33724_cast_fp16 = slice_by_index(begin = var_33724_begin_0, end = var_33724_end_0, end_mask = var_33724_end_mask_0, x = q_151_cast_fp16)[name = tensor<string, []>("op_33724_cast_fp16")];
            tensor<int32, [4]> var_33728_begin_0 = const()[name = tensor<string, []>("op_33728_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_33728_end_0 = const()[name = tensor<string, []>("op_33728_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_33728_end_mask_0 = const()[name = tensor<string, []>("op_33728_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33728_cast_fp16 = slice_by_index(begin = var_33728_begin_0, end = var_33728_end_0, end_mask = var_33728_end_mask_0, x = q_151_cast_fp16)[name = tensor<string, []>("op_33728_cast_fp16")];
            tensor<int32, [4]> var_33732_begin_0 = const()[name = tensor<string, []>("op_33732_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_33732_end_0 = const()[name = tensor<string, []>("op_33732_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_33732_end_mask_0 = const()[name = tensor<string, []>("op_33732_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33732_cast_fp16 = slice_by_index(begin = var_33732_begin_0, end = var_33732_end_0, end_mask = var_33732_end_mask_0, x = q_151_cast_fp16)[name = tensor<string, []>("op_33732_cast_fp16")];
            tensor<int32, [4]> var_33736_begin_0 = const()[name = tensor<string, []>("op_33736_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_33736_end_0 = const()[name = tensor<string, []>("op_33736_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_33736_end_mask_0 = const()[name = tensor<string, []>("op_33736_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33736_cast_fp16 = slice_by_index(begin = var_33736_begin_0, end = var_33736_end_0, end_mask = var_33736_end_mask_0, x = q_151_cast_fp16)[name = tensor<string, []>("op_33736_cast_fp16")];
            tensor<int32, [4]> var_33740_begin_0 = const()[name = tensor<string, []>("op_33740_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_33740_end_0 = const()[name = tensor<string, []>("op_33740_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_33740_end_mask_0 = const()[name = tensor<string, []>("op_33740_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33740_cast_fp16 = slice_by_index(begin = var_33740_begin_0, end = var_33740_end_0, end_mask = var_33740_end_mask_0, x = q_151_cast_fp16)[name = tensor<string, []>("op_33740_cast_fp16")];
            tensor<int32, [4]> var_33744_begin_0 = const()[name = tensor<string, []>("op_33744_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_33744_end_0 = const()[name = tensor<string, []>("op_33744_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_33744_end_mask_0 = const()[name = tensor<string, []>("op_33744_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33744_cast_fp16 = slice_by_index(begin = var_33744_begin_0, end = var_33744_end_0, end_mask = var_33744_end_mask_0, x = q_151_cast_fp16)[name = tensor<string, []>("op_33744_cast_fp16")];
            tensor<int32, [4]> var_33748_begin_0 = const()[name = tensor<string, []>("op_33748_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_33748_end_0 = const()[name = tensor<string, []>("op_33748_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_33748_end_mask_0 = const()[name = tensor<string, []>("op_33748_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33748_cast_fp16 = slice_by_index(begin = var_33748_begin_0, end = var_33748_end_0, end_mask = var_33748_end_mask_0, x = q_151_cast_fp16)[name = tensor<string, []>("op_33748_cast_fp16")];
            tensor<int32, [4]> k_303_perm_0 = const()[name = tensor<string, []>("k_303_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_33755_begin_0 = const()[name = tensor<string, []>("op_33755_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_33755_end_0 = const()[name = tensor<string, []>("op_33755_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_33755_end_mask_0 = const()[name = tensor<string, []>("op_33755_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> k_303_cast_fp16 = transpose(perm = k_303_perm_0, x = k_301_cast_fp16)[name = tensor<string, []>("transpose_64")];
            tensor<fp16, [2, 77, 1, 64]> var_33755_cast_fp16 = slice_by_index(begin = var_33755_begin_0, end = var_33755_end_0, end_mask = var_33755_end_mask_0, x = k_303_cast_fp16)[name = tensor<string, []>("op_33755_cast_fp16")];
            tensor<int32, [4]> var_33759_begin_0 = const()[name = tensor<string, []>("op_33759_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_33759_end_0 = const()[name = tensor<string, []>("op_33759_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_33759_end_mask_0 = const()[name = tensor<string, []>("op_33759_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_33759_cast_fp16 = slice_by_index(begin = var_33759_begin_0, end = var_33759_end_0, end_mask = var_33759_end_mask_0, x = k_303_cast_fp16)[name = tensor<string, []>("op_33759_cast_fp16")];
            tensor<int32, [4]> var_33763_begin_0 = const()[name = tensor<string, []>("op_33763_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_33763_end_0 = const()[name = tensor<string, []>("op_33763_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_33763_end_mask_0 = const()[name = tensor<string, []>("op_33763_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_33763_cast_fp16 = slice_by_index(begin = var_33763_begin_0, end = var_33763_end_0, end_mask = var_33763_end_mask_0, x = k_303_cast_fp16)[name = tensor<string, []>("op_33763_cast_fp16")];
            tensor<int32, [4]> var_33767_begin_0 = const()[name = tensor<string, []>("op_33767_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_33767_end_0 = const()[name = tensor<string, []>("op_33767_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_33767_end_mask_0 = const()[name = tensor<string, []>("op_33767_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_33767_cast_fp16 = slice_by_index(begin = var_33767_begin_0, end = var_33767_end_0, end_mask = var_33767_end_mask_0, x = k_303_cast_fp16)[name = tensor<string, []>("op_33767_cast_fp16")];
            tensor<int32, [4]> var_33771_begin_0 = const()[name = tensor<string, []>("op_33771_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_33771_end_0 = const()[name = tensor<string, []>("op_33771_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_33771_end_mask_0 = const()[name = tensor<string, []>("op_33771_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_33771_cast_fp16 = slice_by_index(begin = var_33771_begin_0, end = var_33771_end_0, end_mask = var_33771_end_mask_0, x = k_303_cast_fp16)[name = tensor<string, []>("op_33771_cast_fp16")];
            tensor<int32, [4]> var_33775_begin_0 = const()[name = tensor<string, []>("op_33775_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_33775_end_0 = const()[name = tensor<string, []>("op_33775_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_33775_end_mask_0 = const()[name = tensor<string, []>("op_33775_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_33775_cast_fp16 = slice_by_index(begin = var_33775_begin_0, end = var_33775_end_0, end_mask = var_33775_end_mask_0, x = k_303_cast_fp16)[name = tensor<string, []>("op_33775_cast_fp16")];
            tensor<int32, [4]> var_33779_begin_0 = const()[name = tensor<string, []>("op_33779_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_33779_end_0 = const()[name = tensor<string, []>("op_33779_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_33779_end_mask_0 = const()[name = tensor<string, []>("op_33779_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_33779_cast_fp16 = slice_by_index(begin = var_33779_begin_0, end = var_33779_end_0, end_mask = var_33779_end_mask_0, x = k_303_cast_fp16)[name = tensor<string, []>("op_33779_cast_fp16")];
            tensor<int32, [4]> var_33783_begin_0 = const()[name = tensor<string, []>("op_33783_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_33783_end_0 = const()[name = tensor<string, []>("op_33783_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_33783_end_mask_0 = const()[name = tensor<string, []>("op_33783_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_33783_cast_fp16 = slice_by_index(begin = var_33783_begin_0, end = var_33783_end_0, end_mask = var_33783_end_mask_0, x = k_303_cast_fp16)[name = tensor<string, []>("op_33783_cast_fp16")];
            tensor<int32, [4]> var_33787_begin_0 = const()[name = tensor<string, []>("op_33787_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_33787_end_0 = const()[name = tensor<string, []>("op_33787_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_33787_end_mask_0 = const()[name = tensor<string, []>("op_33787_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_33787_cast_fp16 = slice_by_index(begin = var_33787_begin_0, end = var_33787_end_0, end_mask = var_33787_end_mask_0, x = k_303_cast_fp16)[name = tensor<string, []>("op_33787_cast_fp16")];
            tensor<int32, [4]> var_33791_begin_0 = const()[name = tensor<string, []>("op_33791_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_33791_end_0 = const()[name = tensor<string, []>("op_33791_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_33791_end_mask_0 = const()[name = tensor<string, []>("op_33791_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_33791_cast_fp16 = slice_by_index(begin = var_33791_begin_0, end = var_33791_end_0, end_mask = var_33791_end_mask_0, x = k_303_cast_fp16)[name = tensor<string, []>("op_33791_cast_fp16")];
            tensor<int32, [4]> var_33795_begin_0 = const()[name = tensor<string, []>("op_33795_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_33795_end_0 = const()[name = tensor<string, []>("op_33795_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_33795_end_mask_0 = const()[name = tensor<string, []>("op_33795_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_33795_cast_fp16 = slice_by_index(begin = var_33795_begin_0, end = var_33795_end_0, end_mask = var_33795_end_mask_0, x = k_303_cast_fp16)[name = tensor<string, []>("op_33795_cast_fp16")];
            tensor<int32, [4]> var_33799_begin_0 = const()[name = tensor<string, []>("op_33799_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_33799_end_0 = const()[name = tensor<string, []>("op_33799_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_33799_end_mask_0 = const()[name = tensor<string, []>("op_33799_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_33799_cast_fp16 = slice_by_index(begin = var_33799_begin_0, end = var_33799_end_0, end_mask = var_33799_end_mask_0, x = k_303_cast_fp16)[name = tensor<string, []>("op_33799_cast_fp16")];
            tensor<int32, [4]> var_33803_begin_0 = const()[name = tensor<string, []>("op_33803_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_33803_end_0 = const()[name = tensor<string, []>("op_33803_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_33803_end_mask_0 = const()[name = tensor<string, []>("op_33803_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_33803_cast_fp16 = slice_by_index(begin = var_33803_begin_0, end = var_33803_end_0, end_mask = var_33803_end_mask_0, x = k_303_cast_fp16)[name = tensor<string, []>("op_33803_cast_fp16")];
            tensor<int32, [4]> var_33807_begin_0 = const()[name = tensor<string, []>("op_33807_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_33807_end_0 = const()[name = tensor<string, []>("op_33807_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_33807_end_mask_0 = const()[name = tensor<string, []>("op_33807_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_33807_cast_fp16 = slice_by_index(begin = var_33807_begin_0, end = var_33807_end_0, end_mask = var_33807_end_mask_0, x = k_303_cast_fp16)[name = tensor<string, []>("op_33807_cast_fp16")];
            tensor<int32, [4]> var_33811_begin_0 = const()[name = tensor<string, []>("op_33811_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_33811_end_0 = const()[name = tensor<string, []>("op_33811_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_33811_end_mask_0 = const()[name = tensor<string, []>("op_33811_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_33811_cast_fp16 = slice_by_index(begin = var_33811_begin_0, end = var_33811_end_0, end_mask = var_33811_end_mask_0, x = k_303_cast_fp16)[name = tensor<string, []>("op_33811_cast_fp16")];
            tensor<int32, [4]> var_33815_begin_0 = const()[name = tensor<string, []>("op_33815_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_33815_end_0 = const()[name = tensor<string, []>("op_33815_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_33815_end_mask_0 = const()[name = tensor<string, []>("op_33815_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_33815_cast_fp16 = slice_by_index(begin = var_33815_begin_0, end = var_33815_end_0, end_mask = var_33815_end_mask_0, x = k_303_cast_fp16)[name = tensor<string, []>("op_33815_cast_fp16")];
            tensor<int32, [4]> var_33819_begin_0 = const()[name = tensor<string, []>("op_33819_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_33819_end_0 = const()[name = tensor<string, []>("op_33819_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_33819_end_mask_0 = const()[name = tensor<string, []>("op_33819_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_33819_cast_fp16 = slice_by_index(begin = var_33819_begin_0, end = var_33819_end_0, end_mask = var_33819_end_mask_0, x = k_303_cast_fp16)[name = tensor<string, []>("op_33819_cast_fp16")];
            tensor<int32, [4]> var_33823_begin_0 = const()[name = tensor<string, []>("op_33823_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_33823_end_0 = const()[name = tensor<string, []>("op_33823_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_33823_end_mask_0 = const()[name = tensor<string, []>("op_33823_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_33823_cast_fp16 = slice_by_index(begin = var_33823_begin_0, end = var_33823_end_0, end_mask = var_33823_end_mask_0, x = k_303_cast_fp16)[name = tensor<string, []>("op_33823_cast_fp16")];
            tensor<int32, [4]> var_33827_begin_0 = const()[name = tensor<string, []>("op_33827_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_33827_end_0 = const()[name = tensor<string, []>("op_33827_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_33827_end_mask_0 = const()[name = tensor<string, []>("op_33827_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_33827_cast_fp16 = slice_by_index(begin = var_33827_begin_0, end = var_33827_end_0, end_mask = var_33827_end_mask_0, x = k_303_cast_fp16)[name = tensor<string, []>("op_33827_cast_fp16")];
            tensor<int32, [4]> var_33831_begin_0 = const()[name = tensor<string, []>("op_33831_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_33831_end_0 = const()[name = tensor<string, []>("op_33831_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_33831_end_mask_0 = const()[name = tensor<string, []>("op_33831_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_33831_cast_fp16 = slice_by_index(begin = var_33831_begin_0, end = var_33831_end_0, end_mask = var_33831_end_mask_0, x = k_303_cast_fp16)[name = tensor<string, []>("op_33831_cast_fp16")];
            tensor<int32, [4]> var_33833_begin_0 = const()[name = tensor<string, []>("op_33833_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_33833_end_0 = const()[name = tensor<string, []>("op_33833_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_33833_end_mask_0 = const()[name = tensor<string, []>("op_33833_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_33833_cast_fp16 = slice_by_index(begin = var_33833_begin_0, end = var_33833_end_0, end_mask = var_33833_end_mask_0, x = v_151_cast_fp16)[name = tensor<string, []>("op_33833_cast_fp16")];
            tensor<int32, [4]> var_33837_begin_0 = const()[name = tensor<string, []>("op_33837_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_33837_end_0 = const()[name = tensor<string, []>("op_33837_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_33837_end_mask_0 = const()[name = tensor<string, []>("op_33837_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_33837_cast_fp16 = slice_by_index(begin = var_33837_begin_0, end = var_33837_end_0, end_mask = var_33837_end_mask_0, x = v_151_cast_fp16)[name = tensor<string, []>("op_33837_cast_fp16")];
            tensor<int32, [4]> var_33841_begin_0 = const()[name = tensor<string, []>("op_33841_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_33841_end_0 = const()[name = tensor<string, []>("op_33841_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_33841_end_mask_0 = const()[name = tensor<string, []>("op_33841_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_33841_cast_fp16 = slice_by_index(begin = var_33841_begin_0, end = var_33841_end_0, end_mask = var_33841_end_mask_0, x = v_151_cast_fp16)[name = tensor<string, []>("op_33841_cast_fp16")];
            tensor<int32, [4]> var_33845_begin_0 = const()[name = tensor<string, []>("op_33845_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_33845_end_0 = const()[name = tensor<string, []>("op_33845_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_33845_end_mask_0 = const()[name = tensor<string, []>("op_33845_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_33845_cast_fp16 = slice_by_index(begin = var_33845_begin_0, end = var_33845_end_0, end_mask = var_33845_end_mask_0, x = v_151_cast_fp16)[name = tensor<string, []>("op_33845_cast_fp16")];
            tensor<int32, [4]> var_33849_begin_0 = const()[name = tensor<string, []>("op_33849_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_33849_end_0 = const()[name = tensor<string, []>("op_33849_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_33849_end_mask_0 = const()[name = tensor<string, []>("op_33849_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_33849_cast_fp16 = slice_by_index(begin = var_33849_begin_0, end = var_33849_end_0, end_mask = var_33849_end_mask_0, x = v_151_cast_fp16)[name = tensor<string, []>("op_33849_cast_fp16")];
            tensor<int32, [4]> var_33853_begin_0 = const()[name = tensor<string, []>("op_33853_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_33853_end_0 = const()[name = tensor<string, []>("op_33853_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_33853_end_mask_0 = const()[name = tensor<string, []>("op_33853_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_33853_cast_fp16 = slice_by_index(begin = var_33853_begin_0, end = var_33853_end_0, end_mask = var_33853_end_mask_0, x = v_151_cast_fp16)[name = tensor<string, []>("op_33853_cast_fp16")];
            tensor<int32, [4]> var_33857_begin_0 = const()[name = tensor<string, []>("op_33857_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_33857_end_0 = const()[name = tensor<string, []>("op_33857_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_33857_end_mask_0 = const()[name = tensor<string, []>("op_33857_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_33857_cast_fp16 = slice_by_index(begin = var_33857_begin_0, end = var_33857_end_0, end_mask = var_33857_end_mask_0, x = v_151_cast_fp16)[name = tensor<string, []>("op_33857_cast_fp16")];
            tensor<int32, [4]> var_33861_begin_0 = const()[name = tensor<string, []>("op_33861_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_33861_end_0 = const()[name = tensor<string, []>("op_33861_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_33861_end_mask_0 = const()[name = tensor<string, []>("op_33861_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_33861_cast_fp16 = slice_by_index(begin = var_33861_begin_0, end = var_33861_end_0, end_mask = var_33861_end_mask_0, x = v_151_cast_fp16)[name = tensor<string, []>("op_33861_cast_fp16")];
            tensor<int32, [4]> var_33865_begin_0 = const()[name = tensor<string, []>("op_33865_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_33865_end_0 = const()[name = tensor<string, []>("op_33865_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_33865_end_mask_0 = const()[name = tensor<string, []>("op_33865_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_33865_cast_fp16 = slice_by_index(begin = var_33865_begin_0, end = var_33865_end_0, end_mask = var_33865_end_mask_0, x = v_151_cast_fp16)[name = tensor<string, []>("op_33865_cast_fp16")];
            tensor<int32, [4]> var_33869_begin_0 = const()[name = tensor<string, []>("op_33869_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_33869_end_0 = const()[name = tensor<string, []>("op_33869_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_33869_end_mask_0 = const()[name = tensor<string, []>("op_33869_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_33869_cast_fp16 = slice_by_index(begin = var_33869_begin_0, end = var_33869_end_0, end_mask = var_33869_end_mask_0, x = v_151_cast_fp16)[name = tensor<string, []>("op_33869_cast_fp16")];
            tensor<int32, [4]> var_33873_begin_0 = const()[name = tensor<string, []>("op_33873_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_33873_end_0 = const()[name = tensor<string, []>("op_33873_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_33873_end_mask_0 = const()[name = tensor<string, []>("op_33873_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_33873_cast_fp16 = slice_by_index(begin = var_33873_begin_0, end = var_33873_end_0, end_mask = var_33873_end_mask_0, x = v_151_cast_fp16)[name = tensor<string, []>("op_33873_cast_fp16")];
            tensor<int32, [4]> var_33877_begin_0 = const()[name = tensor<string, []>("op_33877_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_33877_end_0 = const()[name = tensor<string, []>("op_33877_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_33877_end_mask_0 = const()[name = tensor<string, []>("op_33877_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_33877_cast_fp16 = slice_by_index(begin = var_33877_begin_0, end = var_33877_end_0, end_mask = var_33877_end_mask_0, x = v_151_cast_fp16)[name = tensor<string, []>("op_33877_cast_fp16")];
            tensor<int32, [4]> var_33881_begin_0 = const()[name = tensor<string, []>("op_33881_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_33881_end_0 = const()[name = tensor<string, []>("op_33881_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_33881_end_mask_0 = const()[name = tensor<string, []>("op_33881_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_33881_cast_fp16 = slice_by_index(begin = var_33881_begin_0, end = var_33881_end_0, end_mask = var_33881_end_mask_0, x = v_151_cast_fp16)[name = tensor<string, []>("op_33881_cast_fp16")];
            tensor<int32, [4]> var_33885_begin_0 = const()[name = tensor<string, []>("op_33885_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_33885_end_0 = const()[name = tensor<string, []>("op_33885_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_33885_end_mask_0 = const()[name = tensor<string, []>("op_33885_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_33885_cast_fp16 = slice_by_index(begin = var_33885_begin_0, end = var_33885_end_0, end_mask = var_33885_end_mask_0, x = v_151_cast_fp16)[name = tensor<string, []>("op_33885_cast_fp16")];
            tensor<int32, [4]> var_33889_begin_0 = const()[name = tensor<string, []>("op_33889_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_33889_end_0 = const()[name = tensor<string, []>("op_33889_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_33889_end_mask_0 = const()[name = tensor<string, []>("op_33889_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_33889_cast_fp16 = slice_by_index(begin = var_33889_begin_0, end = var_33889_end_0, end_mask = var_33889_end_mask_0, x = v_151_cast_fp16)[name = tensor<string, []>("op_33889_cast_fp16")];
            tensor<int32, [4]> var_33893_begin_0 = const()[name = tensor<string, []>("op_33893_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_33893_end_0 = const()[name = tensor<string, []>("op_33893_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_33893_end_mask_0 = const()[name = tensor<string, []>("op_33893_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_33893_cast_fp16 = slice_by_index(begin = var_33893_begin_0, end = var_33893_end_0, end_mask = var_33893_end_mask_0, x = v_151_cast_fp16)[name = tensor<string, []>("op_33893_cast_fp16")];
            tensor<int32, [4]> var_33897_begin_0 = const()[name = tensor<string, []>("op_33897_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_33897_end_0 = const()[name = tensor<string, []>("op_33897_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_33897_end_mask_0 = const()[name = tensor<string, []>("op_33897_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_33897_cast_fp16 = slice_by_index(begin = var_33897_begin_0, end = var_33897_end_0, end_mask = var_33897_end_mask_0, x = v_151_cast_fp16)[name = tensor<string, []>("op_33897_cast_fp16")];
            tensor<int32, [4]> var_33901_begin_0 = const()[name = tensor<string, []>("op_33901_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_33901_end_0 = const()[name = tensor<string, []>("op_33901_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_33901_end_mask_0 = const()[name = tensor<string, []>("op_33901_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_33901_cast_fp16 = slice_by_index(begin = var_33901_begin_0, end = var_33901_end_0, end_mask = var_33901_end_mask_0, x = v_151_cast_fp16)[name = tensor<string, []>("op_33901_cast_fp16")];
            tensor<int32, [4]> var_33905_begin_0 = const()[name = tensor<string, []>("op_33905_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_33905_end_0 = const()[name = tensor<string, []>("op_33905_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_33905_end_mask_0 = const()[name = tensor<string, []>("op_33905_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_33905_cast_fp16 = slice_by_index(begin = var_33905_begin_0, end = var_33905_end_0, end_mask = var_33905_end_mask_0, x = v_151_cast_fp16)[name = tensor<string, []>("op_33905_cast_fp16")];
            tensor<int32, [4]> var_33909_begin_0 = const()[name = tensor<string, []>("op_33909_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_33909_end_0 = const()[name = tensor<string, []>("op_33909_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_33909_end_mask_0 = const()[name = tensor<string, []>("op_33909_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_33909_cast_fp16 = slice_by_index(begin = var_33909_begin_0, end = var_33909_end_0, end_mask = var_33909_end_mask_0, x = v_151_cast_fp16)[name = tensor<string, []>("op_33909_cast_fp16")];
            tensor<string, []> var_33913_equation_0 = const()[name = tensor<string, []>("op_33913_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33913_cast_fp16 = einsum(equation = var_33913_equation_0, values = (var_33755_cast_fp16, var_33672_cast_fp16))[name = tensor<string, []>("op_33913_cast_fp16")];
            tensor<fp16, []> var_33914_to_fp16 = const()[name = tensor<string, []>("op_33914_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2841_cast_fp16 = mul(x = var_33913_cast_fp16, y = var_33914_to_fp16)[name = tensor<string, []>("aw_2841_cast_fp16")];
            tensor<string, []> var_33917_equation_0 = const()[name = tensor<string, []>("op_33917_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33917_cast_fp16 = einsum(equation = var_33917_equation_0, values = (var_33759_cast_fp16, var_33676_cast_fp16))[name = tensor<string, []>("op_33917_cast_fp16")];
            tensor<fp16, []> var_33918_to_fp16 = const()[name = tensor<string, []>("op_33918_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2843_cast_fp16 = mul(x = var_33917_cast_fp16, y = var_33918_to_fp16)[name = tensor<string, []>("aw_2843_cast_fp16")];
            tensor<string, []> var_33921_equation_0 = const()[name = tensor<string, []>("op_33921_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33921_cast_fp16 = einsum(equation = var_33921_equation_0, values = (var_33763_cast_fp16, var_33680_cast_fp16))[name = tensor<string, []>("op_33921_cast_fp16")];
            tensor<fp16, []> var_33922_to_fp16 = const()[name = tensor<string, []>("op_33922_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2845_cast_fp16 = mul(x = var_33921_cast_fp16, y = var_33922_to_fp16)[name = tensor<string, []>("aw_2845_cast_fp16")];
            tensor<string, []> var_33925_equation_0 = const()[name = tensor<string, []>("op_33925_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33925_cast_fp16 = einsum(equation = var_33925_equation_0, values = (var_33767_cast_fp16, var_33684_cast_fp16))[name = tensor<string, []>("op_33925_cast_fp16")];
            tensor<fp16, []> var_33926_to_fp16 = const()[name = tensor<string, []>("op_33926_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2847_cast_fp16 = mul(x = var_33925_cast_fp16, y = var_33926_to_fp16)[name = tensor<string, []>("aw_2847_cast_fp16")];
            tensor<string, []> var_33929_equation_0 = const()[name = tensor<string, []>("op_33929_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33929_cast_fp16 = einsum(equation = var_33929_equation_0, values = (var_33771_cast_fp16, var_33688_cast_fp16))[name = tensor<string, []>("op_33929_cast_fp16")];
            tensor<fp16, []> var_33930_to_fp16 = const()[name = tensor<string, []>("op_33930_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2849_cast_fp16 = mul(x = var_33929_cast_fp16, y = var_33930_to_fp16)[name = tensor<string, []>("aw_2849_cast_fp16")];
            tensor<string, []> var_33933_equation_0 = const()[name = tensor<string, []>("op_33933_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33933_cast_fp16 = einsum(equation = var_33933_equation_0, values = (var_33775_cast_fp16, var_33692_cast_fp16))[name = tensor<string, []>("op_33933_cast_fp16")];
            tensor<fp16, []> var_33934_to_fp16 = const()[name = tensor<string, []>("op_33934_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2851_cast_fp16 = mul(x = var_33933_cast_fp16, y = var_33934_to_fp16)[name = tensor<string, []>("aw_2851_cast_fp16")];
            tensor<string, []> var_33937_equation_0 = const()[name = tensor<string, []>("op_33937_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33937_cast_fp16 = einsum(equation = var_33937_equation_0, values = (var_33779_cast_fp16, var_33696_cast_fp16))[name = tensor<string, []>("op_33937_cast_fp16")];
            tensor<fp16, []> var_33938_to_fp16 = const()[name = tensor<string, []>("op_33938_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2853_cast_fp16 = mul(x = var_33937_cast_fp16, y = var_33938_to_fp16)[name = tensor<string, []>("aw_2853_cast_fp16")];
            tensor<string, []> var_33941_equation_0 = const()[name = tensor<string, []>("op_33941_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33941_cast_fp16 = einsum(equation = var_33941_equation_0, values = (var_33783_cast_fp16, var_33700_cast_fp16))[name = tensor<string, []>("op_33941_cast_fp16")];
            tensor<fp16, []> var_33942_to_fp16 = const()[name = tensor<string, []>("op_33942_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2855_cast_fp16 = mul(x = var_33941_cast_fp16, y = var_33942_to_fp16)[name = tensor<string, []>("aw_2855_cast_fp16")];
            tensor<string, []> var_33945_equation_0 = const()[name = tensor<string, []>("op_33945_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33945_cast_fp16 = einsum(equation = var_33945_equation_0, values = (var_33787_cast_fp16, var_33704_cast_fp16))[name = tensor<string, []>("op_33945_cast_fp16")];
            tensor<fp16, []> var_33946_to_fp16 = const()[name = tensor<string, []>("op_33946_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2857_cast_fp16 = mul(x = var_33945_cast_fp16, y = var_33946_to_fp16)[name = tensor<string, []>("aw_2857_cast_fp16")];
            tensor<string, []> var_33949_equation_0 = const()[name = tensor<string, []>("op_33949_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33949_cast_fp16 = einsum(equation = var_33949_equation_0, values = (var_33791_cast_fp16, var_33708_cast_fp16))[name = tensor<string, []>("op_33949_cast_fp16")];
            tensor<fp16, []> var_33950_to_fp16 = const()[name = tensor<string, []>("op_33950_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2859_cast_fp16 = mul(x = var_33949_cast_fp16, y = var_33950_to_fp16)[name = tensor<string, []>("aw_2859_cast_fp16")];
            tensor<string, []> var_33953_equation_0 = const()[name = tensor<string, []>("op_33953_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33953_cast_fp16 = einsum(equation = var_33953_equation_0, values = (var_33795_cast_fp16, var_33712_cast_fp16))[name = tensor<string, []>("op_33953_cast_fp16")];
            tensor<fp16, []> var_33954_to_fp16 = const()[name = tensor<string, []>("op_33954_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2861_cast_fp16 = mul(x = var_33953_cast_fp16, y = var_33954_to_fp16)[name = tensor<string, []>("aw_2861_cast_fp16")];
            tensor<string, []> var_33957_equation_0 = const()[name = tensor<string, []>("op_33957_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33957_cast_fp16 = einsum(equation = var_33957_equation_0, values = (var_33799_cast_fp16, var_33716_cast_fp16))[name = tensor<string, []>("op_33957_cast_fp16")];
            tensor<fp16, []> var_33958_to_fp16 = const()[name = tensor<string, []>("op_33958_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2863_cast_fp16 = mul(x = var_33957_cast_fp16, y = var_33958_to_fp16)[name = tensor<string, []>("aw_2863_cast_fp16")];
            tensor<string, []> var_33961_equation_0 = const()[name = tensor<string, []>("op_33961_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33961_cast_fp16 = einsum(equation = var_33961_equation_0, values = (var_33803_cast_fp16, var_33720_cast_fp16))[name = tensor<string, []>("op_33961_cast_fp16")];
            tensor<fp16, []> var_33962_to_fp16 = const()[name = tensor<string, []>("op_33962_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2865_cast_fp16 = mul(x = var_33961_cast_fp16, y = var_33962_to_fp16)[name = tensor<string, []>("aw_2865_cast_fp16")];
            tensor<string, []> var_33965_equation_0 = const()[name = tensor<string, []>("op_33965_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33965_cast_fp16 = einsum(equation = var_33965_equation_0, values = (var_33807_cast_fp16, var_33724_cast_fp16))[name = tensor<string, []>("op_33965_cast_fp16")];
            tensor<fp16, []> var_33966_to_fp16 = const()[name = tensor<string, []>("op_33966_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2867_cast_fp16 = mul(x = var_33965_cast_fp16, y = var_33966_to_fp16)[name = tensor<string, []>("aw_2867_cast_fp16")];
            tensor<string, []> var_33969_equation_0 = const()[name = tensor<string, []>("op_33969_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33969_cast_fp16 = einsum(equation = var_33969_equation_0, values = (var_33811_cast_fp16, var_33728_cast_fp16))[name = tensor<string, []>("op_33969_cast_fp16")];
            tensor<fp16, []> var_33970_to_fp16 = const()[name = tensor<string, []>("op_33970_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2869_cast_fp16 = mul(x = var_33969_cast_fp16, y = var_33970_to_fp16)[name = tensor<string, []>("aw_2869_cast_fp16")];
            tensor<string, []> var_33973_equation_0 = const()[name = tensor<string, []>("op_33973_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33973_cast_fp16 = einsum(equation = var_33973_equation_0, values = (var_33815_cast_fp16, var_33732_cast_fp16))[name = tensor<string, []>("op_33973_cast_fp16")];
            tensor<fp16, []> var_33974_to_fp16 = const()[name = tensor<string, []>("op_33974_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2871_cast_fp16 = mul(x = var_33973_cast_fp16, y = var_33974_to_fp16)[name = tensor<string, []>("aw_2871_cast_fp16")];
            tensor<string, []> var_33977_equation_0 = const()[name = tensor<string, []>("op_33977_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33977_cast_fp16 = einsum(equation = var_33977_equation_0, values = (var_33819_cast_fp16, var_33736_cast_fp16))[name = tensor<string, []>("op_33977_cast_fp16")];
            tensor<fp16, []> var_33978_to_fp16 = const()[name = tensor<string, []>("op_33978_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2873_cast_fp16 = mul(x = var_33977_cast_fp16, y = var_33978_to_fp16)[name = tensor<string, []>("aw_2873_cast_fp16")];
            tensor<string, []> var_33981_equation_0 = const()[name = tensor<string, []>("op_33981_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33981_cast_fp16 = einsum(equation = var_33981_equation_0, values = (var_33823_cast_fp16, var_33740_cast_fp16))[name = tensor<string, []>("op_33981_cast_fp16")];
            tensor<fp16, []> var_33982_to_fp16 = const()[name = tensor<string, []>("op_33982_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2875_cast_fp16 = mul(x = var_33981_cast_fp16, y = var_33982_to_fp16)[name = tensor<string, []>("aw_2875_cast_fp16")];
            tensor<string, []> var_33985_equation_0 = const()[name = tensor<string, []>("op_33985_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33985_cast_fp16 = einsum(equation = var_33985_equation_0, values = (var_33827_cast_fp16, var_33744_cast_fp16))[name = tensor<string, []>("op_33985_cast_fp16")];
            tensor<fp16, []> var_33986_to_fp16 = const()[name = tensor<string, []>("op_33986_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2877_cast_fp16 = mul(x = var_33985_cast_fp16, y = var_33986_to_fp16)[name = tensor<string, []>("aw_2877_cast_fp16")];
            tensor<string, []> var_33989_equation_0 = const()[name = tensor<string, []>("op_33989_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33989_cast_fp16 = einsum(equation = var_33989_equation_0, values = (var_33831_cast_fp16, var_33748_cast_fp16))[name = tensor<string, []>("op_33989_cast_fp16")];
            tensor<fp16, []> var_33990_to_fp16 = const()[name = tensor<string, []>("op_33990_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2879_cast_fp16 = mul(x = var_33989_cast_fp16, y = var_33990_to_fp16)[name = tensor<string, []>("aw_2879_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_33992_cast_fp16 = softmax(axis = var_30355, x = aw_2841_cast_fp16)[name = tensor<string, []>("op_33992_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_33993_cast_fp16 = softmax(axis = var_30355, x = aw_2843_cast_fp16)[name = tensor<string, []>("op_33993_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_33994_cast_fp16 = softmax(axis = var_30355, x = aw_2845_cast_fp16)[name = tensor<string, []>("op_33994_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_33995_cast_fp16 = softmax(axis = var_30355, x = aw_2847_cast_fp16)[name = tensor<string, []>("op_33995_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_33996_cast_fp16 = softmax(axis = var_30355, x = aw_2849_cast_fp16)[name = tensor<string, []>("op_33996_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_33997_cast_fp16 = softmax(axis = var_30355, x = aw_2851_cast_fp16)[name = tensor<string, []>("op_33997_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_33998_cast_fp16 = softmax(axis = var_30355, x = aw_2853_cast_fp16)[name = tensor<string, []>("op_33998_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_33999_cast_fp16 = softmax(axis = var_30355, x = aw_2855_cast_fp16)[name = tensor<string, []>("op_33999_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_34000_cast_fp16 = softmax(axis = var_30355, x = aw_2857_cast_fp16)[name = tensor<string, []>("op_34000_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_34001_cast_fp16 = softmax(axis = var_30355, x = aw_2859_cast_fp16)[name = tensor<string, []>("op_34001_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_34002_cast_fp16 = softmax(axis = var_30355, x = aw_2861_cast_fp16)[name = tensor<string, []>("op_34002_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_34003_cast_fp16 = softmax(axis = var_30355, x = aw_2863_cast_fp16)[name = tensor<string, []>("op_34003_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_34004_cast_fp16 = softmax(axis = var_30355, x = aw_2865_cast_fp16)[name = tensor<string, []>("op_34004_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_34005_cast_fp16 = softmax(axis = var_30355, x = aw_2867_cast_fp16)[name = tensor<string, []>("op_34005_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_34006_cast_fp16 = softmax(axis = var_30355, x = aw_2869_cast_fp16)[name = tensor<string, []>("op_34006_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_34007_cast_fp16 = softmax(axis = var_30355, x = aw_2871_cast_fp16)[name = tensor<string, []>("op_34007_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_34008_cast_fp16 = softmax(axis = var_30355, x = aw_2873_cast_fp16)[name = tensor<string, []>("op_34008_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_34009_cast_fp16 = softmax(axis = var_30355, x = aw_2875_cast_fp16)[name = tensor<string, []>("op_34009_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_34010_cast_fp16 = softmax(axis = var_30355, x = aw_2877_cast_fp16)[name = tensor<string, []>("op_34010_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_34011_cast_fp16 = softmax(axis = var_30355, x = aw_2879_cast_fp16)[name = tensor<string, []>("op_34011_cast_fp16")];
            tensor<string, []> var_34013_equation_0 = const()[name = tensor<string, []>("op_34013_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34013_cast_fp16 = einsum(equation = var_34013_equation_0, values = (var_33833_cast_fp16, var_33992_cast_fp16))[name = tensor<string, []>("op_34013_cast_fp16")];
            tensor<string, []> var_34015_equation_0 = const()[name = tensor<string, []>("op_34015_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34015_cast_fp16 = einsum(equation = var_34015_equation_0, values = (var_33837_cast_fp16, var_33993_cast_fp16))[name = tensor<string, []>("op_34015_cast_fp16")];
            tensor<string, []> var_34017_equation_0 = const()[name = tensor<string, []>("op_34017_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34017_cast_fp16 = einsum(equation = var_34017_equation_0, values = (var_33841_cast_fp16, var_33994_cast_fp16))[name = tensor<string, []>("op_34017_cast_fp16")];
            tensor<string, []> var_34019_equation_0 = const()[name = tensor<string, []>("op_34019_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34019_cast_fp16 = einsum(equation = var_34019_equation_0, values = (var_33845_cast_fp16, var_33995_cast_fp16))[name = tensor<string, []>("op_34019_cast_fp16")];
            tensor<string, []> var_34021_equation_0 = const()[name = tensor<string, []>("op_34021_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34021_cast_fp16 = einsum(equation = var_34021_equation_0, values = (var_33849_cast_fp16, var_33996_cast_fp16))[name = tensor<string, []>("op_34021_cast_fp16")];
            tensor<string, []> var_34023_equation_0 = const()[name = tensor<string, []>("op_34023_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34023_cast_fp16 = einsum(equation = var_34023_equation_0, values = (var_33853_cast_fp16, var_33997_cast_fp16))[name = tensor<string, []>("op_34023_cast_fp16")];
            tensor<string, []> var_34025_equation_0 = const()[name = tensor<string, []>("op_34025_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34025_cast_fp16 = einsum(equation = var_34025_equation_0, values = (var_33857_cast_fp16, var_33998_cast_fp16))[name = tensor<string, []>("op_34025_cast_fp16")];
            tensor<string, []> var_34027_equation_0 = const()[name = tensor<string, []>("op_34027_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34027_cast_fp16 = einsum(equation = var_34027_equation_0, values = (var_33861_cast_fp16, var_33999_cast_fp16))[name = tensor<string, []>("op_34027_cast_fp16")];
            tensor<string, []> var_34029_equation_0 = const()[name = tensor<string, []>("op_34029_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34029_cast_fp16 = einsum(equation = var_34029_equation_0, values = (var_33865_cast_fp16, var_34000_cast_fp16))[name = tensor<string, []>("op_34029_cast_fp16")];
            tensor<string, []> var_34031_equation_0 = const()[name = tensor<string, []>("op_34031_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34031_cast_fp16 = einsum(equation = var_34031_equation_0, values = (var_33869_cast_fp16, var_34001_cast_fp16))[name = tensor<string, []>("op_34031_cast_fp16")];
            tensor<string, []> var_34033_equation_0 = const()[name = tensor<string, []>("op_34033_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34033_cast_fp16 = einsum(equation = var_34033_equation_0, values = (var_33873_cast_fp16, var_34002_cast_fp16))[name = tensor<string, []>("op_34033_cast_fp16")];
            tensor<string, []> var_34035_equation_0 = const()[name = tensor<string, []>("op_34035_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34035_cast_fp16 = einsum(equation = var_34035_equation_0, values = (var_33877_cast_fp16, var_34003_cast_fp16))[name = tensor<string, []>("op_34035_cast_fp16")];
            tensor<string, []> var_34037_equation_0 = const()[name = tensor<string, []>("op_34037_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34037_cast_fp16 = einsum(equation = var_34037_equation_0, values = (var_33881_cast_fp16, var_34004_cast_fp16))[name = tensor<string, []>("op_34037_cast_fp16")];
            tensor<string, []> var_34039_equation_0 = const()[name = tensor<string, []>("op_34039_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34039_cast_fp16 = einsum(equation = var_34039_equation_0, values = (var_33885_cast_fp16, var_34005_cast_fp16))[name = tensor<string, []>("op_34039_cast_fp16")];
            tensor<string, []> var_34041_equation_0 = const()[name = tensor<string, []>("op_34041_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34041_cast_fp16 = einsum(equation = var_34041_equation_0, values = (var_33889_cast_fp16, var_34006_cast_fp16))[name = tensor<string, []>("op_34041_cast_fp16")];
            tensor<string, []> var_34043_equation_0 = const()[name = tensor<string, []>("op_34043_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34043_cast_fp16 = einsum(equation = var_34043_equation_0, values = (var_33893_cast_fp16, var_34007_cast_fp16))[name = tensor<string, []>("op_34043_cast_fp16")];
            tensor<string, []> var_34045_equation_0 = const()[name = tensor<string, []>("op_34045_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34045_cast_fp16 = einsum(equation = var_34045_equation_0, values = (var_33897_cast_fp16, var_34008_cast_fp16))[name = tensor<string, []>("op_34045_cast_fp16")];
            tensor<string, []> var_34047_equation_0 = const()[name = tensor<string, []>("op_34047_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34047_cast_fp16 = einsum(equation = var_34047_equation_0, values = (var_33901_cast_fp16, var_34009_cast_fp16))[name = tensor<string, []>("op_34047_cast_fp16")];
            tensor<string, []> var_34049_equation_0 = const()[name = tensor<string, []>("op_34049_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34049_cast_fp16 = einsum(equation = var_34049_equation_0, values = (var_33905_cast_fp16, var_34010_cast_fp16))[name = tensor<string, []>("op_34049_cast_fp16")];
            tensor<string, []> var_34051_equation_0 = const()[name = tensor<string, []>("op_34051_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34051_cast_fp16 = einsum(equation = var_34051_equation_0, values = (var_33909_cast_fp16, var_34011_cast_fp16))[name = tensor<string, []>("op_34051_cast_fp16")];
            tensor<bool, []> input_465_interleave_0 = const()[name = tensor<string, []>("input_465_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_465_cast_fp16 = concat(axis = var_30355, interleave = input_465_interleave_0, values = (var_34013_cast_fp16, var_34015_cast_fp16, var_34017_cast_fp16, var_34019_cast_fp16, var_34021_cast_fp16, var_34023_cast_fp16, var_34025_cast_fp16, var_34027_cast_fp16, var_34029_cast_fp16, var_34031_cast_fp16, var_34033_cast_fp16, var_34035_cast_fp16, var_34037_cast_fp16, var_34039_cast_fp16, var_34041_cast_fp16, var_34043_cast_fp16, var_34045_cast_fp16, var_34047_cast_fp16, var_34049_cast_fp16, var_34051_cast_fp16))[name = tensor<string, []>("input_465_cast_fp16")];
            tensor<string, []> var_34061_pad_type_0 = const()[name = tensor<string, []>("op_34061_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_34061_strides_0 = const()[name = tensor<string, []>("op_34061_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_34061_pad_0 = const()[name = tensor<string, []>("op_34061_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_34061_dilations_0 = const()[name = tensor<string, []>("op_34061_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_34061_groups_0 = const()[name = tensor<string, []>("op_34061_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_3_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92100096))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93328960))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_3_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_3_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_3_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93329152)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_34061_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_3_attn2_to_out_0_bias_to_fp16, dilations = var_34061_dilations_0, groups = var_34061_groups_0, pad = var_34061_pad_0, pad_type = var_34061_pad_type_0, strides = var_34061_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_3_attn2_to_out_0_weight_to_fp16_palettized, x = input_465_cast_fp16)[name = tensor<string, []>("op_34061_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_227_cast_fp16 = add(x = var_34061_cast_fp16, y = inputs_225_cast_fp16)[name = tensor<string, []>("inputs_227_cast_fp16")];
            tensor<int32, [1]> input_467_axes_0 = const()[name = tensor<string, []>("input_467_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> input_467_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_467_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93331776)))];
            tensor<fp16, [1280]> input_467_beta_0_to_fp16 = const()[name = tensor<string, []>("input_467_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93334400)))];
            tensor<fp16, []> var_34071_to_fp16 = const()[name = tensor<string, []>("op_34071_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> input_467_cast_fp16 = layer_norm(axes = input_467_axes_0, beta = input_467_beta_0_to_fp16, epsilon = var_34071_to_fp16, gamma = input_467_gamma_0_to_fp16, x = inputs_227_cast_fp16)[name = tensor<string, []>("input_467_cast_fp16")];
            tensor<string, []> var_34091_pad_type_0 = const()[name = tensor<string, []>("op_34091_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_34091_strides_0 = const()[name = tensor<string, []>("op_34091_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_34091_pad_0 = const()[name = tensor<string, []>("op_34091_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_34091_dilations_0 = const()[name = tensor<string, []>("op_34091_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_34091_groups_0 = const()[name = tensor<string, []>("op_34091_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_3_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [9830400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93337024))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103167488))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_3_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([10240, 1280, 1, 1])];
            tensor<fp16, [10240]> up_blocks_0_attentions_0_transformer_blocks_3_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_3_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103167680)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_34091_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_3_ff_net_0_proj_bias_to_fp16, dilations = var_34091_dilations_0, groups = var_34091_groups_0, pad = var_34091_pad_0, pad_type = var_34091_pad_type_0, strides = var_34091_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_3_ff_net_0_proj_weight_to_fp16_palettized, x = input_467_cast_fp16)[name = tensor<string, []>("op_34091_cast_fp16")];
            tensor<int32, [2]> var_34092_split_sizes_0 = const()[name = tensor<string, []>("op_34092_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_34092_axis_0 = const()[name = tensor<string, []>("op_34092_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_34092_cast_fp16_0, tensor<fp16, [2, 5120, 1, 1024]> var_34092_cast_fp16_1 = split(axis = var_34092_axis_0, split_sizes = var_34092_split_sizes_0, x = var_34091_cast_fp16)[name = tensor<string, []>("op_34092_cast_fp16")];
            tensor<string, []> var_34094_mode_0 = const()[name = tensor<string, []>("op_34094_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_34094_cast_fp16 = gelu(mode = var_34094_mode_0, x = var_34092_cast_fp16_1)[name = tensor<string, []>("op_34094_cast_fp16")];
            tensor<fp16, [2, 5120, 1, 1024]> input_469_cast_fp16 = mul(x = var_34092_cast_fp16_0, y = var_34094_cast_fp16)[name = tensor<string, []>("input_469_cast_fp16")];
            tensor<string, []> var_34102_pad_type_0 = const()[name = tensor<string, []>("op_34102_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_34102_strides_0 = const()[name = tensor<string, []>("op_34102_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_34102_pad_0 = const()[name = tensor<string, []>("op_34102_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_34102_dilations_0 = const()[name = tensor<string, []>("op_34102_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_34102_groups_0 = const()[name = tensor<string, []>("op_34102_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_3_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [4915200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103188224))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108103488))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_3_ff_net_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 5120, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_3_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_3_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108103680)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_34102_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_3_ff_net_2_bias_to_fp16, dilations = var_34102_dilations_0, groups = var_34102_groups_0, pad = var_34102_pad_0, pad_type = var_34102_pad_type_0, strides = var_34102_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_3_ff_net_2_weight_to_fp16_palettized, x = input_469_cast_fp16)[name = tensor<string, []>("op_34102_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_229_cast_fp16 = add(x = var_34102_cast_fp16, y = inputs_227_cast_fp16)[name = tensor<string, []>("inputs_229_cast_fp16")];
            tensor<int32, [1]> hidden_states_311_axes_0 = const()[name = tensor<string, []>("hidden_states_311_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_311_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_311_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108106304)))];
            tensor<fp16, [1280]> hidden_states_311_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_311_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108108928)))];
            tensor<fp16, []> var_34118_to_fp16 = const()[name = tensor<string, []>("op_34118_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_311_cast_fp16 = layer_norm(axes = hidden_states_311_axes_0, beta = hidden_states_311_beta_0_to_fp16, epsilon = var_34118_to_fp16, gamma = hidden_states_311_gamma_0_to_fp16, x = inputs_229_cast_fp16)[name = tensor<string, []>("hidden_states_311_cast_fp16")];
            tensor<string, []> q_153_pad_type_0 = const()[name = tensor<string, []>("q_153_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_153_strides_0 = const()[name = tensor<string, []>("q_153_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_153_pad_0 = const()[name = tensor<string, []>("q_153_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_153_dilations_0 = const()[name = tensor<string, []>("q_153_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_153_groups_0 = const()[name = tensor<string, []>("q_153_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_4_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108111552))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109340416))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_4_attn1_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_153_cast_fp16 = conv(dilations = q_153_dilations_0, groups = q_153_groups_0, pad = q_153_pad_0, pad_type = q_153_pad_type_0, strides = q_153_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_4_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_311_cast_fp16)[name = tensor<string, []>("q_153_cast_fp16")];
            tensor<string, []> k_305_pad_type_0 = const()[name = tensor<string, []>("k_305_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_305_strides_0 = const()[name = tensor<string, []>("k_305_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_305_pad_0 = const()[name = tensor<string, []>("k_305_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_305_dilations_0 = const()[name = tensor<string, []>("k_305_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_305_groups_0 = const()[name = tensor<string, []>("k_305_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_4_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109340608))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(110569472))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_4_attn1_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> k_305_cast_fp16 = conv(dilations = k_305_dilations_0, groups = k_305_groups_0, pad = k_305_pad_0, pad_type = k_305_pad_type_0, strides = k_305_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_4_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_311_cast_fp16)[name = tensor<string, []>("k_305_cast_fp16")];
            tensor<string, []> v_153_pad_type_0 = const()[name = tensor<string, []>("v_153_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_153_strides_0 = const()[name = tensor<string, []>("v_153_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_153_pad_0 = const()[name = tensor<string, []>("v_153_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_153_dilations_0 = const()[name = tensor<string, []>("v_153_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_153_groups_0 = const()[name = tensor<string, []>("v_153_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_4_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(110569664))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(111798528))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_4_attn1_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> v_153_cast_fp16 = conv(dilations = v_153_dilations_0, groups = v_153_groups_0, pad = v_153_pad_0, pad_type = v_153_pad_type_0, strides = v_153_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_4_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_311_cast_fp16)[name = tensor<string, []>("v_153_cast_fp16")];
            tensor<int32, [4]> var_34151_begin_0 = const()[name = tensor<string, []>("op_34151_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_34151_end_0 = const()[name = tensor<string, []>("op_34151_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_34151_end_mask_0 = const()[name = tensor<string, []>("op_34151_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34151_cast_fp16 = slice_by_index(begin = var_34151_begin_0, end = var_34151_end_0, end_mask = var_34151_end_mask_0, x = q_153_cast_fp16)[name = tensor<string, []>("op_34151_cast_fp16")];
            tensor<int32, [4]> var_34155_begin_0 = const()[name = tensor<string, []>("op_34155_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_34155_end_0 = const()[name = tensor<string, []>("op_34155_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_34155_end_mask_0 = const()[name = tensor<string, []>("op_34155_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34155_cast_fp16 = slice_by_index(begin = var_34155_begin_0, end = var_34155_end_0, end_mask = var_34155_end_mask_0, x = q_153_cast_fp16)[name = tensor<string, []>("op_34155_cast_fp16")];
            tensor<int32, [4]> var_34159_begin_0 = const()[name = tensor<string, []>("op_34159_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_34159_end_0 = const()[name = tensor<string, []>("op_34159_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_34159_end_mask_0 = const()[name = tensor<string, []>("op_34159_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34159_cast_fp16 = slice_by_index(begin = var_34159_begin_0, end = var_34159_end_0, end_mask = var_34159_end_mask_0, x = q_153_cast_fp16)[name = tensor<string, []>("op_34159_cast_fp16")];
            tensor<int32, [4]> var_34163_begin_0 = const()[name = tensor<string, []>("op_34163_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_34163_end_0 = const()[name = tensor<string, []>("op_34163_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_34163_end_mask_0 = const()[name = tensor<string, []>("op_34163_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34163_cast_fp16 = slice_by_index(begin = var_34163_begin_0, end = var_34163_end_0, end_mask = var_34163_end_mask_0, x = q_153_cast_fp16)[name = tensor<string, []>("op_34163_cast_fp16")];
            tensor<int32, [4]> var_34167_begin_0 = const()[name = tensor<string, []>("op_34167_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_34167_end_0 = const()[name = tensor<string, []>("op_34167_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_34167_end_mask_0 = const()[name = tensor<string, []>("op_34167_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34167_cast_fp16 = slice_by_index(begin = var_34167_begin_0, end = var_34167_end_0, end_mask = var_34167_end_mask_0, x = q_153_cast_fp16)[name = tensor<string, []>("op_34167_cast_fp16")];
            tensor<int32, [4]> var_34171_begin_0 = const()[name = tensor<string, []>("op_34171_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_34171_end_0 = const()[name = tensor<string, []>("op_34171_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_34171_end_mask_0 = const()[name = tensor<string, []>("op_34171_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34171_cast_fp16 = slice_by_index(begin = var_34171_begin_0, end = var_34171_end_0, end_mask = var_34171_end_mask_0, x = q_153_cast_fp16)[name = tensor<string, []>("op_34171_cast_fp16")];
            tensor<int32, [4]> var_34175_begin_0 = const()[name = tensor<string, []>("op_34175_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_34175_end_0 = const()[name = tensor<string, []>("op_34175_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_34175_end_mask_0 = const()[name = tensor<string, []>("op_34175_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34175_cast_fp16 = slice_by_index(begin = var_34175_begin_0, end = var_34175_end_0, end_mask = var_34175_end_mask_0, x = q_153_cast_fp16)[name = tensor<string, []>("op_34175_cast_fp16")];
            tensor<int32, [4]> var_34179_begin_0 = const()[name = tensor<string, []>("op_34179_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_34179_end_0 = const()[name = tensor<string, []>("op_34179_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_34179_end_mask_0 = const()[name = tensor<string, []>("op_34179_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34179_cast_fp16 = slice_by_index(begin = var_34179_begin_0, end = var_34179_end_0, end_mask = var_34179_end_mask_0, x = q_153_cast_fp16)[name = tensor<string, []>("op_34179_cast_fp16")];
            tensor<int32, [4]> var_34183_begin_0 = const()[name = tensor<string, []>("op_34183_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_34183_end_0 = const()[name = tensor<string, []>("op_34183_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_34183_end_mask_0 = const()[name = tensor<string, []>("op_34183_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34183_cast_fp16 = slice_by_index(begin = var_34183_begin_0, end = var_34183_end_0, end_mask = var_34183_end_mask_0, x = q_153_cast_fp16)[name = tensor<string, []>("op_34183_cast_fp16")];
            tensor<int32, [4]> var_34187_begin_0 = const()[name = tensor<string, []>("op_34187_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_34187_end_0 = const()[name = tensor<string, []>("op_34187_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_34187_end_mask_0 = const()[name = tensor<string, []>("op_34187_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34187_cast_fp16 = slice_by_index(begin = var_34187_begin_0, end = var_34187_end_0, end_mask = var_34187_end_mask_0, x = q_153_cast_fp16)[name = tensor<string, []>("op_34187_cast_fp16")];
            tensor<int32, [4]> var_34191_begin_0 = const()[name = tensor<string, []>("op_34191_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_34191_end_0 = const()[name = tensor<string, []>("op_34191_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_34191_end_mask_0 = const()[name = tensor<string, []>("op_34191_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34191_cast_fp16 = slice_by_index(begin = var_34191_begin_0, end = var_34191_end_0, end_mask = var_34191_end_mask_0, x = q_153_cast_fp16)[name = tensor<string, []>("op_34191_cast_fp16")];
            tensor<int32, [4]> var_34195_begin_0 = const()[name = tensor<string, []>("op_34195_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_34195_end_0 = const()[name = tensor<string, []>("op_34195_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_34195_end_mask_0 = const()[name = tensor<string, []>("op_34195_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34195_cast_fp16 = slice_by_index(begin = var_34195_begin_0, end = var_34195_end_0, end_mask = var_34195_end_mask_0, x = q_153_cast_fp16)[name = tensor<string, []>("op_34195_cast_fp16")];
            tensor<int32, [4]> var_34199_begin_0 = const()[name = tensor<string, []>("op_34199_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_34199_end_0 = const()[name = tensor<string, []>("op_34199_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_34199_end_mask_0 = const()[name = tensor<string, []>("op_34199_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34199_cast_fp16 = slice_by_index(begin = var_34199_begin_0, end = var_34199_end_0, end_mask = var_34199_end_mask_0, x = q_153_cast_fp16)[name = tensor<string, []>("op_34199_cast_fp16")];
            tensor<int32, [4]> var_34203_begin_0 = const()[name = tensor<string, []>("op_34203_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_34203_end_0 = const()[name = tensor<string, []>("op_34203_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_34203_end_mask_0 = const()[name = tensor<string, []>("op_34203_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34203_cast_fp16 = slice_by_index(begin = var_34203_begin_0, end = var_34203_end_0, end_mask = var_34203_end_mask_0, x = q_153_cast_fp16)[name = tensor<string, []>("op_34203_cast_fp16")];
            tensor<int32, [4]> var_34207_begin_0 = const()[name = tensor<string, []>("op_34207_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_34207_end_0 = const()[name = tensor<string, []>("op_34207_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_34207_end_mask_0 = const()[name = tensor<string, []>("op_34207_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34207_cast_fp16 = slice_by_index(begin = var_34207_begin_0, end = var_34207_end_0, end_mask = var_34207_end_mask_0, x = q_153_cast_fp16)[name = tensor<string, []>("op_34207_cast_fp16")];
            tensor<int32, [4]> var_34211_begin_0 = const()[name = tensor<string, []>("op_34211_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_34211_end_0 = const()[name = tensor<string, []>("op_34211_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_34211_end_mask_0 = const()[name = tensor<string, []>("op_34211_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34211_cast_fp16 = slice_by_index(begin = var_34211_begin_0, end = var_34211_end_0, end_mask = var_34211_end_mask_0, x = q_153_cast_fp16)[name = tensor<string, []>("op_34211_cast_fp16")];
            tensor<int32, [4]> var_34215_begin_0 = const()[name = tensor<string, []>("op_34215_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_34215_end_0 = const()[name = tensor<string, []>("op_34215_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_34215_end_mask_0 = const()[name = tensor<string, []>("op_34215_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34215_cast_fp16 = slice_by_index(begin = var_34215_begin_0, end = var_34215_end_0, end_mask = var_34215_end_mask_0, x = q_153_cast_fp16)[name = tensor<string, []>("op_34215_cast_fp16")];
            tensor<int32, [4]> var_34219_begin_0 = const()[name = tensor<string, []>("op_34219_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_34219_end_0 = const()[name = tensor<string, []>("op_34219_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_34219_end_mask_0 = const()[name = tensor<string, []>("op_34219_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34219_cast_fp16 = slice_by_index(begin = var_34219_begin_0, end = var_34219_end_0, end_mask = var_34219_end_mask_0, x = q_153_cast_fp16)[name = tensor<string, []>("op_34219_cast_fp16")];
            tensor<int32, [4]> var_34223_begin_0 = const()[name = tensor<string, []>("op_34223_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_34223_end_0 = const()[name = tensor<string, []>("op_34223_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_34223_end_mask_0 = const()[name = tensor<string, []>("op_34223_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34223_cast_fp16 = slice_by_index(begin = var_34223_begin_0, end = var_34223_end_0, end_mask = var_34223_end_mask_0, x = q_153_cast_fp16)[name = tensor<string, []>("op_34223_cast_fp16")];
            tensor<int32, [4]> var_34227_begin_0 = const()[name = tensor<string, []>("op_34227_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_34227_end_0 = const()[name = tensor<string, []>("op_34227_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_34227_end_mask_0 = const()[name = tensor<string, []>("op_34227_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34227_cast_fp16 = slice_by_index(begin = var_34227_begin_0, end = var_34227_end_0, end_mask = var_34227_end_mask_0, x = q_153_cast_fp16)[name = tensor<string, []>("op_34227_cast_fp16")];
            tensor<int32, [4]> k_307_perm_0 = const()[name = tensor<string, []>("k_307_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_34234_begin_0 = const()[name = tensor<string, []>("op_34234_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_34234_end_0 = const()[name = tensor<string, []>("op_34234_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_34234_end_mask_0 = const()[name = tensor<string, []>("op_34234_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> k_307_cast_fp16 = transpose(perm = k_307_perm_0, x = k_305_cast_fp16)[name = tensor<string, []>("transpose_63")];
            tensor<fp16, [2, 1024, 1, 64]> var_34234_cast_fp16 = slice_by_index(begin = var_34234_begin_0, end = var_34234_end_0, end_mask = var_34234_end_mask_0, x = k_307_cast_fp16)[name = tensor<string, []>("op_34234_cast_fp16")];
            tensor<int32, [4]> var_34238_begin_0 = const()[name = tensor<string, []>("op_34238_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_34238_end_0 = const()[name = tensor<string, []>("op_34238_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_34238_end_mask_0 = const()[name = tensor<string, []>("op_34238_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_34238_cast_fp16 = slice_by_index(begin = var_34238_begin_0, end = var_34238_end_0, end_mask = var_34238_end_mask_0, x = k_307_cast_fp16)[name = tensor<string, []>("op_34238_cast_fp16")];
            tensor<int32, [4]> var_34242_begin_0 = const()[name = tensor<string, []>("op_34242_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_34242_end_0 = const()[name = tensor<string, []>("op_34242_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_34242_end_mask_0 = const()[name = tensor<string, []>("op_34242_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_34242_cast_fp16 = slice_by_index(begin = var_34242_begin_0, end = var_34242_end_0, end_mask = var_34242_end_mask_0, x = k_307_cast_fp16)[name = tensor<string, []>("op_34242_cast_fp16")];
            tensor<int32, [4]> var_34246_begin_0 = const()[name = tensor<string, []>("op_34246_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_34246_end_0 = const()[name = tensor<string, []>("op_34246_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_34246_end_mask_0 = const()[name = tensor<string, []>("op_34246_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_34246_cast_fp16 = slice_by_index(begin = var_34246_begin_0, end = var_34246_end_0, end_mask = var_34246_end_mask_0, x = k_307_cast_fp16)[name = tensor<string, []>("op_34246_cast_fp16")];
            tensor<int32, [4]> var_34250_begin_0 = const()[name = tensor<string, []>("op_34250_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_34250_end_0 = const()[name = tensor<string, []>("op_34250_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_34250_end_mask_0 = const()[name = tensor<string, []>("op_34250_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_34250_cast_fp16 = slice_by_index(begin = var_34250_begin_0, end = var_34250_end_0, end_mask = var_34250_end_mask_0, x = k_307_cast_fp16)[name = tensor<string, []>("op_34250_cast_fp16")];
            tensor<int32, [4]> var_34254_begin_0 = const()[name = tensor<string, []>("op_34254_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_34254_end_0 = const()[name = tensor<string, []>("op_34254_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_34254_end_mask_0 = const()[name = tensor<string, []>("op_34254_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_34254_cast_fp16 = slice_by_index(begin = var_34254_begin_0, end = var_34254_end_0, end_mask = var_34254_end_mask_0, x = k_307_cast_fp16)[name = tensor<string, []>("op_34254_cast_fp16")];
            tensor<int32, [4]> var_34258_begin_0 = const()[name = tensor<string, []>("op_34258_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_34258_end_0 = const()[name = tensor<string, []>("op_34258_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_34258_end_mask_0 = const()[name = tensor<string, []>("op_34258_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_34258_cast_fp16 = slice_by_index(begin = var_34258_begin_0, end = var_34258_end_0, end_mask = var_34258_end_mask_0, x = k_307_cast_fp16)[name = tensor<string, []>("op_34258_cast_fp16")];
            tensor<int32, [4]> var_34262_begin_0 = const()[name = tensor<string, []>("op_34262_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_34262_end_0 = const()[name = tensor<string, []>("op_34262_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_34262_end_mask_0 = const()[name = tensor<string, []>("op_34262_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_34262_cast_fp16 = slice_by_index(begin = var_34262_begin_0, end = var_34262_end_0, end_mask = var_34262_end_mask_0, x = k_307_cast_fp16)[name = tensor<string, []>("op_34262_cast_fp16")];
            tensor<int32, [4]> var_34266_begin_0 = const()[name = tensor<string, []>("op_34266_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_34266_end_0 = const()[name = tensor<string, []>("op_34266_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_34266_end_mask_0 = const()[name = tensor<string, []>("op_34266_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_34266_cast_fp16 = slice_by_index(begin = var_34266_begin_0, end = var_34266_end_0, end_mask = var_34266_end_mask_0, x = k_307_cast_fp16)[name = tensor<string, []>("op_34266_cast_fp16")];
            tensor<int32, [4]> var_34270_begin_0 = const()[name = tensor<string, []>("op_34270_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_34270_end_0 = const()[name = tensor<string, []>("op_34270_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_34270_end_mask_0 = const()[name = tensor<string, []>("op_34270_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_34270_cast_fp16 = slice_by_index(begin = var_34270_begin_0, end = var_34270_end_0, end_mask = var_34270_end_mask_0, x = k_307_cast_fp16)[name = tensor<string, []>("op_34270_cast_fp16")];
            tensor<int32, [4]> var_34274_begin_0 = const()[name = tensor<string, []>("op_34274_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_34274_end_0 = const()[name = tensor<string, []>("op_34274_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_34274_end_mask_0 = const()[name = tensor<string, []>("op_34274_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_34274_cast_fp16 = slice_by_index(begin = var_34274_begin_0, end = var_34274_end_0, end_mask = var_34274_end_mask_0, x = k_307_cast_fp16)[name = tensor<string, []>("op_34274_cast_fp16")];
            tensor<int32, [4]> var_34278_begin_0 = const()[name = tensor<string, []>("op_34278_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_34278_end_0 = const()[name = tensor<string, []>("op_34278_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_34278_end_mask_0 = const()[name = tensor<string, []>("op_34278_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_34278_cast_fp16 = slice_by_index(begin = var_34278_begin_0, end = var_34278_end_0, end_mask = var_34278_end_mask_0, x = k_307_cast_fp16)[name = tensor<string, []>("op_34278_cast_fp16")];
            tensor<int32, [4]> var_34282_begin_0 = const()[name = tensor<string, []>("op_34282_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_34282_end_0 = const()[name = tensor<string, []>("op_34282_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_34282_end_mask_0 = const()[name = tensor<string, []>("op_34282_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_34282_cast_fp16 = slice_by_index(begin = var_34282_begin_0, end = var_34282_end_0, end_mask = var_34282_end_mask_0, x = k_307_cast_fp16)[name = tensor<string, []>("op_34282_cast_fp16")];
            tensor<int32, [4]> var_34286_begin_0 = const()[name = tensor<string, []>("op_34286_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_34286_end_0 = const()[name = tensor<string, []>("op_34286_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_34286_end_mask_0 = const()[name = tensor<string, []>("op_34286_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_34286_cast_fp16 = slice_by_index(begin = var_34286_begin_0, end = var_34286_end_0, end_mask = var_34286_end_mask_0, x = k_307_cast_fp16)[name = tensor<string, []>("op_34286_cast_fp16")];
            tensor<int32, [4]> var_34290_begin_0 = const()[name = tensor<string, []>("op_34290_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_34290_end_0 = const()[name = tensor<string, []>("op_34290_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_34290_end_mask_0 = const()[name = tensor<string, []>("op_34290_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_34290_cast_fp16 = slice_by_index(begin = var_34290_begin_0, end = var_34290_end_0, end_mask = var_34290_end_mask_0, x = k_307_cast_fp16)[name = tensor<string, []>("op_34290_cast_fp16")];
            tensor<int32, [4]> var_34294_begin_0 = const()[name = tensor<string, []>("op_34294_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_34294_end_0 = const()[name = tensor<string, []>("op_34294_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_34294_end_mask_0 = const()[name = tensor<string, []>("op_34294_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_34294_cast_fp16 = slice_by_index(begin = var_34294_begin_0, end = var_34294_end_0, end_mask = var_34294_end_mask_0, x = k_307_cast_fp16)[name = tensor<string, []>("op_34294_cast_fp16")];
            tensor<int32, [4]> var_34298_begin_0 = const()[name = tensor<string, []>("op_34298_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_34298_end_0 = const()[name = tensor<string, []>("op_34298_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_34298_end_mask_0 = const()[name = tensor<string, []>("op_34298_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_34298_cast_fp16 = slice_by_index(begin = var_34298_begin_0, end = var_34298_end_0, end_mask = var_34298_end_mask_0, x = k_307_cast_fp16)[name = tensor<string, []>("op_34298_cast_fp16")];
            tensor<int32, [4]> var_34302_begin_0 = const()[name = tensor<string, []>("op_34302_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_34302_end_0 = const()[name = tensor<string, []>("op_34302_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_34302_end_mask_0 = const()[name = tensor<string, []>("op_34302_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_34302_cast_fp16 = slice_by_index(begin = var_34302_begin_0, end = var_34302_end_0, end_mask = var_34302_end_mask_0, x = k_307_cast_fp16)[name = tensor<string, []>("op_34302_cast_fp16")];
            tensor<int32, [4]> var_34306_begin_0 = const()[name = tensor<string, []>("op_34306_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_34306_end_0 = const()[name = tensor<string, []>("op_34306_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_34306_end_mask_0 = const()[name = tensor<string, []>("op_34306_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_34306_cast_fp16 = slice_by_index(begin = var_34306_begin_0, end = var_34306_end_0, end_mask = var_34306_end_mask_0, x = k_307_cast_fp16)[name = tensor<string, []>("op_34306_cast_fp16")];
            tensor<int32, [4]> var_34310_begin_0 = const()[name = tensor<string, []>("op_34310_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_34310_end_0 = const()[name = tensor<string, []>("op_34310_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_34310_end_mask_0 = const()[name = tensor<string, []>("op_34310_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_34310_cast_fp16 = slice_by_index(begin = var_34310_begin_0, end = var_34310_end_0, end_mask = var_34310_end_mask_0, x = k_307_cast_fp16)[name = tensor<string, []>("op_34310_cast_fp16")];
            tensor<int32, [4]> var_34312_begin_0 = const()[name = tensor<string, []>("op_34312_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_34312_end_0 = const()[name = tensor<string, []>("op_34312_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_34312_end_mask_0 = const()[name = tensor<string, []>("op_34312_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34312_cast_fp16 = slice_by_index(begin = var_34312_begin_0, end = var_34312_end_0, end_mask = var_34312_end_mask_0, x = v_153_cast_fp16)[name = tensor<string, []>("op_34312_cast_fp16")];
            tensor<int32, [4]> var_34316_begin_0 = const()[name = tensor<string, []>("op_34316_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_34316_end_0 = const()[name = tensor<string, []>("op_34316_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_34316_end_mask_0 = const()[name = tensor<string, []>("op_34316_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34316_cast_fp16 = slice_by_index(begin = var_34316_begin_0, end = var_34316_end_0, end_mask = var_34316_end_mask_0, x = v_153_cast_fp16)[name = tensor<string, []>("op_34316_cast_fp16")];
            tensor<int32, [4]> var_34320_begin_0 = const()[name = tensor<string, []>("op_34320_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_34320_end_0 = const()[name = tensor<string, []>("op_34320_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_34320_end_mask_0 = const()[name = tensor<string, []>("op_34320_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34320_cast_fp16 = slice_by_index(begin = var_34320_begin_0, end = var_34320_end_0, end_mask = var_34320_end_mask_0, x = v_153_cast_fp16)[name = tensor<string, []>("op_34320_cast_fp16")];
            tensor<int32, [4]> var_34324_begin_0 = const()[name = tensor<string, []>("op_34324_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_34324_end_0 = const()[name = tensor<string, []>("op_34324_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_34324_end_mask_0 = const()[name = tensor<string, []>("op_34324_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34324_cast_fp16 = slice_by_index(begin = var_34324_begin_0, end = var_34324_end_0, end_mask = var_34324_end_mask_0, x = v_153_cast_fp16)[name = tensor<string, []>("op_34324_cast_fp16")];
            tensor<int32, [4]> var_34328_begin_0 = const()[name = tensor<string, []>("op_34328_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_34328_end_0 = const()[name = tensor<string, []>("op_34328_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_34328_end_mask_0 = const()[name = tensor<string, []>("op_34328_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34328_cast_fp16 = slice_by_index(begin = var_34328_begin_0, end = var_34328_end_0, end_mask = var_34328_end_mask_0, x = v_153_cast_fp16)[name = tensor<string, []>("op_34328_cast_fp16")];
            tensor<int32, [4]> var_34332_begin_0 = const()[name = tensor<string, []>("op_34332_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_34332_end_0 = const()[name = tensor<string, []>("op_34332_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_34332_end_mask_0 = const()[name = tensor<string, []>("op_34332_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34332_cast_fp16 = slice_by_index(begin = var_34332_begin_0, end = var_34332_end_0, end_mask = var_34332_end_mask_0, x = v_153_cast_fp16)[name = tensor<string, []>("op_34332_cast_fp16")];
            tensor<int32, [4]> var_34336_begin_0 = const()[name = tensor<string, []>("op_34336_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_34336_end_0 = const()[name = tensor<string, []>("op_34336_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_34336_end_mask_0 = const()[name = tensor<string, []>("op_34336_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34336_cast_fp16 = slice_by_index(begin = var_34336_begin_0, end = var_34336_end_0, end_mask = var_34336_end_mask_0, x = v_153_cast_fp16)[name = tensor<string, []>("op_34336_cast_fp16")];
            tensor<int32, [4]> var_34340_begin_0 = const()[name = tensor<string, []>("op_34340_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_34340_end_0 = const()[name = tensor<string, []>("op_34340_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_34340_end_mask_0 = const()[name = tensor<string, []>("op_34340_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34340_cast_fp16 = slice_by_index(begin = var_34340_begin_0, end = var_34340_end_0, end_mask = var_34340_end_mask_0, x = v_153_cast_fp16)[name = tensor<string, []>("op_34340_cast_fp16")];
            tensor<int32, [4]> var_34344_begin_0 = const()[name = tensor<string, []>("op_34344_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_34344_end_0 = const()[name = tensor<string, []>("op_34344_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_34344_end_mask_0 = const()[name = tensor<string, []>("op_34344_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34344_cast_fp16 = slice_by_index(begin = var_34344_begin_0, end = var_34344_end_0, end_mask = var_34344_end_mask_0, x = v_153_cast_fp16)[name = tensor<string, []>("op_34344_cast_fp16")];
            tensor<int32, [4]> var_34348_begin_0 = const()[name = tensor<string, []>("op_34348_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_34348_end_0 = const()[name = tensor<string, []>("op_34348_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_34348_end_mask_0 = const()[name = tensor<string, []>("op_34348_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34348_cast_fp16 = slice_by_index(begin = var_34348_begin_0, end = var_34348_end_0, end_mask = var_34348_end_mask_0, x = v_153_cast_fp16)[name = tensor<string, []>("op_34348_cast_fp16")];
            tensor<int32, [4]> var_34352_begin_0 = const()[name = tensor<string, []>("op_34352_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_34352_end_0 = const()[name = tensor<string, []>("op_34352_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_34352_end_mask_0 = const()[name = tensor<string, []>("op_34352_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34352_cast_fp16 = slice_by_index(begin = var_34352_begin_0, end = var_34352_end_0, end_mask = var_34352_end_mask_0, x = v_153_cast_fp16)[name = tensor<string, []>("op_34352_cast_fp16")];
            tensor<int32, [4]> var_34356_begin_0 = const()[name = tensor<string, []>("op_34356_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_34356_end_0 = const()[name = tensor<string, []>("op_34356_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_34356_end_mask_0 = const()[name = tensor<string, []>("op_34356_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34356_cast_fp16 = slice_by_index(begin = var_34356_begin_0, end = var_34356_end_0, end_mask = var_34356_end_mask_0, x = v_153_cast_fp16)[name = tensor<string, []>("op_34356_cast_fp16")];
            tensor<int32, [4]> var_34360_begin_0 = const()[name = tensor<string, []>("op_34360_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_34360_end_0 = const()[name = tensor<string, []>("op_34360_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_34360_end_mask_0 = const()[name = tensor<string, []>("op_34360_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34360_cast_fp16 = slice_by_index(begin = var_34360_begin_0, end = var_34360_end_0, end_mask = var_34360_end_mask_0, x = v_153_cast_fp16)[name = tensor<string, []>("op_34360_cast_fp16")];
            tensor<int32, [4]> var_34364_begin_0 = const()[name = tensor<string, []>("op_34364_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_34364_end_0 = const()[name = tensor<string, []>("op_34364_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_34364_end_mask_0 = const()[name = tensor<string, []>("op_34364_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34364_cast_fp16 = slice_by_index(begin = var_34364_begin_0, end = var_34364_end_0, end_mask = var_34364_end_mask_0, x = v_153_cast_fp16)[name = tensor<string, []>("op_34364_cast_fp16")];
            tensor<int32, [4]> var_34368_begin_0 = const()[name = tensor<string, []>("op_34368_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_34368_end_0 = const()[name = tensor<string, []>("op_34368_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_34368_end_mask_0 = const()[name = tensor<string, []>("op_34368_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34368_cast_fp16 = slice_by_index(begin = var_34368_begin_0, end = var_34368_end_0, end_mask = var_34368_end_mask_0, x = v_153_cast_fp16)[name = tensor<string, []>("op_34368_cast_fp16")];
            tensor<int32, [4]> var_34372_begin_0 = const()[name = tensor<string, []>("op_34372_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_34372_end_0 = const()[name = tensor<string, []>("op_34372_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_34372_end_mask_0 = const()[name = tensor<string, []>("op_34372_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34372_cast_fp16 = slice_by_index(begin = var_34372_begin_0, end = var_34372_end_0, end_mask = var_34372_end_mask_0, x = v_153_cast_fp16)[name = tensor<string, []>("op_34372_cast_fp16")];
            tensor<int32, [4]> var_34376_begin_0 = const()[name = tensor<string, []>("op_34376_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_34376_end_0 = const()[name = tensor<string, []>("op_34376_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_34376_end_mask_0 = const()[name = tensor<string, []>("op_34376_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34376_cast_fp16 = slice_by_index(begin = var_34376_begin_0, end = var_34376_end_0, end_mask = var_34376_end_mask_0, x = v_153_cast_fp16)[name = tensor<string, []>("op_34376_cast_fp16")];
            tensor<int32, [4]> var_34380_begin_0 = const()[name = tensor<string, []>("op_34380_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_34380_end_0 = const()[name = tensor<string, []>("op_34380_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_34380_end_mask_0 = const()[name = tensor<string, []>("op_34380_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34380_cast_fp16 = slice_by_index(begin = var_34380_begin_0, end = var_34380_end_0, end_mask = var_34380_end_mask_0, x = v_153_cast_fp16)[name = tensor<string, []>("op_34380_cast_fp16")];
            tensor<int32, [4]> var_34384_begin_0 = const()[name = tensor<string, []>("op_34384_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_34384_end_0 = const()[name = tensor<string, []>("op_34384_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_34384_end_mask_0 = const()[name = tensor<string, []>("op_34384_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34384_cast_fp16 = slice_by_index(begin = var_34384_begin_0, end = var_34384_end_0, end_mask = var_34384_end_mask_0, x = v_153_cast_fp16)[name = tensor<string, []>("op_34384_cast_fp16")];
            tensor<int32, [4]> var_34388_begin_0 = const()[name = tensor<string, []>("op_34388_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_34388_end_0 = const()[name = tensor<string, []>("op_34388_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_34388_end_mask_0 = const()[name = tensor<string, []>("op_34388_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34388_cast_fp16 = slice_by_index(begin = var_34388_begin_0, end = var_34388_end_0, end_mask = var_34388_end_mask_0, x = v_153_cast_fp16)[name = tensor<string, []>("op_34388_cast_fp16")];
            tensor<string, []> var_34392_equation_0 = const()[name = tensor<string, []>("op_34392_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34392_cast_fp16 = einsum(equation = var_34392_equation_0, values = (var_34234_cast_fp16, var_34151_cast_fp16))[name = tensor<string, []>("op_34392_cast_fp16")];
            tensor<fp16, []> var_34393_to_fp16 = const()[name = tensor<string, []>("op_34393_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2881_cast_fp16 = mul(x = var_34392_cast_fp16, y = var_34393_to_fp16)[name = tensor<string, []>("aw_2881_cast_fp16")];
            tensor<string, []> var_34396_equation_0 = const()[name = tensor<string, []>("op_34396_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34396_cast_fp16 = einsum(equation = var_34396_equation_0, values = (var_34238_cast_fp16, var_34155_cast_fp16))[name = tensor<string, []>("op_34396_cast_fp16")];
            tensor<fp16, []> var_34397_to_fp16 = const()[name = tensor<string, []>("op_34397_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2883_cast_fp16 = mul(x = var_34396_cast_fp16, y = var_34397_to_fp16)[name = tensor<string, []>("aw_2883_cast_fp16")];
            tensor<string, []> var_34400_equation_0 = const()[name = tensor<string, []>("op_34400_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34400_cast_fp16 = einsum(equation = var_34400_equation_0, values = (var_34242_cast_fp16, var_34159_cast_fp16))[name = tensor<string, []>("op_34400_cast_fp16")];
            tensor<fp16, []> var_34401_to_fp16 = const()[name = tensor<string, []>("op_34401_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2885_cast_fp16 = mul(x = var_34400_cast_fp16, y = var_34401_to_fp16)[name = tensor<string, []>("aw_2885_cast_fp16")];
            tensor<string, []> var_34404_equation_0 = const()[name = tensor<string, []>("op_34404_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34404_cast_fp16 = einsum(equation = var_34404_equation_0, values = (var_34246_cast_fp16, var_34163_cast_fp16))[name = tensor<string, []>("op_34404_cast_fp16")];
            tensor<fp16, []> var_34405_to_fp16 = const()[name = tensor<string, []>("op_34405_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2887_cast_fp16 = mul(x = var_34404_cast_fp16, y = var_34405_to_fp16)[name = tensor<string, []>("aw_2887_cast_fp16")];
            tensor<string, []> var_34408_equation_0 = const()[name = tensor<string, []>("op_34408_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34408_cast_fp16 = einsum(equation = var_34408_equation_0, values = (var_34250_cast_fp16, var_34167_cast_fp16))[name = tensor<string, []>("op_34408_cast_fp16")];
            tensor<fp16, []> var_34409_to_fp16 = const()[name = tensor<string, []>("op_34409_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2889_cast_fp16 = mul(x = var_34408_cast_fp16, y = var_34409_to_fp16)[name = tensor<string, []>("aw_2889_cast_fp16")];
            tensor<string, []> var_34412_equation_0 = const()[name = tensor<string, []>("op_34412_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34412_cast_fp16 = einsum(equation = var_34412_equation_0, values = (var_34254_cast_fp16, var_34171_cast_fp16))[name = tensor<string, []>("op_34412_cast_fp16")];
            tensor<fp16, []> var_34413_to_fp16 = const()[name = tensor<string, []>("op_34413_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2891_cast_fp16 = mul(x = var_34412_cast_fp16, y = var_34413_to_fp16)[name = tensor<string, []>("aw_2891_cast_fp16")];
            tensor<string, []> var_34416_equation_0 = const()[name = tensor<string, []>("op_34416_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34416_cast_fp16 = einsum(equation = var_34416_equation_0, values = (var_34258_cast_fp16, var_34175_cast_fp16))[name = tensor<string, []>("op_34416_cast_fp16")];
            tensor<fp16, []> var_34417_to_fp16 = const()[name = tensor<string, []>("op_34417_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2893_cast_fp16 = mul(x = var_34416_cast_fp16, y = var_34417_to_fp16)[name = tensor<string, []>("aw_2893_cast_fp16")];
            tensor<string, []> var_34420_equation_0 = const()[name = tensor<string, []>("op_34420_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34420_cast_fp16 = einsum(equation = var_34420_equation_0, values = (var_34262_cast_fp16, var_34179_cast_fp16))[name = tensor<string, []>("op_34420_cast_fp16")];
            tensor<fp16, []> var_34421_to_fp16 = const()[name = tensor<string, []>("op_34421_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2895_cast_fp16 = mul(x = var_34420_cast_fp16, y = var_34421_to_fp16)[name = tensor<string, []>("aw_2895_cast_fp16")];
            tensor<string, []> var_34424_equation_0 = const()[name = tensor<string, []>("op_34424_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34424_cast_fp16 = einsum(equation = var_34424_equation_0, values = (var_34266_cast_fp16, var_34183_cast_fp16))[name = tensor<string, []>("op_34424_cast_fp16")];
            tensor<fp16, []> var_34425_to_fp16 = const()[name = tensor<string, []>("op_34425_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2897_cast_fp16 = mul(x = var_34424_cast_fp16, y = var_34425_to_fp16)[name = tensor<string, []>("aw_2897_cast_fp16")];
            tensor<string, []> var_34428_equation_0 = const()[name = tensor<string, []>("op_34428_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34428_cast_fp16 = einsum(equation = var_34428_equation_0, values = (var_34270_cast_fp16, var_34187_cast_fp16))[name = tensor<string, []>("op_34428_cast_fp16")];
            tensor<fp16, []> var_34429_to_fp16 = const()[name = tensor<string, []>("op_34429_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2899_cast_fp16 = mul(x = var_34428_cast_fp16, y = var_34429_to_fp16)[name = tensor<string, []>("aw_2899_cast_fp16")];
            tensor<string, []> var_34432_equation_0 = const()[name = tensor<string, []>("op_34432_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34432_cast_fp16 = einsum(equation = var_34432_equation_0, values = (var_34274_cast_fp16, var_34191_cast_fp16))[name = tensor<string, []>("op_34432_cast_fp16")];
            tensor<fp16, []> var_34433_to_fp16 = const()[name = tensor<string, []>("op_34433_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2901_cast_fp16 = mul(x = var_34432_cast_fp16, y = var_34433_to_fp16)[name = tensor<string, []>("aw_2901_cast_fp16")];
            tensor<string, []> var_34436_equation_0 = const()[name = tensor<string, []>("op_34436_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34436_cast_fp16 = einsum(equation = var_34436_equation_0, values = (var_34278_cast_fp16, var_34195_cast_fp16))[name = tensor<string, []>("op_34436_cast_fp16")];
            tensor<fp16, []> var_34437_to_fp16 = const()[name = tensor<string, []>("op_34437_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2903_cast_fp16 = mul(x = var_34436_cast_fp16, y = var_34437_to_fp16)[name = tensor<string, []>("aw_2903_cast_fp16")];
            tensor<string, []> var_34440_equation_0 = const()[name = tensor<string, []>("op_34440_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34440_cast_fp16 = einsum(equation = var_34440_equation_0, values = (var_34282_cast_fp16, var_34199_cast_fp16))[name = tensor<string, []>("op_34440_cast_fp16")];
            tensor<fp16, []> var_34441_to_fp16 = const()[name = tensor<string, []>("op_34441_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2905_cast_fp16 = mul(x = var_34440_cast_fp16, y = var_34441_to_fp16)[name = tensor<string, []>("aw_2905_cast_fp16")];
            tensor<string, []> var_34444_equation_0 = const()[name = tensor<string, []>("op_34444_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34444_cast_fp16 = einsum(equation = var_34444_equation_0, values = (var_34286_cast_fp16, var_34203_cast_fp16))[name = tensor<string, []>("op_34444_cast_fp16")];
            tensor<fp16, []> var_34445_to_fp16 = const()[name = tensor<string, []>("op_34445_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2907_cast_fp16 = mul(x = var_34444_cast_fp16, y = var_34445_to_fp16)[name = tensor<string, []>("aw_2907_cast_fp16")];
            tensor<string, []> var_34448_equation_0 = const()[name = tensor<string, []>("op_34448_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34448_cast_fp16 = einsum(equation = var_34448_equation_0, values = (var_34290_cast_fp16, var_34207_cast_fp16))[name = tensor<string, []>("op_34448_cast_fp16")];
            tensor<fp16, []> var_34449_to_fp16 = const()[name = tensor<string, []>("op_34449_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2909_cast_fp16 = mul(x = var_34448_cast_fp16, y = var_34449_to_fp16)[name = tensor<string, []>("aw_2909_cast_fp16")];
            tensor<string, []> var_34452_equation_0 = const()[name = tensor<string, []>("op_34452_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34452_cast_fp16 = einsum(equation = var_34452_equation_0, values = (var_34294_cast_fp16, var_34211_cast_fp16))[name = tensor<string, []>("op_34452_cast_fp16")];
            tensor<fp16, []> var_34453_to_fp16 = const()[name = tensor<string, []>("op_34453_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2911_cast_fp16 = mul(x = var_34452_cast_fp16, y = var_34453_to_fp16)[name = tensor<string, []>("aw_2911_cast_fp16")];
            tensor<string, []> var_34456_equation_0 = const()[name = tensor<string, []>("op_34456_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34456_cast_fp16 = einsum(equation = var_34456_equation_0, values = (var_34298_cast_fp16, var_34215_cast_fp16))[name = tensor<string, []>("op_34456_cast_fp16")];
            tensor<fp16, []> var_34457_to_fp16 = const()[name = tensor<string, []>("op_34457_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2913_cast_fp16 = mul(x = var_34456_cast_fp16, y = var_34457_to_fp16)[name = tensor<string, []>("aw_2913_cast_fp16")];
            tensor<string, []> var_34460_equation_0 = const()[name = tensor<string, []>("op_34460_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34460_cast_fp16 = einsum(equation = var_34460_equation_0, values = (var_34302_cast_fp16, var_34219_cast_fp16))[name = tensor<string, []>("op_34460_cast_fp16")];
            tensor<fp16, []> var_34461_to_fp16 = const()[name = tensor<string, []>("op_34461_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2915_cast_fp16 = mul(x = var_34460_cast_fp16, y = var_34461_to_fp16)[name = tensor<string, []>("aw_2915_cast_fp16")];
            tensor<string, []> var_34464_equation_0 = const()[name = tensor<string, []>("op_34464_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34464_cast_fp16 = einsum(equation = var_34464_equation_0, values = (var_34306_cast_fp16, var_34223_cast_fp16))[name = tensor<string, []>("op_34464_cast_fp16")];
            tensor<fp16, []> var_34465_to_fp16 = const()[name = tensor<string, []>("op_34465_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2917_cast_fp16 = mul(x = var_34464_cast_fp16, y = var_34465_to_fp16)[name = tensor<string, []>("aw_2917_cast_fp16")];
            tensor<string, []> var_34468_equation_0 = const()[name = tensor<string, []>("op_34468_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34468_cast_fp16 = einsum(equation = var_34468_equation_0, values = (var_34310_cast_fp16, var_34227_cast_fp16))[name = tensor<string, []>("op_34468_cast_fp16")];
            tensor<fp16, []> var_34469_to_fp16 = const()[name = tensor<string, []>("op_34469_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2919_cast_fp16 = mul(x = var_34468_cast_fp16, y = var_34469_to_fp16)[name = tensor<string, []>("aw_2919_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34471_cast_fp16 = softmax(axis = var_30355, x = aw_2881_cast_fp16)[name = tensor<string, []>("op_34471_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34472_cast_fp16 = softmax(axis = var_30355, x = aw_2883_cast_fp16)[name = tensor<string, []>("op_34472_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34473_cast_fp16 = softmax(axis = var_30355, x = aw_2885_cast_fp16)[name = tensor<string, []>("op_34473_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34474_cast_fp16 = softmax(axis = var_30355, x = aw_2887_cast_fp16)[name = tensor<string, []>("op_34474_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34475_cast_fp16 = softmax(axis = var_30355, x = aw_2889_cast_fp16)[name = tensor<string, []>("op_34475_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34476_cast_fp16 = softmax(axis = var_30355, x = aw_2891_cast_fp16)[name = tensor<string, []>("op_34476_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34477_cast_fp16 = softmax(axis = var_30355, x = aw_2893_cast_fp16)[name = tensor<string, []>("op_34477_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34478_cast_fp16 = softmax(axis = var_30355, x = aw_2895_cast_fp16)[name = tensor<string, []>("op_34478_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34479_cast_fp16 = softmax(axis = var_30355, x = aw_2897_cast_fp16)[name = tensor<string, []>("op_34479_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34480_cast_fp16 = softmax(axis = var_30355, x = aw_2899_cast_fp16)[name = tensor<string, []>("op_34480_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34481_cast_fp16 = softmax(axis = var_30355, x = aw_2901_cast_fp16)[name = tensor<string, []>("op_34481_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34482_cast_fp16 = softmax(axis = var_30355, x = aw_2903_cast_fp16)[name = tensor<string, []>("op_34482_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34483_cast_fp16 = softmax(axis = var_30355, x = aw_2905_cast_fp16)[name = tensor<string, []>("op_34483_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34484_cast_fp16 = softmax(axis = var_30355, x = aw_2907_cast_fp16)[name = tensor<string, []>("op_34484_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34485_cast_fp16 = softmax(axis = var_30355, x = aw_2909_cast_fp16)[name = tensor<string, []>("op_34485_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34486_cast_fp16 = softmax(axis = var_30355, x = aw_2911_cast_fp16)[name = tensor<string, []>("op_34486_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34487_cast_fp16 = softmax(axis = var_30355, x = aw_2913_cast_fp16)[name = tensor<string, []>("op_34487_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34488_cast_fp16 = softmax(axis = var_30355, x = aw_2915_cast_fp16)[name = tensor<string, []>("op_34488_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34489_cast_fp16 = softmax(axis = var_30355, x = aw_2917_cast_fp16)[name = tensor<string, []>("op_34489_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34490_cast_fp16 = softmax(axis = var_30355, x = aw_2919_cast_fp16)[name = tensor<string, []>("op_34490_cast_fp16")];
            tensor<string, []> var_34492_equation_0 = const()[name = tensor<string, []>("op_34492_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34492_cast_fp16 = einsum(equation = var_34492_equation_0, values = (var_34312_cast_fp16, var_34471_cast_fp16))[name = tensor<string, []>("op_34492_cast_fp16")];
            tensor<string, []> var_34494_equation_0 = const()[name = tensor<string, []>("op_34494_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34494_cast_fp16 = einsum(equation = var_34494_equation_0, values = (var_34316_cast_fp16, var_34472_cast_fp16))[name = tensor<string, []>("op_34494_cast_fp16")];
            tensor<string, []> var_34496_equation_0 = const()[name = tensor<string, []>("op_34496_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34496_cast_fp16 = einsum(equation = var_34496_equation_0, values = (var_34320_cast_fp16, var_34473_cast_fp16))[name = tensor<string, []>("op_34496_cast_fp16")];
            tensor<string, []> var_34498_equation_0 = const()[name = tensor<string, []>("op_34498_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34498_cast_fp16 = einsum(equation = var_34498_equation_0, values = (var_34324_cast_fp16, var_34474_cast_fp16))[name = tensor<string, []>("op_34498_cast_fp16")];
            tensor<string, []> var_34500_equation_0 = const()[name = tensor<string, []>("op_34500_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34500_cast_fp16 = einsum(equation = var_34500_equation_0, values = (var_34328_cast_fp16, var_34475_cast_fp16))[name = tensor<string, []>("op_34500_cast_fp16")];
            tensor<string, []> var_34502_equation_0 = const()[name = tensor<string, []>("op_34502_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34502_cast_fp16 = einsum(equation = var_34502_equation_0, values = (var_34332_cast_fp16, var_34476_cast_fp16))[name = tensor<string, []>("op_34502_cast_fp16")];
            tensor<string, []> var_34504_equation_0 = const()[name = tensor<string, []>("op_34504_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34504_cast_fp16 = einsum(equation = var_34504_equation_0, values = (var_34336_cast_fp16, var_34477_cast_fp16))[name = tensor<string, []>("op_34504_cast_fp16")];
            tensor<string, []> var_34506_equation_0 = const()[name = tensor<string, []>("op_34506_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34506_cast_fp16 = einsum(equation = var_34506_equation_0, values = (var_34340_cast_fp16, var_34478_cast_fp16))[name = tensor<string, []>("op_34506_cast_fp16")];
            tensor<string, []> var_34508_equation_0 = const()[name = tensor<string, []>("op_34508_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34508_cast_fp16 = einsum(equation = var_34508_equation_0, values = (var_34344_cast_fp16, var_34479_cast_fp16))[name = tensor<string, []>("op_34508_cast_fp16")];
            tensor<string, []> var_34510_equation_0 = const()[name = tensor<string, []>("op_34510_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34510_cast_fp16 = einsum(equation = var_34510_equation_0, values = (var_34348_cast_fp16, var_34480_cast_fp16))[name = tensor<string, []>("op_34510_cast_fp16")];
            tensor<string, []> var_34512_equation_0 = const()[name = tensor<string, []>("op_34512_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34512_cast_fp16 = einsum(equation = var_34512_equation_0, values = (var_34352_cast_fp16, var_34481_cast_fp16))[name = tensor<string, []>("op_34512_cast_fp16")];
            tensor<string, []> var_34514_equation_0 = const()[name = tensor<string, []>("op_34514_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34514_cast_fp16 = einsum(equation = var_34514_equation_0, values = (var_34356_cast_fp16, var_34482_cast_fp16))[name = tensor<string, []>("op_34514_cast_fp16")];
            tensor<string, []> var_34516_equation_0 = const()[name = tensor<string, []>("op_34516_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34516_cast_fp16 = einsum(equation = var_34516_equation_0, values = (var_34360_cast_fp16, var_34483_cast_fp16))[name = tensor<string, []>("op_34516_cast_fp16")];
            tensor<string, []> var_34518_equation_0 = const()[name = tensor<string, []>("op_34518_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34518_cast_fp16 = einsum(equation = var_34518_equation_0, values = (var_34364_cast_fp16, var_34484_cast_fp16))[name = tensor<string, []>("op_34518_cast_fp16")];
            tensor<string, []> var_34520_equation_0 = const()[name = tensor<string, []>("op_34520_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34520_cast_fp16 = einsum(equation = var_34520_equation_0, values = (var_34368_cast_fp16, var_34485_cast_fp16))[name = tensor<string, []>("op_34520_cast_fp16")];
            tensor<string, []> var_34522_equation_0 = const()[name = tensor<string, []>("op_34522_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34522_cast_fp16 = einsum(equation = var_34522_equation_0, values = (var_34372_cast_fp16, var_34486_cast_fp16))[name = tensor<string, []>("op_34522_cast_fp16")];
            tensor<string, []> var_34524_equation_0 = const()[name = tensor<string, []>("op_34524_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34524_cast_fp16 = einsum(equation = var_34524_equation_0, values = (var_34376_cast_fp16, var_34487_cast_fp16))[name = tensor<string, []>("op_34524_cast_fp16")];
            tensor<string, []> var_34526_equation_0 = const()[name = tensor<string, []>("op_34526_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34526_cast_fp16 = einsum(equation = var_34526_equation_0, values = (var_34380_cast_fp16, var_34488_cast_fp16))[name = tensor<string, []>("op_34526_cast_fp16")];
            tensor<string, []> var_34528_equation_0 = const()[name = tensor<string, []>("op_34528_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34528_cast_fp16 = einsum(equation = var_34528_equation_0, values = (var_34384_cast_fp16, var_34489_cast_fp16))[name = tensor<string, []>("op_34528_cast_fp16")];
            tensor<string, []> var_34530_equation_0 = const()[name = tensor<string, []>("op_34530_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34530_cast_fp16 = einsum(equation = var_34530_equation_0, values = (var_34388_cast_fp16, var_34490_cast_fp16))[name = tensor<string, []>("op_34530_cast_fp16")];
            tensor<bool, []> input_471_interleave_0 = const()[name = tensor<string, []>("input_471_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_471_cast_fp16 = concat(axis = var_30355, interleave = input_471_interleave_0, values = (var_34492_cast_fp16, var_34494_cast_fp16, var_34496_cast_fp16, var_34498_cast_fp16, var_34500_cast_fp16, var_34502_cast_fp16, var_34504_cast_fp16, var_34506_cast_fp16, var_34508_cast_fp16, var_34510_cast_fp16, var_34512_cast_fp16, var_34514_cast_fp16, var_34516_cast_fp16, var_34518_cast_fp16, var_34520_cast_fp16, var_34522_cast_fp16, var_34524_cast_fp16, var_34526_cast_fp16, var_34528_cast_fp16, var_34530_cast_fp16))[name = tensor<string, []>("input_471_cast_fp16")];
            tensor<string, []> var_34540_pad_type_0 = const()[name = tensor<string, []>("op_34540_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_34540_strides_0 = const()[name = tensor<string, []>("op_34540_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_34540_pad_0 = const()[name = tensor<string, []>("op_34540_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_34540_dilations_0 = const()[name = tensor<string, []>("op_34540_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_34540_groups_0 = const()[name = tensor<string, []>("op_34540_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_4_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(111798720))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113027584))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_4_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_4_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_4_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113027776)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_34540_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_4_attn1_to_out_0_bias_to_fp16, dilations = var_34540_dilations_0, groups = var_34540_groups_0, pad = var_34540_pad_0, pad_type = var_34540_pad_type_0, strides = var_34540_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_4_attn1_to_out_0_weight_to_fp16_palettized, x = input_471_cast_fp16)[name = tensor<string, []>("op_34540_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_231_cast_fp16 = add(x = var_34540_cast_fp16, y = inputs_229_cast_fp16)[name = tensor<string, []>("inputs_231_cast_fp16")];
            tensor<int32, [1]> hidden_states_313_axes_0 = const()[name = tensor<string, []>("hidden_states_313_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_313_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_313_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113030400)))];
            tensor<fp16, [1280]> hidden_states_313_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_313_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113033024)))];
            tensor<fp16, []> var_34550_to_fp16 = const()[name = tensor<string, []>("op_34550_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_313_cast_fp16 = layer_norm(axes = hidden_states_313_axes_0, beta = hidden_states_313_beta_0_to_fp16, epsilon = var_34550_to_fp16, gamma = hidden_states_313_gamma_0_to_fp16, x = inputs_231_cast_fp16)[name = tensor<string, []>("hidden_states_313_cast_fp16")];
            tensor<string, []> q_155_pad_type_0 = const()[name = tensor<string, []>("q_155_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_155_strides_0 = const()[name = tensor<string, []>("q_155_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_155_pad_0 = const()[name = tensor<string, []>("q_155_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_155_dilations_0 = const()[name = tensor<string, []>("q_155_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_155_groups_0 = const()[name = tensor<string, []>("q_155_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_4_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113035648))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(114264512))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_4_attn2_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_155_cast_fp16 = conv(dilations = q_155_dilations_0, groups = q_155_groups_0, pad = q_155_pad_0, pad_type = q_155_pad_type_0, strides = q_155_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_4_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_313_cast_fp16)[name = tensor<string, []>("q_155_cast_fp16")];
            tensor<string, []> k_309_pad_type_0 = const()[name = tensor<string, []>("k_309_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_309_strides_0 = const()[name = tensor<string, []>("k_309_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_309_pad_0 = const()[name = tensor<string, []>("k_309_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_309_dilations_0 = const()[name = tensor<string, []>("k_309_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_309_groups_0 = const()[name = tensor<string, []>("k_309_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_4_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(114264704))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(116230848))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_4_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> k_309_cast_fp16 = conv(dilations = k_309_dilations_0, groups = k_309_groups_0, pad = k_309_pad_0, pad_type = k_309_pad_type_0, strides = k_309_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_4_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("k_309_cast_fp16")];
            tensor<string, []> v_155_pad_type_0 = const()[name = tensor<string, []>("v_155_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_155_strides_0 = const()[name = tensor<string, []>("v_155_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_155_pad_0 = const()[name = tensor<string, []>("v_155_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_155_dilations_0 = const()[name = tensor<string, []>("v_155_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_155_groups_0 = const()[name = tensor<string, []>("v_155_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_4_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(116231040))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118197184))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_4_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> v_155_cast_fp16 = conv(dilations = v_155_dilations_0, groups = v_155_groups_0, pad = v_155_pad_0, pad_type = v_155_pad_type_0, strides = v_155_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_4_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("v_155_cast_fp16")];
            tensor<int32, [4]> var_34583_begin_0 = const()[name = tensor<string, []>("op_34583_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_34583_end_0 = const()[name = tensor<string, []>("op_34583_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_34583_end_mask_0 = const()[name = tensor<string, []>("op_34583_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34583_cast_fp16 = slice_by_index(begin = var_34583_begin_0, end = var_34583_end_0, end_mask = var_34583_end_mask_0, x = q_155_cast_fp16)[name = tensor<string, []>("op_34583_cast_fp16")];
            tensor<int32, [4]> var_34587_begin_0 = const()[name = tensor<string, []>("op_34587_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_34587_end_0 = const()[name = tensor<string, []>("op_34587_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_34587_end_mask_0 = const()[name = tensor<string, []>("op_34587_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34587_cast_fp16 = slice_by_index(begin = var_34587_begin_0, end = var_34587_end_0, end_mask = var_34587_end_mask_0, x = q_155_cast_fp16)[name = tensor<string, []>("op_34587_cast_fp16")];
            tensor<int32, [4]> var_34591_begin_0 = const()[name = tensor<string, []>("op_34591_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_34591_end_0 = const()[name = tensor<string, []>("op_34591_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_34591_end_mask_0 = const()[name = tensor<string, []>("op_34591_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34591_cast_fp16 = slice_by_index(begin = var_34591_begin_0, end = var_34591_end_0, end_mask = var_34591_end_mask_0, x = q_155_cast_fp16)[name = tensor<string, []>("op_34591_cast_fp16")];
            tensor<int32, [4]> var_34595_begin_0 = const()[name = tensor<string, []>("op_34595_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_34595_end_0 = const()[name = tensor<string, []>("op_34595_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_34595_end_mask_0 = const()[name = tensor<string, []>("op_34595_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34595_cast_fp16 = slice_by_index(begin = var_34595_begin_0, end = var_34595_end_0, end_mask = var_34595_end_mask_0, x = q_155_cast_fp16)[name = tensor<string, []>("op_34595_cast_fp16")];
            tensor<int32, [4]> var_34599_begin_0 = const()[name = tensor<string, []>("op_34599_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_34599_end_0 = const()[name = tensor<string, []>("op_34599_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_34599_end_mask_0 = const()[name = tensor<string, []>("op_34599_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34599_cast_fp16 = slice_by_index(begin = var_34599_begin_0, end = var_34599_end_0, end_mask = var_34599_end_mask_0, x = q_155_cast_fp16)[name = tensor<string, []>("op_34599_cast_fp16")];
            tensor<int32, [4]> var_34603_begin_0 = const()[name = tensor<string, []>("op_34603_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_34603_end_0 = const()[name = tensor<string, []>("op_34603_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_34603_end_mask_0 = const()[name = tensor<string, []>("op_34603_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34603_cast_fp16 = slice_by_index(begin = var_34603_begin_0, end = var_34603_end_0, end_mask = var_34603_end_mask_0, x = q_155_cast_fp16)[name = tensor<string, []>("op_34603_cast_fp16")];
            tensor<int32, [4]> var_34607_begin_0 = const()[name = tensor<string, []>("op_34607_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_34607_end_0 = const()[name = tensor<string, []>("op_34607_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_34607_end_mask_0 = const()[name = tensor<string, []>("op_34607_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34607_cast_fp16 = slice_by_index(begin = var_34607_begin_0, end = var_34607_end_0, end_mask = var_34607_end_mask_0, x = q_155_cast_fp16)[name = tensor<string, []>("op_34607_cast_fp16")];
            tensor<int32, [4]> var_34611_begin_0 = const()[name = tensor<string, []>("op_34611_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_34611_end_0 = const()[name = tensor<string, []>("op_34611_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_34611_end_mask_0 = const()[name = tensor<string, []>("op_34611_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34611_cast_fp16 = slice_by_index(begin = var_34611_begin_0, end = var_34611_end_0, end_mask = var_34611_end_mask_0, x = q_155_cast_fp16)[name = tensor<string, []>("op_34611_cast_fp16")];
            tensor<int32, [4]> var_34615_begin_0 = const()[name = tensor<string, []>("op_34615_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_34615_end_0 = const()[name = tensor<string, []>("op_34615_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_34615_end_mask_0 = const()[name = tensor<string, []>("op_34615_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34615_cast_fp16 = slice_by_index(begin = var_34615_begin_0, end = var_34615_end_0, end_mask = var_34615_end_mask_0, x = q_155_cast_fp16)[name = tensor<string, []>("op_34615_cast_fp16")];
            tensor<int32, [4]> var_34619_begin_0 = const()[name = tensor<string, []>("op_34619_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_34619_end_0 = const()[name = tensor<string, []>("op_34619_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_34619_end_mask_0 = const()[name = tensor<string, []>("op_34619_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34619_cast_fp16 = slice_by_index(begin = var_34619_begin_0, end = var_34619_end_0, end_mask = var_34619_end_mask_0, x = q_155_cast_fp16)[name = tensor<string, []>("op_34619_cast_fp16")];
            tensor<int32, [4]> var_34623_begin_0 = const()[name = tensor<string, []>("op_34623_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_34623_end_0 = const()[name = tensor<string, []>("op_34623_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_34623_end_mask_0 = const()[name = tensor<string, []>("op_34623_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34623_cast_fp16 = slice_by_index(begin = var_34623_begin_0, end = var_34623_end_0, end_mask = var_34623_end_mask_0, x = q_155_cast_fp16)[name = tensor<string, []>("op_34623_cast_fp16")];
            tensor<int32, [4]> var_34627_begin_0 = const()[name = tensor<string, []>("op_34627_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_34627_end_0 = const()[name = tensor<string, []>("op_34627_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_34627_end_mask_0 = const()[name = tensor<string, []>("op_34627_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34627_cast_fp16 = slice_by_index(begin = var_34627_begin_0, end = var_34627_end_0, end_mask = var_34627_end_mask_0, x = q_155_cast_fp16)[name = tensor<string, []>("op_34627_cast_fp16")];
            tensor<int32, [4]> var_34631_begin_0 = const()[name = tensor<string, []>("op_34631_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_34631_end_0 = const()[name = tensor<string, []>("op_34631_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_34631_end_mask_0 = const()[name = tensor<string, []>("op_34631_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34631_cast_fp16 = slice_by_index(begin = var_34631_begin_0, end = var_34631_end_0, end_mask = var_34631_end_mask_0, x = q_155_cast_fp16)[name = tensor<string, []>("op_34631_cast_fp16")];
            tensor<int32, [4]> var_34635_begin_0 = const()[name = tensor<string, []>("op_34635_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_34635_end_0 = const()[name = tensor<string, []>("op_34635_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_34635_end_mask_0 = const()[name = tensor<string, []>("op_34635_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34635_cast_fp16 = slice_by_index(begin = var_34635_begin_0, end = var_34635_end_0, end_mask = var_34635_end_mask_0, x = q_155_cast_fp16)[name = tensor<string, []>("op_34635_cast_fp16")];
            tensor<int32, [4]> var_34639_begin_0 = const()[name = tensor<string, []>("op_34639_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_34639_end_0 = const()[name = tensor<string, []>("op_34639_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_34639_end_mask_0 = const()[name = tensor<string, []>("op_34639_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34639_cast_fp16 = slice_by_index(begin = var_34639_begin_0, end = var_34639_end_0, end_mask = var_34639_end_mask_0, x = q_155_cast_fp16)[name = tensor<string, []>("op_34639_cast_fp16")];
            tensor<int32, [4]> var_34643_begin_0 = const()[name = tensor<string, []>("op_34643_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_34643_end_0 = const()[name = tensor<string, []>("op_34643_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_34643_end_mask_0 = const()[name = tensor<string, []>("op_34643_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34643_cast_fp16 = slice_by_index(begin = var_34643_begin_0, end = var_34643_end_0, end_mask = var_34643_end_mask_0, x = q_155_cast_fp16)[name = tensor<string, []>("op_34643_cast_fp16")];
            tensor<int32, [4]> var_34647_begin_0 = const()[name = tensor<string, []>("op_34647_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_34647_end_0 = const()[name = tensor<string, []>("op_34647_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_34647_end_mask_0 = const()[name = tensor<string, []>("op_34647_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34647_cast_fp16 = slice_by_index(begin = var_34647_begin_0, end = var_34647_end_0, end_mask = var_34647_end_mask_0, x = q_155_cast_fp16)[name = tensor<string, []>("op_34647_cast_fp16")];
            tensor<int32, [4]> var_34651_begin_0 = const()[name = tensor<string, []>("op_34651_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_34651_end_0 = const()[name = tensor<string, []>("op_34651_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_34651_end_mask_0 = const()[name = tensor<string, []>("op_34651_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34651_cast_fp16 = slice_by_index(begin = var_34651_begin_0, end = var_34651_end_0, end_mask = var_34651_end_mask_0, x = q_155_cast_fp16)[name = tensor<string, []>("op_34651_cast_fp16")];
            tensor<int32, [4]> var_34655_begin_0 = const()[name = tensor<string, []>("op_34655_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_34655_end_0 = const()[name = tensor<string, []>("op_34655_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_34655_end_mask_0 = const()[name = tensor<string, []>("op_34655_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34655_cast_fp16 = slice_by_index(begin = var_34655_begin_0, end = var_34655_end_0, end_mask = var_34655_end_mask_0, x = q_155_cast_fp16)[name = tensor<string, []>("op_34655_cast_fp16")];
            tensor<int32, [4]> var_34659_begin_0 = const()[name = tensor<string, []>("op_34659_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_34659_end_0 = const()[name = tensor<string, []>("op_34659_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_34659_end_mask_0 = const()[name = tensor<string, []>("op_34659_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34659_cast_fp16 = slice_by_index(begin = var_34659_begin_0, end = var_34659_end_0, end_mask = var_34659_end_mask_0, x = q_155_cast_fp16)[name = tensor<string, []>("op_34659_cast_fp16")];
            tensor<int32, [4]> k_311_perm_0 = const()[name = tensor<string, []>("k_311_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_34666_begin_0 = const()[name = tensor<string, []>("op_34666_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_34666_end_0 = const()[name = tensor<string, []>("op_34666_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_34666_end_mask_0 = const()[name = tensor<string, []>("op_34666_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> k_311_cast_fp16 = transpose(perm = k_311_perm_0, x = k_309_cast_fp16)[name = tensor<string, []>("transpose_62")];
            tensor<fp16, [2, 77, 1, 64]> var_34666_cast_fp16 = slice_by_index(begin = var_34666_begin_0, end = var_34666_end_0, end_mask = var_34666_end_mask_0, x = k_311_cast_fp16)[name = tensor<string, []>("op_34666_cast_fp16")];
            tensor<int32, [4]> var_34670_begin_0 = const()[name = tensor<string, []>("op_34670_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_34670_end_0 = const()[name = tensor<string, []>("op_34670_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_34670_end_mask_0 = const()[name = tensor<string, []>("op_34670_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_34670_cast_fp16 = slice_by_index(begin = var_34670_begin_0, end = var_34670_end_0, end_mask = var_34670_end_mask_0, x = k_311_cast_fp16)[name = tensor<string, []>("op_34670_cast_fp16")];
            tensor<int32, [4]> var_34674_begin_0 = const()[name = tensor<string, []>("op_34674_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_34674_end_0 = const()[name = tensor<string, []>("op_34674_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_34674_end_mask_0 = const()[name = tensor<string, []>("op_34674_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_34674_cast_fp16 = slice_by_index(begin = var_34674_begin_0, end = var_34674_end_0, end_mask = var_34674_end_mask_0, x = k_311_cast_fp16)[name = tensor<string, []>("op_34674_cast_fp16")];
            tensor<int32, [4]> var_34678_begin_0 = const()[name = tensor<string, []>("op_34678_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_34678_end_0 = const()[name = tensor<string, []>("op_34678_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_34678_end_mask_0 = const()[name = tensor<string, []>("op_34678_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_34678_cast_fp16 = slice_by_index(begin = var_34678_begin_0, end = var_34678_end_0, end_mask = var_34678_end_mask_0, x = k_311_cast_fp16)[name = tensor<string, []>("op_34678_cast_fp16")];
            tensor<int32, [4]> var_34682_begin_0 = const()[name = tensor<string, []>("op_34682_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_34682_end_0 = const()[name = tensor<string, []>("op_34682_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_34682_end_mask_0 = const()[name = tensor<string, []>("op_34682_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_34682_cast_fp16 = slice_by_index(begin = var_34682_begin_0, end = var_34682_end_0, end_mask = var_34682_end_mask_0, x = k_311_cast_fp16)[name = tensor<string, []>("op_34682_cast_fp16")];
            tensor<int32, [4]> var_34686_begin_0 = const()[name = tensor<string, []>("op_34686_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_34686_end_0 = const()[name = tensor<string, []>("op_34686_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_34686_end_mask_0 = const()[name = tensor<string, []>("op_34686_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_34686_cast_fp16 = slice_by_index(begin = var_34686_begin_0, end = var_34686_end_0, end_mask = var_34686_end_mask_0, x = k_311_cast_fp16)[name = tensor<string, []>("op_34686_cast_fp16")];
            tensor<int32, [4]> var_34690_begin_0 = const()[name = tensor<string, []>("op_34690_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_34690_end_0 = const()[name = tensor<string, []>("op_34690_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_34690_end_mask_0 = const()[name = tensor<string, []>("op_34690_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_34690_cast_fp16 = slice_by_index(begin = var_34690_begin_0, end = var_34690_end_0, end_mask = var_34690_end_mask_0, x = k_311_cast_fp16)[name = tensor<string, []>("op_34690_cast_fp16")];
            tensor<int32, [4]> var_34694_begin_0 = const()[name = tensor<string, []>("op_34694_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_34694_end_0 = const()[name = tensor<string, []>("op_34694_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_34694_end_mask_0 = const()[name = tensor<string, []>("op_34694_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_34694_cast_fp16 = slice_by_index(begin = var_34694_begin_0, end = var_34694_end_0, end_mask = var_34694_end_mask_0, x = k_311_cast_fp16)[name = tensor<string, []>("op_34694_cast_fp16")];
            tensor<int32, [4]> var_34698_begin_0 = const()[name = tensor<string, []>("op_34698_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_34698_end_0 = const()[name = tensor<string, []>("op_34698_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_34698_end_mask_0 = const()[name = tensor<string, []>("op_34698_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_34698_cast_fp16 = slice_by_index(begin = var_34698_begin_0, end = var_34698_end_0, end_mask = var_34698_end_mask_0, x = k_311_cast_fp16)[name = tensor<string, []>("op_34698_cast_fp16")];
            tensor<int32, [4]> var_34702_begin_0 = const()[name = tensor<string, []>("op_34702_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_34702_end_0 = const()[name = tensor<string, []>("op_34702_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_34702_end_mask_0 = const()[name = tensor<string, []>("op_34702_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_34702_cast_fp16 = slice_by_index(begin = var_34702_begin_0, end = var_34702_end_0, end_mask = var_34702_end_mask_0, x = k_311_cast_fp16)[name = tensor<string, []>("op_34702_cast_fp16")];
            tensor<int32, [4]> var_34706_begin_0 = const()[name = tensor<string, []>("op_34706_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_34706_end_0 = const()[name = tensor<string, []>("op_34706_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_34706_end_mask_0 = const()[name = tensor<string, []>("op_34706_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_34706_cast_fp16 = slice_by_index(begin = var_34706_begin_0, end = var_34706_end_0, end_mask = var_34706_end_mask_0, x = k_311_cast_fp16)[name = tensor<string, []>("op_34706_cast_fp16")];
            tensor<int32, [4]> var_34710_begin_0 = const()[name = tensor<string, []>("op_34710_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_34710_end_0 = const()[name = tensor<string, []>("op_34710_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_34710_end_mask_0 = const()[name = tensor<string, []>("op_34710_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_34710_cast_fp16 = slice_by_index(begin = var_34710_begin_0, end = var_34710_end_0, end_mask = var_34710_end_mask_0, x = k_311_cast_fp16)[name = tensor<string, []>("op_34710_cast_fp16")];
            tensor<int32, [4]> var_34714_begin_0 = const()[name = tensor<string, []>("op_34714_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_34714_end_0 = const()[name = tensor<string, []>("op_34714_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_34714_end_mask_0 = const()[name = tensor<string, []>("op_34714_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_34714_cast_fp16 = slice_by_index(begin = var_34714_begin_0, end = var_34714_end_0, end_mask = var_34714_end_mask_0, x = k_311_cast_fp16)[name = tensor<string, []>("op_34714_cast_fp16")];
            tensor<int32, [4]> var_34718_begin_0 = const()[name = tensor<string, []>("op_34718_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_34718_end_0 = const()[name = tensor<string, []>("op_34718_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_34718_end_mask_0 = const()[name = tensor<string, []>("op_34718_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_34718_cast_fp16 = slice_by_index(begin = var_34718_begin_0, end = var_34718_end_0, end_mask = var_34718_end_mask_0, x = k_311_cast_fp16)[name = tensor<string, []>("op_34718_cast_fp16")];
            tensor<int32, [4]> var_34722_begin_0 = const()[name = tensor<string, []>("op_34722_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_34722_end_0 = const()[name = tensor<string, []>("op_34722_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_34722_end_mask_0 = const()[name = tensor<string, []>("op_34722_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_34722_cast_fp16 = slice_by_index(begin = var_34722_begin_0, end = var_34722_end_0, end_mask = var_34722_end_mask_0, x = k_311_cast_fp16)[name = tensor<string, []>("op_34722_cast_fp16")];
            tensor<int32, [4]> var_34726_begin_0 = const()[name = tensor<string, []>("op_34726_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_34726_end_0 = const()[name = tensor<string, []>("op_34726_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_34726_end_mask_0 = const()[name = tensor<string, []>("op_34726_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_34726_cast_fp16 = slice_by_index(begin = var_34726_begin_0, end = var_34726_end_0, end_mask = var_34726_end_mask_0, x = k_311_cast_fp16)[name = tensor<string, []>("op_34726_cast_fp16")];
            tensor<int32, [4]> var_34730_begin_0 = const()[name = tensor<string, []>("op_34730_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_34730_end_0 = const()[name = tensor<string, []>("op_34730_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_34730_end_mask_0 = const()[name = tensor<string, []>("op_34730_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_34730_cast_fp16 = slice_by_index(begin = var_34730_begin_0, end = var_34730_end_0, end_mask = var_34730_end_mask_0, x = k_311_cast_fp16)[name = tensor<string, []>("op_34730_cast_fp16")];
            tensor<int32, [4]> var_34734_begin_0 = const()[name = tensor<string, []>("op_34734_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_34734_end_0 = const()[name = tensor<string, []>("op_34734_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_34734_end_mask_0 = const()[name = tensor<string, []>("op_34734_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_34734_cast_fp16 = slice_by_index(begin = var_34734_begin_0, end = var_34734_end_0, end_mask = var_34734_end_mask_0, x = k_311_cast_fp16)[name = tensor<string, []>("op_34734_cast_fp16")];
            tensor<int32, [4]> var_34738_begin_0 = const()[name = tensor<string, []>("op_34738_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_34738_end_0 = const()[name = tensor<string, []>("op_34738_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_34738_end_mask_0 = const()[name = tensor<string, []>("op_34738_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_34738_cast_fp16 = slice_by_index(begin = var_34738_begin_0, end = var_34738_end_0, end_mask = var_34738_end_mask_0, x = k_311_cast_fp16)[name = tensor<string, []>("op_34738_cast_fp16")];
            tensor<int32, [4]> var_34742_begin_0 = const()[name = tensor<string, []>("op_34742_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_34742_end_0 = const()[name = tensor<string, []>("op_34742_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_34742_end_mask_0 = const()[name = tensor<string, []>("op_34742_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_34742_cast_fp16 = slice_by_index(begin = var_34742_begin_0, end = var_34742_end_0, end_mask = var_34742_end_mask_0, x = k_311_cast_fp16)[name = tensor<string, []>("op_34742_cast_fp16")];
            tensor<int32, [4]> var_34744_begin_0 = const()[name = tensor<string, []>("op_34744_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_34744_end_0 = const()[name = tensor<string, []>("op_34744_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_34744_end_mask_0 = const()[name = tensor<string, []>("op_34744_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_34744_cast_fp16 = slice_by_index(begin = var_34744_begin_0, end = var_34744_end_0, end_mask = var_34744_end_mask_0, x = v_155_cast_fp16)[name = tensor<string, []>("op_34744_cast_fp16")];
            tensor<int32, [4]> var_34748_begin_0 = const()[name = tensor<string, []>("op_34748_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_34748_end_0 = const()[name = tensor<string, []>("op_34748_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_34748_end_mask_0 = const()[name = tensor<string, []>("op_34748_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_34748_cast_fp16 = slice_by_index(begin = var_34748_begin_0, end = var_34748_end_0, end_mask = var_34748_end_mask_0, x = v_155_cast_fp16)[name = tensor<string, []>("op_34748_cast_fp16")];
            tensor<int32, [4]> var_34752_begin_0 = const()[name = tensor<string, []>("op_34752_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_34752_end_0 = const()[name = tensor<string, []>("op_34752_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_34752_end_mask_0 = const()[name = tensor<string, []>("op_34752_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_34752_cast_fp16 = slice_by_index(begin = var_34752_begin_0, end = var_34752_end_0, end_mask = var_34752_end_mask_0, x = v_155_cast_fp16)[name = tensor<string, []>("op_34752_cast_fp16")];
            tensor<int32, [4]> var_34756_begin_0 = const()[name = tensor<string, []>("op_34756_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_34756_end_0 = const()[name = tensor<string, []>("op_34756_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_34756_end_mask_0 = const()[name = tensor<string, []>("op_34756_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_34756_cast_fp16 = slice_by_index(begin = var_34756_begin_0, end = var_34756_end_0, end_mask = var_34756_end_mask_0, x = v_155_cast_fp16)[name = tensor<string, []>("op_34756_cast_fp16")];
            tensor<int32, [4]> var_34760_begin_0 = const()[name = tensor<string, []>("op_34760_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_34760_end_0 = const()[name = tensor<string, []>("op_34760_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_34760_end_mask_0 = const()[name = tensor<string, []>("op_34760_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_34760_cast_fp16 = slice_by_index(begin = var_34760_begin_0, end = var_34760_end_0, end_mask = var_34760_end_mask_0, x = v_155_cast_fp16)[name = tensor<string, []>("op_34760_cast_fp16")];
            tensor<int32, [4]> var_34764_begin_0 = const()[name = tensor<string, []>("op_34764_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_34764_end_0 = const()[name = tensor<string, []>("op_34764_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_34764_end_mask_0 = const()[name = tensor<string, []>("op_34764_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_34764_cast_fp16 = slice_by_index(begin = var_34764_begin_0, end = var_34764_end_0, end_mask = var_34764_end_mask_0, x = v_155_cast_fp16)[name = tensor<string, []>("op_34764_cast_fp16")];
            tensor<int32, [4]> var_34768_begin_0 = const()[name = tensor<string, []>("op_34768_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_34768_end_0 = const()[name = tensor<string, []>("op_34768_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_34768_end_mask_0 = const()[name = tensor<string, []>("op_34768_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_34768_cast_fp16 = slice_by_index(begin = var_34768_begin_0, end = var_34768_end_0, end_mask = var_34768_end_mask_0, x = v_155_cast_fp16)[name = tensor<string, []>("op_34768_cast_fp16")];
            tensor<int32, [4]> var_34772_begin_0 = const()[name = tensor<string, []>("op_34772_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_34772_end_0 = const()[name = tensor<string, []>("op_34772_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_34772_end_mask_0 = const()[name = tensor<string, []>("op_34772_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_34772_cast_fp16 = slice_by_index(begin = var_34772_begin_0, end = var_34772_end_0, end_mask = var_34772_end_mask_0, x = v_155_cast_fp16)[name = tensor<string, []>("op_34772_cast_fp16")];
            tensor<int32, [4]> var_34776_begin_0 = const()[name = tensor<string, []>("op_34776_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_34776_end_0 = const()[name = tensor<string, []>("op_34776_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_34776_end_mask_0 = const()[name = tensor<string, []>("op_34776_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_34776_cast_fp16 = slice_by_index(begin = var_34776_begin_0, end = var_34776_end_0, end_mask = var_34776_end_mask_0, x = v_155_cast_fp16)[name = tensor<string, []>("op_34776_cast_fp16")];
            tensor<int32, [4]> var_34780_begin_0 = const()[name = tensor<string, []>("op_34780_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_34780_end_0 = const()[name = tensor<string, []>("op_34780_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_34780_end_mask_0 = const()[name = tensor<string, []>("op_34780_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_34780_cast_fp16 = slice_by_index(begin = var_34780_begin_0, end = var_34780_end_0, end_mask = var_34780_end_mask_0, x = v_155_cast_fp16)[name = tensor<string, []>("op_34780_cast_fp16")];
            tensor<int32, [4]> var_34784_begin_0 = const()[name = tensor<string, []>("op_34784_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_34784_end_0 = const()[name = tensor<string, []>("op_34784_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_34784_end_mask_0 = const()[name = tensor<string, []>("op_34784_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_34784_cast_fp16 = slice_by_index(begin = var_34784_begin_0, end = var_34784_end_0, end_mask = var_34784_end_mask_0, x = v_155_cast_fp16)[name = tensor<string, []>("op_34784_cast_fp16")];
            tensor<int32, [4]> var_34788_begin_0 = const()[name = tensor<string, []>("op_34788_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_34788_end_0 = const()[name = tensor<string, []>("op_34788_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_34788_end_mask_0 = const()[name = tensor<string, []>("op_34788_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_34788_cast_fp16 = slice_by_index(begin = var_34788_begin_0, end = var_34788_end_0, end_mask = var_34788_end_mask_0, x = v_155_cast_fp16)[name = tensor<string, []>("op_34788_cast_fp16")];
            tensor<int32, [4]> var_34792_begin_0 = const()[name = tensor<string, []>("op_34792_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_34792_end_0 = const()[name = tensor<string, []>("op_34792_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_34792_end_mask_0 = const()[name = tensor<string, []>("op_34792_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_34792_cast_fp16 = slice_by_index(begin = var_34792_begin_0, end = var_34792_end_0, end_mask = var_34792_end_mask_0, x = v_155_cast_fp16)[name = tensor<string, []>("op_34792_cast_fp16")];
            tensor<int32, [4]> var_34796_begin_0 = const()[name = tensor<string, []>("op_34796_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_34796_end_0 = const()[name = tensor<string, []>("op_34796_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_34796_end_mask_0 = const()[name = tensor<string, []>("op_34796_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_34796_cast_fp16 = slice_by_index(begin = var_34796_begin_0, end = var_34796_end_0, end_mask = var_34796_end_mask_0, x = v_155_cast_fp16)[name = tensor<string, []>("op_34796_cast_fp16")];
            tensor<int32, [4]> var_34800_begin_0 = const()[name = tensor<string, []>("op_34800_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_34800_end_0 = const()[name = tensor<string, []>("op_34800_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_34800_end_mask_0 = const()[name = tensor<string, []>("op_34800_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_34800_cast_fp16 = slice_by_index(begin = var_34800_begin_0, end = var_34800_end_0, end_mask = var_34800_end_mask_0, x = v_155_cast_fp16)[name = tensor<string, []>("op_34800_cast_fp16")];
            tensor<int32, [4]> var_34804_begin_0 = const()[name = tensor<string, []>("op_34804_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_34804_end_0 = const()[name = tensor<string, []>("op_34804_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_34804_end_mask_0 = const()[name = tensor<string, []>("op_34804_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_34804_cast_fp16 = slice_by_index(begin = var_34804_begin_0, end = var_34804_end_0, end_mask = var_34804_end_mask_0, x = v_155_cast_fp16)[name = tensor<string, []>("op_34804_cast_fp16")];
            tensor<int32, [4]> var_34808_begin_0 = const()[name = tensor<string, []>("op_34808_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_34808_end_0 = const()[name = tensor<string, []>("op_34808_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_34808_end_mask_0 = const()[name = tensor<string, []>("op_34808_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_34808_cast_fp16 = slice_by_index(begin = var_34808_begin_0, end = var_34808_end_0, end_mask = var_34808_end_mask_0, x = v_155_cast_fp16)[name = tensor<string, []>("op_34808_cast_fp16")];
            tensor<int32, [4]> var_34812_begin_0 = const()[name = tensor<string, []>("op_34812_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_34812_end_0 = const()[name = tensor<string, []>("op_34812_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_34812_end_mask_0 = const()[name = tensor<string, []>("op_34812_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_34812_cast_fp16 = slice_by_index(begin = var_34812_begin_0, end = var_34812_end_0, end_mask = var_34812_end_mask_0, x = v_155_cast_fp16)[name = tensor<string, []>("op_34812_cast_fp16")];
            tensor<int32, [4]> var_34816_begin_0 = const()[name = tensor<string, []>("op_34816_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_34816_end_0 = const()[name = tensor<string, []>("op_34816_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_34816_end_mask_0 = const()[name = tensor<string, []>("op_34816_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_34816_cast_fp16 = slice_by_index(begin = var_34816_begin_0, end = var_34816_end_0, end_mask = var_34816_end_mask_0, x = v_155_cast_fp16)[name = tensor<string, []>("op_34816_cast_fp16")];
            tensor<int32, [4]> var_34820_begin_0 = const()[name = tensor<string, []>("op_34820_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_34820_end_0 = const()[name = tensor<string, []>("op_34820_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_34820_end_mask_0 = const()[name = tensor<string, []>("op_34820_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_34820_cast_fp16 = slice_by_index(begin = var_34820_begin_0, end = var_34820_end_0, end_mask = var_34820_end_mask_0, x = v_155_cast_fp16)[name = tensor<string, []>("op_34820_cast_fp16")];
            tensor<string, []> var_34824_equation_0 = const()[name = tensor<string, []>("op_34824_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_34824_cast_fp16 = einsum(equation = var_34824_equation_0, values = (var_34666_cast_fp16, var_34583_cast_fp16))[name = tensor<string, []>("op_34824_cast_fp16")];
            tensor<fp16, []> var_34825_to_fp16 = const()[name = tensor<string, []>("op_34825_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2921_cast_fp16 = mul(x = var_34824_cast_fp16, y = var_34825_to_fp16)[name = tensor<string, []>("aw_2921_cast_fp16")];
            tensor<string, []> var_34828_equation_0 = const()[name = tensor<string, []>("op_34828_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_34828_cast_fp16 = einsum(equation = var_34828_equation_0, values = (var_34670_cast_fp16, var_34587_cast_fp16))[name = tensor<string, []>("op_34828_cast_fp16")];
            tensor<fp16, []> var_34829_to_fp16 = const()[name = tensor<string, []>("op_34829_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2923_cast_fp16 = mul(x = var_34828_cast_fp16, y = var_34829_to_fp16)[name = tensor<string, []>("aw_2923_cast_fp16")];
            tensor<string, []> var_34832_equation_0 = const()[name = tensor<string, []>("op_34832_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_34832_cast_fp16 = einsum(equation = var_34832_equation_0, values = (var_34674_cast_fp16, var_34591_cast_fp16))[name = tensor<string, []>("op_34832_cast_fp16")];
            tensor<fp16, []> var_34833_to_fp16 = const()[name = tensor<string, []>("op_34833_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2925_cast_fp16 = mul(x = var_34832_cast_fp16, y = var_34833_to_fp16)[name = tensor<string, []>("aw_2925_cast_fp16")];
            tensor<string, []> var_34836_equation_0 = const()[name = tensor<string, []>("op_34836_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_34836_cast_fp16 = einsum(equation = var_34836_equation_0, values = (var_34678_cast_fp16, var_34595_cast_fp16))[name = tensor<string, []>("op_34836_cast_fp16")];
            tensor<fp16, []> var_34837_to_fp16 = const()[name = tensor<string, []>("op_34837_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2927_cast_fp16 = mul(x = var_34836_cast_fp16, y = var_34837_to_fp16)[name = tensor<string, []>("aw_2927_cast_fp16")];
            tensor<string, []> var_34840_equation_0 = const()[name = tensor<string, []>("op_34840_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_34840_cast_fp16 = einsum(equation = var_34840_equation_0, values = (var_34682_cast_fp16, var_34599_cast_fp16))[name = tensor<string, []>("op_34840_cast_fp16")];
            tensor<fp16, []> var_34841_to_fp16 = const()[name = tensor<string, []>("op_34841_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2929_cast_fp16 = mul(x = var_34840_cast_fp16, y = var_34841_to_fp16)[name = tensor<string, []>("aw_2929_cast_fp16")];
            tensor<string, []> var_34844_equation_0 = const()[name = tensor<string, []>("op_34844_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_34844_cast_fp16 = einsum(equation = var_34844_equation_0, values = (var_34686_cast_fp16, var_34603_cast_fp16))[name = tensor<string, []>("op_34844_cast_fp16")];
            tensor<fp16, []> var_34845_to_fp16 = const()[name = tensor<string, []>("op_34845_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2931_cast_fp16 = mul(x = var_34844_cast_fp16, y = var_34845_to_fp16)[name = tensor<string, []>("aw_2931_cast_fp16")];
            tensor<string, []> var_34848_equation_0 = const()[name = tensor<string, []>("op_34848_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_34848_cast_fp16 = einsum(equation = var_34848_equation_0, values = (var_34690_cast_fp16, var_34607_cast_fp16))[name = tensor<string, []>("op_34848_cast_fp16")];
            tensor<fp16, []> var_34849_to_fp16 = const()[name = tensor<string, []>("op_34849_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2933_cast_fp16 = mul(x = var_34848_cast_fp16, y = var_34849_to_fp16)[name = tensor<string, []>("aw_2933_cast_fp16")];
            tensor<string, []> var_34852_equation_0 = const()[name = tensor<string, []>("op_34852_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_34852_cast_fp16 = einsum(equation = var_34852_equation_0, values = (var_34694_cast_fp16, var_34611_cast_fp16))[name = tensor<string, []>("op_34852_cast_fp16")];
            tensor<fp16, []> var_34853_to_fp16 = const()[name = tensor<string, []>("op_34853_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2935_cast_fp16 = mul(x = var_34852_cast_fp16, y = var_34853_to_fp16)[name = tensor<string, []>("aw_2935_cast_fp16")];
            tensor<string, []> var_34856_equation_0 = const()[name = tensor<string, []>("op_34856_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_34856_cast_fp16 = einsum(equation = var_34856_equation_0, values = (var_34698_cast_fp16, var_34615_cast_fp16))[name = tensor<string, []>("op_34856_cast_fp16")];
            tensor<fp16, []> var_34857_to_fp16 = const()[name = tensor<string, []>("op_34857_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2937_cast_fp16 = mul(x = var_34856_cast_fp16, y = var_34857_to_fp16)[name = tensor<string, []>("aw_2937_cast_fp16")];
            tensor<string, []> var_34860_equation_0 = const()[name = tensor<string, []>("op_34860_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_34860_cast_fp16 = einsum(equation = var_34860_equation_0, values = (var_34702_cast_fp16, var_34619_cast_fp16))[name = tensor<string, []>("op_34860_cast_fp16")];
            tensor<fp16, []> var_34861_to_fp16 = const()[name = tensor<string, []>("op_34861_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2939_cast_fp16 = mul(x = var_34860_cast_fp16, y = var_34861_to_fp16)[name = tensor<string, []>("aw_2939_cast_fp16")];
            tensor<string, []> var_34864_equation_0 = const()[name = tensor<string, []>("op_34864_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_34864_cast_fp16 = einsum(equation = var_34864_equation_0, values = (var_34706_cast_fp16, var_34623_cast_fp16))[name = tensor<string, []>("op_34864_cast_fp16")];
            tensor<fp16, []> var_34865_to_fp16 = const()[name = tensor<string, []>("op_34865_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2941_cast_fp16 = mul(x = var_34864_cast_fp16, y = var_34865_to_fp16)[name = tensor<string, []>("aw_2941_cast_fp16")];
            tensor<string, []> var_34868_equation_0 = const()[name = tensor<string, []>("op_34868_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_34868_cast_fp16 = einsum(equation = var_34868_equation_0, values = (var_34710_cast_fp16, var_34627_cast_fp16))[name = tensor<string, []>("op_34868_cast_fp16")];
            tensor<fp16, []> var_34869_to_fp16 = const()[name = tensor<string, []>("op_34869_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2943_cast_fp16 = mul(x = var_34868_cast_fp16, y = var_34869_to_fp16)[name = tensor<string, []>("aw_2943_cast_fp16")];
            tensor<string, []> var_34872_equation_0 = const()[name = tensor<string, []>("op_34872_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_34872_cast_fp16 = einsum(equation = var_34872_equation_0, values = (var_34714_cast_fp16, var_34631_cast_fp16))[name = tensor<string, []>("op_34872_cast_fp16")];
            tensor<fp16, []> var_34873_to_fp16 = const()[name = tensor<string, []>("op_34873_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2945_cast_fp16 = mul(x = var_34872_cast_fp16, y = var_34873_to_fp16)[name = tensor<string, []>("aw_2945_cast_fp16")];
            tensor<string, []> var_34876_equation_0 = const()[name = tensor<string, []>("op_34876_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_34876_cast_fp16 = einsum(equation = var_34876_equation_0, values = (var_34718_cast_fp16, var_34635_cast_fp16))[name = tensor<string, []>("op_34876_cast_fp16")];
            tensor<fp16, []> var_34877_to_fp16 = const()[name = tensor<string, []>("op_34877_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2947_cast_fp16 = mul(x = var_34876_cast_fp16, y = var_34877_to_fp16)[name = tensor<string, []>("aw_2947_cast_fp16")];
            tensor<string, []> var_34880_equation_0 = const()[name = tensor<string, []>("op_34880_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_34880_cast_fp16 = einsum(equation = var_34880_equation_0, values = (var_34722_cast_fp16, var_34639_cast_fp16))[name = tensor<string, []>("op_34880_cast_fp16")];
            tensor<fp16, []> var_34881_to_fp16 = const()[name = tensor<string, []>("op_34881_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2949_cast_fp16 = mul(x = var_34880_cast_fp16, y = var_34881_to_fp16)[name = tensor<string, []>("aw_2949_cast_fp16")];
            tensor<string, []> var_34884_equation_0 = const()[name = tensor<string, []>("op_34884_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_34884_cast_fp16 = einsum(equation = var_34884_equation_0, values = (var_34726_cast_fp16, var_34643_cast_fp16))[name = tensor<string, []>("op_34884_cast_fp16")];
            tensor<fp16, []> var_34885_to_fp16 = const()[name = tensor<string, []>("op_34885_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2951_cast_fp16 = mul(x = var_34884_cast_fp16, y = var_34885_to_fp16)[name = tensor<string, []>("aw_2951_cast_fp16")];
            tensor<string, []> var_34888_equation_0 = const()[name = tensor<string, []>("op_34888_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_34888_cast_fp16 = einsum(equation = var_34888_equation_0, values = (var_34730_cast_fp16, var_34647_cast_fp16))[name = tensor<string, []>("op_34888_cast_fp16")];
            tensor<fp16, []> var_34889_to_fp16 = const()[name = tensor<string, []>("op_34889_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2953_cast_fp16 = mul(x = var_34888_cast_fp16, y = var_34889_to_fp16)[name = tensor<string, []>("aw_2953_cast_fp16")];
            tensor<string, []> var_34892_equation_0 = const()[name = tensor<string, []>("op_34892_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_34892_cast_fp16 = einsum(equation = var_34892_equation_0, values = (var_34734_cast_fp16, var_34651_cast_fp16))[name = tensor<string, []>("op_34892_cast_fp16")];
            tensor<fp16, []> var_34893_to_fp16 = const()[name = tensor<string, []>("op_34893_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2955_cast_fp16 = mul(x = var_34892_cast_fp16, y = var_34893_to_fp16)[name = tensor<string, []>("aw_2955_cast_fp16")];
            tensor<string, []> var_34896_equation_0 = const()[name = tensor<string, []>("op_34896_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_34896_cast_fp16 = einsum(equation = var_34896_equation_0, values = (var_34738_cast_fp16, var_34655_cast_fp16))[name = tensor<string, []>("op_34896_cast_fp16")];
            tensor<fp16, []> var_34897_to_fp16 = const()[name = tensor<string, []>("op_34897_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2957_cast_fp16 = mul(x = var_34896_cast_fp16, y = var_34897_to_fp16)[name = tensor<string, []>("aw_2957_cast_fp16")];
            tensor<string, []> var_34900_equation_0 = const()[name = tensor<string, []>("op_34900_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_34900_cast_fp16 = einsum(equation = var_34900_equation_0, values = (var_34742_cast_fp16, var_34659_cast_fp16))[name = tensor<string, []>("op_34900_cast_fp16")];
            tensor<fp16, []> var_34901_to_fp16 = const()[name = tensor<string, []>("op_34901_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2959_cast_fp16 = mul(x = var_34900_cast_fp16, y = var_34901_to_fp16)[name = tensor<string, []>("aw_2959_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_34903_cast_fp16 = softmax(axis = var_30355, x = aw_2921_cast_fp16)[name = tensor<string, []>("op_34903_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_34904_cast_fp16 = softmax(axis = var_30355, x = aw_2923_cast_fp16)[name = tensor<string, []>("op_34904_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_34905_cast_fp16 = softmax(axis = var_30355, x = aw_2925_cast_fp16)[name = tensor<string, []>("op_34905_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_34906_cast_fp16 = softmax(axis = var_30355, x = aw_2927_cast_fp16)[name = tensor<string, []>("op_34906_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_34907_cast_fp16 = softmax(axis = var_30355, x = aw_2929_cast_fp16)[name = tensor<string, []>("op_34907_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_34908_cast_fp16 = softmax(axis = var_30355, x = aw_2931_cast_fp16)[name = tensor<string, []>("op_34908_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_34909_cast_fp16 = softmax(axis = var_30355, x = aw_2933_cast_fp16)[name = tensor<string, []>("op_34909_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_34910_cast_fp16 = softmax(axis = var_30355, x = aw_2935_cast_fp16)[name = tensor<string, []>("op_34910_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_34911_cast_fp16 = softmax(axis = var_30355, x = aw_2937_cast_fp16)[name = tensor<string, []>("op_34911_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_34912_cast_fp16 = softmax(axis = var_30355, x = aw_2939_cast_fp16)[name = tensor<string, []>("op_34912_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_34913_cast_fp16 = softmax(axis = var_30355, x = aw_2941_cast_fp16)[name = tensor<string, []>("op_34913_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_34914_cast_fp16 = softmax(axis = var_30355, x = aw_2943_cast_fp16)[name = tensor<string, []>("op_34914_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_34915_cast_fp16 = softmax(axis = var_30355, x = aw_2945_cast_fp16)[name = tensor<string, []>("op_34915_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_34916_cast_fp16 = softmax(axis = var_30355, x = aw_2947_cast_fp16)[name = tensor<string, []>("op_34916_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_34917_cast_fp16 = softmax(axis = var_30355, x = aw_2949_cast_fp16)[name = tensor<string, []>("op_34917_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_34918_cast_fp16 = softmax(axis = var_30355, x = aw_2951_cast_fp16)[name = tensor<string, []>("op_34918_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_34919_cast_fp16 = softmax(axis = var_30355, x = aw_2953_cast_fp16)[name = tensor<string, []>("op_34919_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_34920_cast_fp16 = softmax(axis = var_30355, x = aw_2955_cast_fp16)[name = tensor<string, []>("op_34920_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_34921_cast_fp16 = softmax(axis = var_30355, x = aw_2957_cast_fp16)[name = tensor<string, []>("op_34921_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_34922_cast_fp16 = softmax(axis = var_30355, x = aw_2959_cast_fp16)[name = tensor<string, []>("op_34922_cast_fp16")];
            tensor<string, []> var_34924_equation_0 = const()[name = tensor<string, []>("op_34924_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34924_cast_fp16 = einsum(equation = var_34924_equation_0, values = (var_34744_cast_fp16, var_34903_cast_fp16))[name = tensor<string, []>("op_34924_cast_fp16")];
            tensor<string, []> var_34926_equation_0 = const()[name = tensor<string, []>("op_34926_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34926_cast_fp16 = einsum(equation = var_34926_equation_0, values = (var_34748_cast_fp16, var_34904_cast_fp16))[name = tensor<string, []>("op_34926_cast_fp16")];
            tensor<string, []> var_34928_equation_0 = const()[name = tensor<string, []>("op_34928_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34928_cast_fp16 = einsum(equation = var_34928_equation_0, values = (var_34752_cast_fp16, var_34905_cast_fp16))[name = tensor<string, []>("op_34928_cast_fp16")];
            tensor<string, []> var_34930_equation_0 = const()[name = tensor<string, []>("op_34930_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34930_cast_fp16 = einsum(equation = var_34930_equation_0, values = (var_34756_cast_fp16, var_34906_cast_fp16))[name = tensor<string, []>("op_34930_cast_fp16")];
            tensor<string, []> var_34932_equation_0 = const()[name = tensor<string, []>("op_34932_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34932_cast_fp16 = einsum(equation = var_34932_equation_0, values = (var_34760_cast_fp16, var_34907_cast_fp16))[name = tensor<string, []>("op_34932_cast_fp16")];
            tensor<string, []> var_34934_equation_0 = const()[name = tensor<string, []>("op_34934_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34934_cast_fp16 = einsum(equation = var_34934_equation_0, values = (var_34764_cast_fp16, var_34908_cast_fp16))[name = tensor<string, []>("op_34934_cast_fp16")];
            tensor<string, []> var_34936_equation_0 = const()[name = tensor<string, []>("op_34936_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34936_cast_fp16 = einsum(equation = var_34936_equation_0, values = (var_34768_cast_fp16, var_34909_cast_fp16))[name = tensor<string, []>("op_34936_cast_fp16")];
            tensor<string, []> var_34938_equation_0 = const()[name = tensor<string, []>("op_34938_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34938_cast_fp16 = einsum(equation = var_34938_equation_0, values = (var_34772_cast_fp16, var_34910_cast_fp16))[name = tensor<string, []>("op_34938_cast_fp16")];
            tensor<string, []> var_34940_equation_0 = const()[name = tensor<string, []>("op_34940_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34940_cast_fp16 = einsum(equation = var_34940_equation_0, values = (var_34776_cast_fp16, var_34911_cast_fp16))[name = tensor<string, []>("op_34940_cast_fp16")];
            tensor<string, []> var_34942_equation_0 = const()[name = tensor<string, []>("op_34942_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34942_cast_fp16 = einsum(equation = var_34942_equation_0, values = (var_34780_cast_fp16, var_34912_cast_fp16))[name = tensor<string, []>("op_34942_cast_fp16")];
            tensor<string, []> var_34944_equation_0 = const()[name = tensor<string, []>("op_34944_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34944_cast_fp16 = einsum(equation = var_34944_equation_0, values = (var_34784_cast_fp16, var_34913_cast_fp16))[name = tensor<string, []>("op_34944_cast_fp16")];
            tensor<string, []> var_34946_equation_0 = const()[name = tensor<string, []>("op_34946_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34946_cast_fp16 = einsum(equation = var_34946_equation_0, values = (var_34788_cast_fp16, var_34914_cast_fp16))[name = tensor<string, []>("op_34946_cast_fp16")];
            tensor<string, []> var_34948_equation_0 = const()[name = tensor<string, []>("op_34948_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34948_cast_fp16 = einsum(equation = var_34948_equation_0, values = (var_34792_cast_fp16, var_34915_cast_fp16))[name = tensor<string, []>("op_34948_cast_fp16")];
            tensor<string, []> var_34950_equation_0 = const()[name = tensor<string, []>("op_34950_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34950_cast_fp16 = einsum(equation = var_34950_equation_0, values = (var_34796_cast_fp16, var_34916_cast_fp16))[name = tensor<string, []>("op_34950_cast_fp16")];
            tensor<string, []> var_34952_equation_0 = const()[name = tensor<string, []>("op_34952_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34952_cast_fp16 = einsum(equation = var_34952_equation_0, values = (var_34800_cast_fp16, var_34917_cast_fp16))[name = tensor<string, []>("op_34952_cast_fp16")];
            tensor<string, []> var_34954_equation_0 = const()[name = tensor<string, []>("op_34954_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34954_cast_fp16 = einsum(equation = var_34954_equation_0, values = (var_34804_cast_fp16, var_34918_cast_fp16))[name = tensor<string, []>("op_34954_cast_fp16")];
            tensor<string, []> var_34956_equation_0 = const()[name = tensor<string, []>("op_34956_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34956_cast_fp16 = einsum(equation = var_34956_equation_0, values = (var_34808_cast_fp16, var_34919_cast_fp16))[name = tensor<string, []>("op_34956_cast_fp16")];
            tensor<string, []> var_34958_equation_0 = const()[name = tensor<string, []>("op_34958_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34958_cast_fp16 = einsum(equation = var_34958_equation_0, values = (var_34812_cast_fp16, var_34920_cast_fp16))[name = tensor<string, []>("op_34958_cast_fp16")];
            tensor<string, []> var_34960_equation_0 = const()[name = tensor<string, []>("op_34960_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34960_cast_fp16 = einsum(equation = var_34960_equation_0, values = (var_34816_cast_fp16, var_34921_cast_fp16))[name = tensor<string, []>("op_34960_cast_fp16")];
            tensor<string, []> var_34962_equation_0 = const()[name = tensor<string, []>("op_34962_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34962_cast_fp16 = einsum(equation = var_34962_equation_0, values = (var_34820_cast_fp16, var_34922_cast_fp16))[name = tensor<string, []>("op_34962_cast_fp16")];
            tensor<bool, []> input_473_interleave_0 = const()[name = tensor<string, []>("input_473_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_473_cast_fp16 = concat(axis = var_30355, interleave = input_473_interleave_0, values = (var_34924_cast_fp16, var_34926_cast_fp16, var_34928_cast_fp16, var_34930_cast_fp16, var_34932_cast_fp16, var_34934_cast_fp16, var_34936_cast_fp16, var_34938_cast_fp16, var_34940_cast_fp16, var_34942_cast_fp16, var_34944_cast_fp16, var_34946_cast_fp16, var_34948_cast_fp16, var_34950_cast_fp16, var_34952_cast_fp16, var_34954_cast_fp16, var_34956_cast_fp16, var_34958_cast_fp16, var_34960_cast_fp16, var_34962_cast_fp16))[name = tensor<string, []>("input_473_cast_fp16")];
            tensor<string, []> var_34972_pad_type_0 = const()[name = tensor<string, []>("op_34972_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_34972_strides_0 = const()[name = tensor<string, []>("op_34972_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_34972_pad_0 = const()[name = tensor<string, []>("op_34972_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_34972_dilations_0 = const()[name = tensor<string, []>("op_34972_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_34972_groups_0 = const()[name = tensor<string, []>("op_34972_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_4_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118197376))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119426240))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_4_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_4_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_4_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119426432)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_34972_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_4_attn2_to_out_0_bias_to_fp16, dilations = var_34972_dilations_0, groups = var_34972_groups_0, pad = var_34972_pad_0, pad_type = var_34972_pad_type_0, strides = var_34972_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_4_attn2_to_out_0_weight_to_fp16_palettized, x = input_473_cast_fp16)[name = tensor<string, []>("op_34972_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_233_cast_fp16 = add(x = var_34972_cast_fp16, y = inputs_231_cast_fp16)[name = tensor<string, []>("inputs_233_cast_fp16")];
            tensor<int32, [1]> input_475_axes_0 = const()[name = tensor<string, []>("input_475_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> input_475_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_475_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119429056)))];
            tensor<fp16, [1280]> input_475_beta_0_to_fp16 = const()[name = tensor<string, []>("input_475_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119431680)))];
            tensor<fp16, []> var_34982_to_fp16 = const()[name = tensor<string, []>("op_34982_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> input_475_cast_fp16 = layer_norm(axes = input_475_axes_0, beta = input_475_beta_0_to_fp16, epsilon = var_34982_to_fp16, gamma = input_475_gamma_0_to_fp16, x = inputs_233_cast_fp16)[name = tensor<string, []>("input_475_cast_fp16")];
            tensor<string, []> var_35002_pad_type_0 = const()[name = tensor<string, []>("op_35002_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_35002_strides_0 = const()[name = tensor<string, []>("op_35002_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_35002_pad_0 = const()[name = tensor<string, []>("op_35002_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_35002_dilations_0 = const()[name = tensor<string, []>("op_35002_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_35002_groups_0 = const()[name = tensor<string, []>("op_35002_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_4_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [9830400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119434304))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129264768))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_4_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([10240, 1280, 1, 1])];
            tensor<fp16, [10240]> up_blocks_0_attentions_0_transformer_blocks_4_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_4_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129264960)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_35002_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_4_ff_net_0_proj_bias_to_fp16, dilations = var_35002_dilations_0, groups = var_35002_groups_0, pad = var_35002_pad_0, pad_type = var_35002_pad_type_0, strides = var_35002_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_4_ff_net_0_proj_weight_to_fp16_palettized, x = input_475_cast_fp16)[name = tensor<string, []>("op_35002_cast_fp16")];
            tensor<int32, [2]> var_35003_split_sizes_0 = const()[name = tensor<string, []>("op_35003_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_35003_axis_0 = const()[name = tensor<string, []>("op_35003_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_35003_cast_fp16_0, tensor<fp16, [2, 5120, 1, 1024]> var_35003_cast_fp16_1 = split(axis = var_35003_axis_0, split_sizes = var_35003_split_sizes_0, x = var_35002_cast_fp16)[name = tensor<string, []>("op_35003_cast_fp16")];
            tensor<string, []> var_35005_mode_0 = const()[name = tensor<string, []>("op_35005_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_35005_cast_fp16 = gelu(mode = var_35005_mode_0, x = var_35003_cast_fp16_1)[name = tensor<string, []>("op_35005_cast_fp16")];
            tensor<fp16, [2, 5120, 1, 1024]> input_477_cast_fp16 = mul(x = var_35003_cast_fp16_0, y = var_35005_cast_fp16)[name = tensor<string, []>("input_477_cast_fp16")];
            tensor<string, []> var_35013_pad_type_0 = const()[name = tensor<string, []>("op_35013_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_35013_strides_0 = const()[name = tensor<string, []>("op_35013_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_35013_pad_0 = const()[name = tensor<string, []>("op_35013_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_35013_dilations_0 = const()[name = tensor<string, []>("op_35013_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_35013_groups_0 = const()[name = tensor<string, []>("op_35013_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_4_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [4915200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129285504))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134200768))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_4_ff_net_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 5120, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_4_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_4_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134200960)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_35013_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_4_ff_net_2_bias_to_fp16, dilations = var_35013_dilations_0, groups = var_35013_groups_0, pad = var_35013_pad_0, pad_type = var_35013_pad_type_0, strides = var_35013_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_4_ff_net_2_weight_to_fp16_palettized, x = input_477_cast_fp16)[name = tensor<string, []>("op_35013_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_235_cast_fp16 = add(x = var_35013_cast_fp16, y = inputs_233_cast_fp16)[name = tensor<string, []>("inputs_235_cast_fp16")];
            tensor<int32, [1]> hidden_states_317_axes_0 = const()[name = tensor<string, []>("hidden_states_317_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_317_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_317_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134203584)))];
            tensor<fp16, [1280]> hidden_states_317_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_317_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134206208)))];
            tensor<fp16, []> var_35029_to_fp16 = const()[name = tensor<string, []>("op_35029_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_317_cast_fp16 = layer_norm(axes = hidden_states_317_axes_0, beta = hidden_states_317_beta_0_to_fp16, epsilon = var_35029_to_fp16, gamma = hidden_states_317_gamma_0_to_fp16, x = inputs_235_cast_fp16)[name = tensor<string, []>("hidden_states_317_cast_fp16")];
            tensor<string, []> q_157_pad_type_0 = const()[name = tensor<string, []>("q_157_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_157_strides_0 = const()[name = tensor<string, []>("q_157_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_157_pad_0 = const()[name = tensor<string, []>("q_157_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_157_dilations_0 = const()[name = tensor<string, []>("q_157_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_157_groups_0 = const()[name = tensor<string, []>("q_157_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_5_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134208832))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135437696))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_5_attn1_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_157_cast_fp16 = conv(dilations = q_157_dilations_0, groups = q_157_groups_0, pad = q_157_pad_0, pad_type = q_157_pad_type_0, strides = q_157_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_5_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_317_cast_fp16)[name = tensor<string, []>("q_157_cast_fp16")];
            tensor<string, []> k_313_pad_type_0 = const()[name = tensor<string, []>("k_313_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_313_strides_0 = const()[name = tensor<string, []>("k_313_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_313_pad_0 = const()[name = tensor<string, []>("k_313_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_313_dilations_0 = const()[name = tensor<string, []>("k_313_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_313_groups_0 = const()[name = tensor<string, []>("k_313_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_5_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135437888))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(136666752))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_5_attn1_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> k_313_cast_fp16 = conv(dilations = k_313_dilations_0, groups = k_313_groups_0, pad = k_313_pad_0, pad_type = k_313_pad_type_0, strides = k_313_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_5_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_317_cast_fp16)[name = tensor<string, []>("k_313_cast_fp16")];
            tensor<string, []> v_157_pad_type_0 = const()[name = tensor<string, []>("v_157_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_157_strides_0 = const()[name = tensor<string, []>("v_157_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_157_pad_0 = const()[name = tensor<string, []>("v_157_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_157_dilations_0 = const()[name = tensor<string, []>("v_157_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_157_groups_0 = const()[name = tensor<string, []>("v_157_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_5_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(136666944))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(137895808))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_5_attn1_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> v_157_cast_fp16 = conv(dilations = v_157_dilations_0, groups = v_157_groups_0, pad = v_157_pad_0, pad_type = v_157_pad_type_0, strides = v_157_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_5_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_317_cast_fp16)[name = tensor<string, []>("v_157_cast_fp16")];
            tensor<int32, [4]> var_35062_begin_0 = const()[name = tensor<string, []>("op_35062_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_35062_end_0 = const()[name = tensor<string, []>("op_35062_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_35062_end_mask_0 = const()[name = tensor<string, []>("op_35062_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35062_cast_fp16 = slice_by_index(begin = var_35062_begin_0, end = var_35062_end_0, end_mask = var_35062_end_mask_0, x = q_157_cast_fp16)[name = tensor<string, []>("op_35062_cast_fp16")];
            tensor<int32, [4]> var_35066_begin_0 = const()[name = tensor<string, []>("op_35066_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_35066_end_0 = const()[name = tensor<string, []>("op_35066_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_35066_end_mask_0 = const()[name = tensor<string, []>("op_35066_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35066_cast_fp16 = slice_by_index(begin = var_35066_begin_0, end = var_35066_end_0, end_mask = var_35066_end_mask_0, x = q_157_cast_fp16)[name = tensor<string, []>("op_35066_cast_fp16")];
            tensor<int32, [4]> var_35070_begin_0 = const()[name = tensor<string, []>("op_35070_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_35070_end_0 = const()[name = tensor<string, []>("op_35070_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_35070_end_mask_0 = const()[name = tensor<string, []>("op_35070_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35070_cast_fp16 = slice_by_index(begin = var_35070_begin_0, end = var_35070_end_0, end_mask = var_35070_end_mask_0, x = q_157_cast_fp16)[name = tensor<string, []>("op_35070_cast_fp16")];
            tensor<int32, [4]> var_35074_begin_0 = const()[name = tensor<string, []>("op_35074_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_35074_end_0 = const()[name = tensor<string, []>("op_35074_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_35074_end_mask_0 = const()[name = tensor<string, []>("op_35074_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35074_cast_fp16 = slice_by_index(begin = var_35074_begin_0, end = var_35074_end_0, end_mask = var_35074_end_mask_0, x = q_157_cast_fp16)[name = tensor<string, []>("op_35074_cast_fp16")];
            tensor<int32, [4]> var_35078_begin_0 = const()[name = tensor<string, []>("op_35078_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_35078_end_0 = const()[name = tensor<string, []>("op_35078_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_35078_end_mask_0 = const()[name = tensor<string, []>("op_35078_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35078_cast_fp16 = slice_by_index(begin = var_35078_begin_0, end = var_35078_end_0, end_mask = var_35078_end_mask_0, x = q_157_cast_fp16)[name = tensor<string, []>("op_35078_cast_fp16")];
            tensor<int32, [4]> var_35082_begin_0 = const()[name = tensor<string, []>("op_35082_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_35082_end_0 = const()[name = tensor<string, []>("op_35082_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_35082_end_mask_0 = const()[name = tensor<string, []>("op_35082_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35082_cast_fp16 = slice_by_index(begin = var_35082_begin_0, end = var_35082_end_0, end_mask = var_35082_end_mask_0, x = q_157_cast_fp16)[name = tensor<string, []>("op_35082_cast_fp16")];
            tensor<int32, [4]> var_35086_begin_0 = const()[name = tensor<string, []>("op_35086_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_35086_end_0 = const()[name = tensor<string, []>("op_35086_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_35086_end_mask_0 = const()[name = tensor<string, []>("op_35086_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35086_cast_fp16 = slice_by_index(begin = var_35086_begin_0, end = var_35086_end_0, end_mask = var_35086_end_mask_0, x = q_157_cast_fp16)[name = tensor<string, []>("op_35086_cast_fp16")];
            tensor<int32, [4]> var_35090_begin_0 = const()[name = tensor<string, []>("op_35090_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_35090_end_0 = const()[name = tensor<string, []>("op_35090_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_35090_end_mask_0 = const()[name = tensor<string, []>("op_35090_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35090_cast_fp16 = slice_by_index(begin = var_35090_begin_0, end = var_35090_end_0, end_mask = var_35090_end_mask_0, x = q_157_cast_fp16)[name = tensor<string, []>("op_35090_cast_fp16")];
            tensor<int32, [4]> var_35094_begin_0 = const()[name = tensor<string, []>("op_35094_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_35094_end_0 = const()[name = tensor<string, []>("op_35094_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_35094_end_mask_0 = const()[name = tensor<string, []>("op_35094_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35094_cast_fp16 = slice_by_index(begin = var_35094_begin_0, end = var_35094_end_0, end_mask = var_35094_end_mask_0, x = q_157_cast_fp16)[name = tensor<string, []>("op_35094_cast_fp16")];
            tensor<int32, [4]> var_35098_begin_0 = const()[name = tensor<string, []>("op_35098_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_35098_end_0 = const()[name = tensor<string, []>("op_35098_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_35098_end_mask_0 = const()[name = tensor<string, []>("op_35098_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35098_cast_fp16 = slice_by_index(begin = var_35098_begin_0, end = var_35098_end_0, end_mask = var_35098_end_mask_0, x = q_157_cast_fp16)[name = tensor<string, []>("op_35098_cast_fp16")];
            tensor<int32, [4]> var_35102_begin_0 = const()[name = tensor<string, []>("op_35102_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_35102_end_0 = const()[name = tensor<string, []>("op_35102_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_35102_end_mask_0 = const()[name = tensor<string, []>("op_35102_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35102_cast_fp16 = slice_by_index(begin = var_35102_begin_0, end = var_35102_end_0, end_mask = var_35102_end_mask_0, x = q_157_cast_fp16)[name = tensor<string, []>("op_35102_cast_fp16")];
            tensor<int32, [4]> var_35106_begin_0 = const()[name = tensor<string, []>("op_35106_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_35106_end_0 = const()[name = tensor<string, []>("op_35106_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_35106_end_mask_0 = const()[name = tensor<string, []>("op_35106_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35106_cast_fp16 = slice_by_index(begin = var_35106_begin_0, end = var_35106_end_0, end_mask = var_35106_end_mask_0, x = q_157_cast_fp16)[name = tensor<string, []>("op_35106_cast_fp16")];
            tensor<int32, [4]> var_35110_begin_0 = const()[name = tensor<string, []>("op_35110_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_35110_end_0 = const()[name = tensor<string, []>("op_35110_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_35110_end_mask_0 = const()[name = tensor<string, []>("op_35110_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35110_cast_fp16 = slice_by_index(begin = var_35110_begin_0, end = var_35110_end_0, end_mask = var_35110_end_mask_0, x = q_157_cast_fp16)[name = tensor<string, []>("op_35110_cast_fp16")];
            tensor<int32, [4]> var_35114_begin_0 = const()[name = tensor<string, []>("op_35114_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_35114_end_0 = const()[name = tensor<string, []>("op_35114_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_35114_end_mask_0 = const()[name = tensor<string, []>("op_35114_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35114_cast_fp16 = slice_by_index(begin = var_35114_begin_0, end = var_35114_end_0, end_mask = var_35114_end_mask_0, x = q_157_cast_fp16)[name = tensor<string, []>("op_35114_cast_fp16")];
            tensor<int32, [4]> var_35118_begin_0 = const()[name = tensor<string, []>("op_35118_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_35118_end_0 = const()[name = tensor<string, []>("op_35118_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_35118_end_mask_0 = const()[name = tensor<string, []>("op_35118_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35118_cast_fp16 = slice_by_index(begin = var_35118_begin_0, end = var_35118_end_0, end_mask = var_35118_end_mask_0, x = q_157_cast_fp16)[name = tensor<string, []>("op_35118_cast_fp16")];
            tensor<int32, [4]> var_35122_begin_0 = const()[name = tensor<string, []>("op_35122_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_35122_end_0 = const()[name = tensor<string, []>("op_35122_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_35122_end_mask_0 = const()[name = tensor<string, []>("op_35122_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35122_cast_fp16 = slice_by_index(begin = var_35122_begin_0, end = var_35122_end_0, end_mask = var_35122_end_mask_0, x = q_157_cast_fp16)[name = tensor<string, []>("op_35122_cast_fp16")];
            tensor<int32, [4]> var_35126_begin_0 = const()[name = tensor<string, []>("op_35126_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_35126_end_0 = const()[name = tensor<string, []>("op_35126_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_35126_end_mask_0 = const()[name = tensor<string, []>("op_35126_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35126_cast_fp16 = slice_by_index(begin = var_35126_begin_0, end = var_35126_end_0, end_mask = var_35126_end_mask_0, x = q_157_cast_fp16)[name = tensor<string, []>("op_35126_cast_fp16")];
            tensor<int32, [4]> var_35130_begin_0 = const()[name = tensor<string, []>("op_35130_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_35130_end_0 = const()[name = tensor<string, []>("op_35130_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_35130_end_mask_0 = const()[name = tensor<string, []>("op_35130_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35130_cast_fp16 = slice_by_index(begin = var_35130_begin_0, end = var_35130_end_0, end_mask = var_35130_end_mask_0, x = q_157_cast_fp16)[name = tensor<string, []>("op_35130_cast_fp16")];
            tensor<int32, [4]> var_35134_begin_0 = const()[name = tensor<string, []>("op_35134_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_35134_end_0 = const()[name = tensor<string, []>("op_35134_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_35134_end_mask_0 = const()[name = tensor<string, []>("op_35134_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35134_cast_fp16 = slice_by_index(begin = var_35134_begin_0, end = var_35134_end_0, end_mask = var_35134_end_mask_0, x = q_157_cast_fp16)[name = tensor<string, []>("op_35134_cast_fp16")];
            tensor<int32, [4]> var_35138_begin_0 = const()[name = tensor<string, []>("op_35138_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_35138_end_0 = const()[name = tensor<string, []>("op_35138_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_35138_end_mask_0 = const()[name = tensor<string, []>("op_35138_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35138_cast_fp16 = slice_by_index(begin = var_35138_begin_0, end = var_35138_end_0, end_mask = var_35138_end_mask_0, x = q_157_cast_fp16)[name = tensor<string, []>("op_35138_cast_fp16")];
            tensor<int32, [4]> k_315_perm_0 = const()[name = tensor<string, []>("k_315_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_35145_begin_0 = const()[name = tensor<string, []>("op_35145_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_35145_end_0 = const()[name = tensor<string, []>("op_35145_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_35145_end_mask_0 = const()[name = tensor<string, []>("op_35145_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> k_315_cast_fp16 = transpose(perm = k_315_perm_0, x = k_313_cast_fp16)[name = tensor<string, []>("transpose_61")];
            tensor<fp16, [2, 1024, 1, 64]> var_35145_cast_fp16 = slice_by_index(begin = var_35145_begin_0, end = var_35145_end_0, end_mask = var_35145_end_mask_0, x = k_315_cast_fp16)[name = tensor<string, []>("op_35145_cast_fp16")];
            tensor<int32, [4]> var_35149_begin_0 = const()[name = tensor<string, []>("op_35149_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_35149_end_0 = const()[name = tensor<string, []>("op_35149_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_35149_end_mask_0 = const()[name = tensor<string, []>("op_35149_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_35149_cast_fp16 = slice_by_index(begin = var_35149_begin_0, end = var_35149_end_0, end_mask = var_35149_end_mask_0, x = k_315_cast_fp16)[name = tensor<string, []>("op_35149_cast_fp16")];
            tensor<int32, [4]> var_35153_begin_0 = const()[name = tensor<string, []>("op_35153_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_35153_end_0 = const()[name = tensor<string, []>("op_35153_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_35153_end_mask_0 = const()[name = tensor<string, []>("op_35153_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_35153_cast_fp16 = slice_by_index(begin = var_35153_begin_0, end = var_35153_end_0, end_mask = var_35153_end_mask_0, x = k_315_cast_fp16)[name = tensor<string, []>("op_35153_cast_fp16")];
            tensor<int32, [4]> var_35157_begin_0 = const()[name = tensor<string, []>("op_35157_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_35157_end_0 = const()[name = tensor<string, []>("op_35157_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_35157_end_mask_0 = const()[name = tensor<string, []>("op_35157_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_35157_cast_fp16 = slice_by_index(begin = var_35157_begin_0, end = var_35157_end_0, end_mask = var_35157_end_mask_0, x = k_315_cast_fp16)[name = tensor<string, []>("op_35157_cast_fp16")];
            tensor<int32, [4]> var_35161_begin_0 = const()[name = tensor<string, []>("op_35161_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_35161_end_0 = const()[name = tensor<string, []>("op_35161_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_35161_end_mask_0 = const()[name = tensor<string, []>("op_35161_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_35161_cast_fp16 = slice_by_index(begin = var_35161_begin_0, end = var_35161_end_0, end_mask = var_35161_end_mask_0, x = k_315_cast_fp16)[name = tensor<string, []>("op_35161_cast_fp16")];
            tensor<int32, [4]> var_35165_begin_0 = const()[name = tensor<string, []>("op_35165_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_35165_end_0 = const()[name = tensor<string, []>("op_35165_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_35165_end_mask_0 = const()[name = tensor<string, []>("op_35165_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_35165_cast_fp16 = slice_by_index(begin = var_35165_begin_0, end = var_35165_end_0, end_mask = var_35165_end_mask_0, x = k_315_cast_fp16)[name = tensor<string, []>("op_35165_cast_fp16")];
            tensor<int32, [4]> var_35169_begin_0 = const()[name = tensor<string, []>("op_35169_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_35169_end_0 = const()[name = tensor<string, []>("op_35169_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_35169_end_mask_0 = const()[name = tensor<string, []>("op_35169_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_35169_cast_fp16 = slice_by_index(begin = var_35169_begin_0, end = var_35169_end_0, end_mask = var_35169_end_mask_0, x = k_315_cast_fp16)[name = tensor<string, []>("op_35169_cast_fp16")];
            tensor<int32, [4]> var_35173_begin_0 = const()[name = tensor<string, []>("op_35173_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_35173_end_0 = const()[name = tensor<string, []>("op_35173_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_35173_end_mask_0 = const()[name = tensor<string, []>("op_35173_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_35173_cast_fp16 = slice_by_index(begin = var_35173_begin_0, end = var_35173_end_0, end_mask = var_35173_end_mask_0, x = k_315_cast_fp16)[name = tensor<string, []>("op_35173_cast_fp16")];
            tensor<int32, [4]> var_35177_begin_0 = const()[name = tensor<string, []>("op_35177_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_35177_end_0 = const()[name = tensor<string, []>("op_35177_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_35177_end_mask_0 = const()[name = tensor<string, []>("op_35177_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_35177_cast_fp16 = slice_by_index(begin = var_35177_begin_0, end = var_35177_end_0, end_mask = var_35177_end_mask_0, x = k_315_cast_fp16)[name = tensor<string, []>("op_35177_cast_fp16")];
            tensor<int32, [4]> var_35181_begin_0 = const()[name = tensor<string, []>("op_35181_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_35181_end_0 = const()[name = tensor<string, []>("op_35181_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_35181_end_mask_0 = const()[name = tensor<string, []>("op_35181_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_35181_cast_fp16 = slice_by_index(begin = var_35181_begin_0, end = var_35181_end_0, end_mask = var_35181_end_mask_0, x = k_315_cast_fp16)[name = tensor<string, []>("op_35181_cast_fp16")];
            tensor<int32, [4]> var_35185_begin_0 = const()[name = tensor<string, []>("op_35185_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_35185_end_0 = const()[name = tensor<string, []>("op_35185_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_35185_end_mask_0 = const()[name = tensor<string, []>("op_35185_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_35185_cast_fp16 = slice_by_index(begin = var_35185_begin_0, end = var_35185_end_0, end_mask = var_35185_end_mask_0, x = k_315_cast_fp16)[name = tensor<string, []>("op_35185_cast_fp16")];
            tensor<int32, [4]> var_35189_begin_0 = const()[name = tensor<string, []>("op_35189_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_35189_end_0 = const()[name = tensor<string, []>("op_35189_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_35189_end_mask_0 = const()[name = tensor<string, []>("op_35189_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_35189_cast_fp16 = slice_by_index(begin = var_35189_begin_0, end = var_35189_end_0, end_mask = var_35189_end_mask_0, x = k_315_cast_fp16)[name = tensor<string, []>("op_35189_cast_fp16")];
            tensor<int32, [4]> var_35193_begin_0 = const()[name = tensor<string, []>("op_35193_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_35193_end_0 = const()[name = tensor<string, []>("op_35193_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_35193_end_mask_0 = const()[name = tensor<string, []>("op_35193_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_35193_cast_fp16 = slice_by_index(begin = var_35193_begin_0, end = var_35193_end_0, end_mask = var_35193_end_mask_0, x = k_315_cast_fp16)[name = tensor<string, []>("op_35193_cast_fp16")];
            tensor<int32, [4]> var_35197_begin_0 = const()[name = tensor<string, []>("op_35197_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_35197_end_0 = const()[name = tensor<string, []>("op_35197_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_35197_end_mask_0 = const()[name = tensor<string, []>("op_35197_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_35197_cast_fp16 = slice_by_index(begin = var_35197_begin_0, end = var_35197_end_0, end_mask = var_35197_end_mask_0, x = k_315_cast_fp16)[name = tensor<string, []>("op_35197_cast_fp16")];
            tensor<int32, [4]> var_35201_begin_0 = const()[name = tensor<string, []>("op_35201_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_35201_end_0 = const()[name = tensor<string, []>("op_35201_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_35201_end_mask_0 = const()[name = tensor<string, []>("op_35201_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_35201_cast_fp16 = slice_by_index(begin = var_35201_begin_0, end = var_35201_end_0, end_mask = var_35201_end_mask_0, x = k_315_cast_fp16)[name = tensor<string, []>("op_35201_cast_fp16")];
            tensor<int32, [4]> var_35205_begin_0 = const()[name = tensor<string, []>("op_35205_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_35205_end_0 = const()[name = tensor<string, []>("op_35205_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_35205_end_mask_0 = const()[name = tensor<string, []>("op_35205_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_35205_cast_fp16 = slice_by_index(begin = var_35205_begin_0, end = var_35205_end_0, end_mask = var_35205_end_mask_0, x = k_315_cast_fp16)[name = tensor<string, []>("op_35205_cast_fp16")];
            tensor<int32, [4]> var_35209_begin_0 = const()[name = tensor<string, []>("op_35209_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_35209_end_0 = const()[name = tensor<string, []>("op_35209_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_35209_end_mask_0 = const()[name = tensor<string, []>("op_35209_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_35209_cast_fp16 = slice_by_index(begin = var_35209_begin_0, end = var_35209_end_0, end_mask = var_35209_end_mask_0, x = k_315_cast_fp16)[name = tensor<string, []>("op_35209_cast_fp16")];
            tensor<int32, [4]> var_35213_begin_0 = const()[name = tensor<string, []>("op_35213_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_35213_end_0 = const()[name = tensor<string, []>("op_35213_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_35213_end_mask_0 = const()[name = tensor<string, []>("op_35213_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_35213_cast_fp16 = slice_by_index(begin = var_35213_begin_0, end = var_35213_end_0, end_mask = var_35213_end_mask_0, x = k_315_cast_fp16)[name = tensor<string, []>("op_35213_cast_fp16")];
            tensor<int32, [4]> var_35217_begin_0 = const()[name = tensor<string, []>("op_35217_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_35217_end_0 = const()[name = tensor<string, []>("op_35217_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_35217_end_mask_0 = const()[name = tensor<string, []>("op_35217_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_35217_cast_fp16 = slice_by_index(begin = var_35217_begin_0, end = var_35217_end_0, end_mask = var_35217_end_mask_0, x = k_315_cast_fp16)[name = tensor<string, []>("op_35217_cast_fp16")];
            tensor<int32, [4]> var_35221_begin_0 = const()[name = tensor<string, []>("op_35221_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_35221_end_0 = const()[name = tensor<string, []>("op_35221_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_35221_end_mask_0 = const()[name = tensor<string, []>("op_35221_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_35221_cast_fp16 = slice_by_index(begin = var_35221_begin_0, end = var_35221_end_0, end_mask = var_35221_end_mask_0, x = k_315_cast_fp16)[name = tensor<string, []>("op_35221_cast_fp16")];
            tensor<int32, [4]> var_35223_begin_0 = const()[name = tensor<string, []>("op_35223_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_35223_end_0 = const()[name = tensor<string, []>("op_35223_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_35223_end_mask_0 = const()[name = tensor<string, []>("op_35223_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35223_cast_fp16 = slice_by_index(begin = var_35223_begin_0, end = var_35223_end_0, end_mask = var_35223_end_mask_0, x = v_157_cast_fp16)[name = tensor<string, []>("op_35223_cast_fp16")];
            tensor<int32, [4]> var_35227_begin_0 = const()[name = tensor<string, []>("op_35227_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_35227_end_0 = const()[name = tensor<string, []>("op_35227_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_35227_end_mask_0 = const()[name = tensor<string, []>("op_35227_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35227_cast_fp16 = slice_by_index(begin = var_35227_begin_0, end = var_35227_end_0, end_mask = var_35227_end_mask_0, x = v_157_cast_fp16)[name = tensor<string, []>("op_35227_cast_fp16")];
            tensor<int32, [4]> var_35231_begin_0 = const()[name = tensor<string, []>("op_35231_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_35231_end_0 = const()[name = tensor<string, []>("op_35231_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_35231_end_mask_0 = const()[name = tensor<string, []>("op_35231_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35231_cast_fp16 = slice_by_index(begin = var_35231_begin_0, end = var_35231_end_0, end_mask = var_35231_end_mask_0, x = v_157_cast_fp16)[name = tensor<string, []>("op_35231_cast_fp16")];
            tensor<int32, [4]> var_35235_begin_0 = const()[name = tensor<string, []>("op_35235_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_35235_end_0 = const()[name = tensor<string, []>("op_35235_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_35235_end_mask_0 = const()[name = tensor<string, []>("op_35235_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35235_cast_fp16 = slice_by_index(begin = var_35235_begin_0, end = var_35235_end_0, end_mask = var_35235_end_mask_0, x = v_157_cast_fp16)[name = tensor<string, []>("op_35235_cast_fp16")];
            tensor<int32, [4]> var_35239_begin_0 = const()[name = tensor<string, []>("op_35239_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_35239_end_0 = const()[name = tensor<string, []>("op_35239_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_35239_end_mask_0 = const()[name = tensor<string, []>("op_35239_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35239_cast_fp16 = slice_by_index(begin = var_35239_begin_0, end = var_35239_end_0, end_mask = var_35239_end_mask_0, x = v_157_cast_fp16)[name = tensor<string, []>("op_35239_cast_fp16")];
            tensor<int32, [4]> var_35243_begin_0 = const()[name = tensor<string, []>("op_35243_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_35243_end_0 = const()[name = tensor<string, []>("op_35243_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_35243_end_mask_0 = const()[name = tensor<string, []>("op_35243_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35243_cast_fp16 = slice_by_index(begin = var_35243_begin_0, end = var_35243_end_0, end_mask = var_35243_end_mask_0, x = v_157_cast_fp16)[name = tensor<string, []>("op_35243_cast_fp16")];
            tensor<int32, [4]> var_35247_begin_0 = const()[name = tensor<string, []>("op_35247_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_35247_end_0 = const()[name = tensor<string, []>("op_35247_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_35247_end_mask_0 = const()[name = tensor<string, []>("op_35247_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35247_cast_fp16 = slice_by_index(begin = var_35247_begin_0, end = var_35247_end_0, end_mask = var_35247_end_mask_0, x = v_157_cast_fp16)[name = tensor<string, []>("op_35247_cast_fp16")];
            tensor<int32, [4]> var_35251_begin_0 = const()[name = tensor<string, []>("op_35251_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_35251_end_0 = const()[name = tensor<string, []>("op_35251_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_35251_end_mask_0 = const()[name = tensor<string, []>("op_35251_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35251_cast_fp16 = slice_by_index(begin = var_35251_begin_0, end = var_35251_end_0, end_mask = var_35251_end_mask_0, x = v_157_cast_fp16)[name = tensor<string, []>("op_35251_cast_fp16")];
            tensor<int32, [4]> var_35255_begin_0 = const()[name = tensor<string, []>("op_35255_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_35255_end_0 = const()[name = tensor<string, []>("op_35255_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_35255_end_mask_0 = const()[name = tensor<string, []>("op_35255_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35255_cast_fp16 = slice_by_index(begin = var_35255_begin_0, end = var_35255_end_0, end_mask = var_35255_end_mask_0, x = v_157_cast_fp16)[name = tensor<string, []>("op_35255_cast_fp16")];
            tensor<int32, [4]> var_35259_begin_0 = const()[name = tensor<string, []>("op_35259_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_35259_end_0 = const()[name = tensor<string, []>("op_35259_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_35259_end_mask_0 = const()[name = tensor<string, []>("op_35259_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35259_cast_fp16 = slice_by_index(begin = var_35259_begin_0, end = var_35259_end_0, end_mask = var_35259_end_mask_0, x = v_157_cast_fp16)[name = tensor<string, []>("op_35259_cast_fp16")];
            tensor<int32, [4]> var_35263_begin_0 = const()[name = tensor<string, []>("op_35263_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_35263_end_0 = const()[name = tensor<string, []>("op_35263_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_35263_end_mask_0 = const()[name = tensor<string, []>("op_35263_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35263_cast_fp16 = slice_by_index(begin = var_35263_begin_0, end = var_35263_end_0, end_mask = var_35263_end_mask_0, x = v_157_cast_fp16)[name = tensor<string, []>("op_35263_cast_fp16")];
            tensor<int32, [4]> var_35267_begin_0 = const()[name = tensor<string, []>("op_35267_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_35267_end_0 = const()[name = tensor<string, []>("op_35267_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_35267_end_mask_0 = const()[name = tensor<string, []>("op_35267_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35267_cast_fp16 = slice_by_index(begin = var_35267_begin_0, end = var_35267_end_0, end_mask = var_35267_end_mask_0, x = v_157_cast_fp16)[name = tensor<string, []>("op_35267_cast_fp16")];
            tensor<int32, [4]> var_35271_begin_0 = const()[name = tensor<string, []>("op_35271_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_35271_end_0 = const()[name = tensor<string, []>("op_35271_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_35271_end_mask_0 = const()[name = tensor<string, []>("op_35271_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35271_cast_fp16 = slice_by_index(begin = var_35271_begin_0, end = var_35271_end_0, end_mask = var_35271_end_mask_0, x = v_157_cast_fp16)[name = tensor<string, []>("op_35271_cast_fp16")];
            tensor<int32, [4]> var_35275_begin_0 = const()[name = tensor<string, []>("op_35275_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_35275_end_0 = const()[name = tensor<string, []>("op_35275_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_35275_end_mask_0 = const()[name = tensor<string, []>("op_35275_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35275_cast_fp16 = slice_by_index(begin = var_35275_begin_0, end = var_35275_end_0, end_mask = var_35275_end_mask_0, x = v_157_cast_fp16)[name = tensor<string, []>("op_35275_cast_fp16")];
            tensor<int32, [4]> var_35279_begin_0 = const()[name = tensor<string, []>("op_35279_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_35279_end_0 = const()[name = tensor<string, []>("op_35279_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_35279_end_mask_0 = const()[name = tensor<string, []>("op_35279_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35279_cast_fp16 = slice_by_index(begin = var_35279_begin_0, end = var_35279_end_0, end_mask = var_35279_end_mask_0, x = v_157_cast_fp16)[name = tensor<string, []>("op_35279_cast_fp16")];
            tensor<int32, [4]> var_35283_begin_0 = const()[name = tensor<string, []>("op_35283_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_35283_end_0 = const()[name = tensor<string, []>("op_35283_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_35283_end_mask_0 = const()[name = tensor<string, []>("op_35283_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35283_cast_fp16 = slice_by_index(begin = var_35283_begin_0, end = var_35283_end_0, end_mask = var_35283_end_mask_0, x = v_157_cast_fp16)[name = tensor<string, []>("op_35283_cast_fp16")];
            tensor<int32, [4]> var_35287_begin_0 = const()[name = tensor<string, []>("op_35287_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_35287_end_0 = const()[name = tensor<string, []>("op_35287_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_35287_end_mask_0 = const()[name = tensor<string, []>("op_35287_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35287_cast_fp16 = slice_by_index(begin = var_35287_begin_0, end = var_35287_end_0, end_mask = var_35287_end_mask_0, x = v_157_cast_fp16)[name = tensor<string, []>("op_35287_cast_fp16")];
            tensor<int32, [4]> var_35291_begin_0 = const()[name = tensor<string, []>("op_35291_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_35291_end_0 = const()[name = tensor<string, []>("op_35291_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_35291_end_mask_0 = const()[name = tensor<string, []>("op_35291_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35291_cast_fp16 = slice_by_index(begin = var_35291_begin_0, end = var_35291_end_0, end_mask = var_35291_end_mask_0, x = v_157_cast_fp16)[name = tensor<string, []>("op_35291_cast_fp16")];
            tensor<int32, [4]> var_35295_begin_0 = const()[name = tensor<string, []>("op_35295_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_35295_end_0 = const()[name = tensor<string, []>("op_35295_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_35295_end_mask_0 = const()[name = tensor<string, []>("op_35295_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35295_cast_fp16 = slice_by_index(begin = var_35295_begin_0, end = var_35295_end_0, end_mask = var_35295_end_mask_0, x = v_157_cast_fp16)[name = tensor<string, []>("op_35295_cast_fp16")];
            tensor<int32, [4]> var_35299_begin_0 = const()[name = tensor<string, []>("op_35299_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_35299_end_0 = const()[name = tensor<string, []>("op_35299_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_35299_end_mask_0 = const()[name = tensor<string, []>("op_35299_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35299_cast_fp16 = slice_by_index(begin = var_35299_begin_0, end = var_35299_end_0, end_mask = var_35299_end_mask_0, x = v_157_cast_fp16)[name = tensor<string, []>("op_35299_cast_fp16")];
            tensor<string, []> var_35303_equation_0 = const()[name = tensor<string, []>("op_35303_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35303_cast_fp16 = einsum(equation = var_35303_equation_0, values = (var_35145_cast_fp16, var_35062_cast_fp16))[name = tensor<string, []>("op_35303_cast_fp16")];
            tensor<fp16, []> var_35304_to_fp16 = const()[name = tensor<string, []>("op_35304_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2961_cast_fp16 = mul(x = var_35303_cast_fp16, y = var_35304_to_fp16)[name = tensor<string, []>("aw_2961_cast_fp16")];
            tensor<string, []> var_35307_equation_0 = const()[name = tensor<string, []>("op_35307_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35307_cast_fp16 = einsum(equation = var_35307_equation_0, values = (var_35149_cast_fp16, var_35066_cast_fp16))[name = tensor<string, []>("op_35307_cast_fp16")];
            tensor<fp16, []> var_35308_to_fp16 = const()[name = tensor<string, []>("op_35308_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2963_cast_fp16 = mul(x = var_35307_cast_fp16, y = var_35308_to_fp16)[name = tensor<string, []>("aw_2963_cast_fp16")];
            tensor<string, []> var_35311_equation_0 = const()[name = tensor<string, []>("op_35311_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35311_cast_fp16 = einsum(equation = var_35311_equation_0, values = (var_35153_cast_fp16, var_35070_cast_fp16))[name = tensor<string, []>("op_35311_cast_fp16")];
            tensor<fp16, []> var_35312_to_fp16 = const()[name = tensor<string, []>("op_35312_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2965_cast_fp16 = mul(x = var_35311_cast_fp16, y = var_35312_to_fp16)[name = tensor<string, []>("aw_2965_cast_fp16")];
            tensor<string, []> var_35315_equation_0 = const()[name = tensor<string, []>("op_35315_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35315_cast_fp16 = einsum(equation = var_35315_equation_0, values = (var_35157_cast_fp16, var_35074_cast_fp16))[name = tensor<string, []>("op_35315_cast_fp16")];
            tensor<fp16, []> var_35316_to_fp16 = const()[name = tensor<string, []>("op_35316_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2967_cast_fp16 = mul(x = var_35315_cast_fp16, y = var_35316_to_fp16)[name = tensor<string, []>("aw_2967_cast_fp16")];
            tensor<string, []> var_35319_equation_0 = const()[name = tensor<string, []>("op_35319_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35319_cast_fp16 = einsum(equation = var_35319_equation_0, values = (var_35161_cast_fp16, var_35078_cast_fp16))[name = tensor<string, []>("op_35319_cast_fp16")];
            tensor<fp16, []> var_35320_to_fp16 = const()[name = tensor<string, []>("op_35320_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2969_cast_fp16 = mul(x = var_35319_cast_fp16, y = var_35320_to_fp16)[name = tensor<string, []>("aw_2969_cast_fp16")];
            tensor<string, []> var_35323_equation_0 = const()[name = tensor<string, []>("op_35323_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35323_cast_fp16 = einsum(equation = var_35323_equation_0, values = (var_35165_cast_fp16, var_35082_cast_fp16))[name = tensor<string, []>("op_35323_cast_fp16")];
            tensor<fp16, []> var_35324_to_fp16 = const()[name = tensor<string, []>("op_35324_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2971_cast_fp16 = mul(x = var_35323_cast_fp16, y = var_35324_to_fp16)[name = tensor<string, []>("aw_2971_cast_fp16")];
            tensor<string, []> var_35327_equation_0 = const()[name = tensor<string, []>("op_35327_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35327_cast_fp16 = einsum(equation = var_35327_equation_0, values = (var_35169_cast_fp16, var_35086_cast_fp16))[name = tensor<string, []>("op_35327_cast_fp16")];
            tensor<fp16, []> var_35328_to_fp16 = const()[name = tensor<string, []>("op_35328_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2973_cast_fp16 = mul(x = var_35327_cast_fp16, y = var_35328_to_fp16)[name = tensor<string, []>("aw_2973_cast_fp16")];
            tensor<string, []> var_35331_equation_0 = const()[name = tensor<string, []>("op_35331_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35331_cast_fp16 = einsum(equation = var_35331_equation_0, values = (var_35173_cast_fp16, var_35090_cast_fp16))[name = tensor<string, []>("op_35331_cast_fp16")];
            tensor<fp16, []> var_35332_to_fp16 = const()[name = tensor<string, []>("op_35332_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2975_cast_fp16 = mul(x = var_35331_cast_fp16, y = var_35332_to_fp16)[name = tensor<string, []>("aw_2975_cast_fp16")];
            tensor<string, []> var_35335_equation_0 = const()[name = tensor<string, []>("op_35335_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35335_cast_fp16 = einsum(equation = var_35335_equation_0, values = (var_35177_cast_fp16, var_35094_cast_fp16))[name = tensor<string, []>("op_35335_cast_fp16")];
            tensor<fp16, []> var_35336_to_fp16 = const()[name = tensor<string, []>("op_35336_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2977_cast_fp16 = mul(x = var_35335_cast_fp16, y = var_35336_to_fp16)[name = tensor<string, []>("aw_2977_cast_fp16")];
            tensor<string, []> var_35339_equation_0 = const()[name = tensor<string, []>("op_35339_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35339_cast_fp16 = einsum(equation = var_35339_equation_0, values = (var_35181_cast_fp16, var_35098_cast_fp16))[name = tensor<string, []>("op_35339_cast_fp16")];
            tensor<fp16, []> var_35340_to_fp16 = const()[name = tensor<string, []>("op_35340_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2979_cast_fp16 = mul(x = var_35339_cast_fp16, y = var_35340_to_fp16)[name = tensor<string, []>("aw_2979_cast_fp16")];
            tensor<string, []> var_35343_equation_0 = const()[name = tensor<string, []>("op_35343_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35343_cast_fp16 = einsum(equation = var_35343_equation_0, values = (var_35185_cast_fp16, var_35102_cast_fp16))[name = tensor<string, []>("op_35343_cast_fp16")];
            tensor<fp16, []> var_35344_to_fp16 = const()[name = tensor<string, []>("op_35344_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2981_cast_fp16 = mul(x = var_35343_cast_fp16, y = var_35344_to_fp16)[name = tensor<string, []>("aw_2981_cast_fp16")];
            tensor<string, []> var_35347_equation_0 = const()[name = tensor<string, []>("op_35347_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35347_cast_fp16 = einsum(equation = var_35347_equation_0, values = (var_35189_cast_fp16, var_35106_cast_fp16))[name = tensor<string, []>("op_35347_cast_fp16")];
            tensor<fp16, []> var_35348_to_fp16 = const()[name = tensor<string, []>("op_35348_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2983_cast_fp16 = mul(x = var_35347_cast_fp16, y = var_35348_to_fp16)[name = tensor<string, []>("aw_2983_cast_fp16")];
            tensor<string, []> var_35351_equation_0 = const()[name = tensor<string, []>("op_35351_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35351_cast_fp16 = einsum(equation = var_35351_equation_0, values = (var_35193_cast_fp16, var_35110_cast_fp16))[name = tensor<string, []>("op_35351_cast_fp16")];
            tensor<fp16, []> var_35352_to_fp16 = const()[name = tensor<string, []>("op_35352_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2985_cast_fp16 = mul(x = var_35351_cast_fp16, y = var_35352_to_fp16)[name = tensor<string, []>("aw_2985_cast_fp16")];
            tensor<string, []> var_35355_equation_0 = const()[name = tensor<string, []>("op_35355_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35355_cast_fp16 = einsum(equation = var_35355_equation_0, values = (var_35197_cast_fp16, var_35114_cast_fp16))[name = tensor<string, []>("op_35355_cast_fp16")];
            tensor<fp16, []> var_35356_to_fp16 = const()[name = tensor<string, []>("op_35356_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2987_cast_fp16 = mul(x = var_35355_cast_fp16, y = var_35356_to_fp16)[name = tensor<string, []>("aw_2987_cast_fp16")];
            tensor<string, []> var_35359_equation_0 = const()[name = tensor<string, []>("op_35359_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35359_cast_fp16 = einsum(equation = var_35359_equation_0, values = (var_35201_cast_fp16, var_35118_cast_fp16))[name = tensor<string, []>("op_35359_cast_fp16")];
            tensor<fp16, []> var_35360_to_fp16 = const()[name = tensor<string, []>("op_35360_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2989_cast_fp16 = mul(x = var_35359_cast_fp16, y = var_35360_to_fp16)[name = tensor<string, []>("aw_2989_cast_fp16")];
            tensor<string, []> var_35363_equation_0 = const()[name = tensor<string, []>("op_35363_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35363_cast_fp16 = einsum(equation = var_35363_equation_0, values = (var_35205_cast_fp16, var_35122_cast_fp16))[name = tensor<string, []>("op_35363_cast_fp16")];
            tensor<fp16, []> var_35364_to_fp16 = const()[name = tensor<string, []>("op_35364_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2991_cast_fp16 = mul(x = var_35363_cast_fp16, y = var_35364_to_fp16)[name = tensor<string, []>("aw_2991_cast_fp16")];
            tensor<string, []> var_35367_equation_0 = const()[name = tensor<string, []>("op_35367_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35367_cast_fp16 = einsum(equation = var_35367_equation_0, values = (var_35209_cast_fp16, var_35126_cast_fp16))[name = tensor<string, []>("op_35367_cast_fp16")];
            tensor<fp16, []> var_35368_to_fp16 = const()[name = tensor<string, []>("op_35368_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2993_cast_fp16 = mul(x = var_35367_cast_fp16, y = var_35368_to_fp16)[name = tensor<string, []>("aw_2993_cast_fp16")];
            tensor<string, []> var_35371_equation_0 = const()[name = tensor<string, []>("op_35371_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35371_cast_fp16 = einsum(equation = var_35371_equation_0, values = (var_35213_cast_fp16, var_35130_cast_fp16))[name = tensor<string, []>("op_35371_cast_fp16")];
            tensor<fp16, []> var_35372_to_fp16 = const()[name = tensor<string, []>("op_35372_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2995_cast_fp16 = mul(x = var_35371_cast_fp16, y = var_35372_to_fp16)[name = tensor<string, []>("aw_2995_cast_fp16")];
            tensor<string, []> var_35375_equation_0 = const()[name = tensor<string, []>("op_35375_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35375_cast_fp16 = einsum(equation = var_35375_equation_0, values = (var_35217_cast_fp16, var_35134_cast_fp16))[name = tensor<string, []>("op_35375_cast_fp16")];
            tensor<fp16, []> var_35376_to_fp16 = const()[name = tensor<string, []>("op_35376_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2997_cast_fp16 = mul(x = var_35375_cast_fp16, y = var_35376_to_fp16)[name = tensor<string, []>("aw_2997_cast_fp16")];
            tensor<string, []> var_35379_equation_0 = const()[name = tensor<string, []>("op_35379_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35379_cast_fp16 = einsum(equation = var_35379_equation_0, values = (var_35221_cast_fp16, var_35138_cast_fp16))[name = tensor<string, []>("op_35379_cast_fp16")];
            tensor<fp16, []> var_35380_to_fp16 = const()[name = tensor<string, []>("op_35380_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2999_cast_fp16 = mul(x = var_35379_cast_fp16, y = var_35380_to_fp16)[name = tensor<string, []>("aw_2999_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35382_cast_fp16 = softmax(axis = var_30355, x = aw_2961_cast_fp16)[name = tensor<string, []>("op_35382_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35383_cast_fp16 = softmax(axis = var_30355, x = aw_2963_cast_fp16)[name = tensor<string, []>("op_35383_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35384_cast_fp16 = softmax(axis = var_30355, x = aw_2965_cast_fp16)[name = tensor<string, []>("op_35384_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35385_cast_fp16 = softmax(axis = var_30355, x = aw_2967_cast_fp16)[name = tensor<string, []>("op_35385_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35386_cast_fp16 = softmax(axis = var_30355, x = aw_2969_cast_fp16)[name = tensor<string, []>("op_35386_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35387_cast_fp16 = softmax(axis = var_30355, x = aw_2971_cast_fp16)[name = tensor<string, []>("op_35387_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35388_cast_fp16 = softmax(axis = var_30355, x = aw_2973_cast_fp16)[name = tensor<string, []>("op_35388_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35389_cast_fp16 = softmax(axis = var_30355, x = aw_2975_cast_fp16)[name = tensor<string, []>("op_35389_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35390_cast_fp16 = softmax(axis = var_30355, x = aw_2977_cast_fp16)[name = tensor<string, []>("op_35390_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35391_cast_fp16 = softmax(axis = var_30355, x = aw_2979_cast_fp16)[name = tensor<string, []>("op_35391_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35392_cast_fp16 = softmax(axis = var_30355, x = aw_2981_cast_fp16)[name = tensor<string, []>("op_35392_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35393_cast_fp16 = softmax(axis = var_30355, x = aw_2983_cast_fp16)[name = tensor<string, []>("op_35393_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35394_cast_fp16 = softmax(axis = var_30355, x = aw_2985_cast_fp16)[name = tensor<string, []>("op_35394_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35395_cast_fp16 = softmax(axis = var_30355, x = aw_2987_cast_fp16)[name = tensor<string, []>("op_35395_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35396_cast_fp16 = softmax(axis = var_30355, x = aw_2989_cast_fp16)[name = tensor<string, []>("op_35396_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35397_cast_fp16 = softmax(axis = var_30355, x = aw_2991_cast_fp16)[name = tensor<string, []>("op_35397_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35398_cast_fp16 = softmax(axis = var_30355, x = aw_2993_cast_fp16)[name = tensor<string, []>("op_35398_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35399_cast_fp16 = softmax(axis = var_30355, x = aw_2995_cast_fp16)[name = tensor<string, []>("op_35399_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35400_cast_fp16 = softmax(axis = var_30355, x = aw_2997_cast_fp16)[name = tensor<string, []>("op_35400_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35401_cast_fp16 = softmax(axis = var_30355, x = aw_2999_cast_fp16)[name = tensor<string, []>("op_35401_cast_fp16")];
            tensor<string, []> var_35403_equation_0 = const()[name = tensor<string, []>("op_35403_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35403_cast_fp16 = einsum(equation = var_35403_equation_0, values = (var_35223_cast_fp16, var_35382_cast_fp16))[name = tensor<string, []>("op_35403_cast_fp16")];
            tensor<string, []> var_35405_equation_0 = const()[name = tensor<string, []>("op_35405_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35405_cast_fp16 = einsum(equation = var_35405_equation_0, values = (var_35227_cast_fp16, var_35383_cast_fp16))[name = tensor<string, []>("op_35405_cast_fp16")];
            tensor<string, []> var_35407_equation_0 = const()[name = tensor<string, []>("op_35407_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35407_cast_fp16 = einsum(equation = var_35407_equation_0, values = (var_35231_cast_fp16, var_35384_cast_fp16))[name = tensor<string, []>("op_35407_cast_fp16")];
            tensor<string, []> var_35409_equation_0 = const()[name = tensor<string, []>("op_35409_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35409_cast_fp16 = einsum(equation = var_35409_equation_0, values = (var_35235_cast_fp16, var_35385_cast_fp16))[name = tensor<string, []>("op_35409_cast_fp16")];
            tensor<string, []> var_35411_equation_0 = const()[name = tensor<string, []>("op_35411_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35411_cast_fp16 = einsum(equation = var_35411_equation_0, values = (var_35239_cast_fp16, var_35386_cast_fp16))[name = tensor<string, []>("op_35411_cast_fp16")];
            tensor<string, []> var_35413_equation_0 = const()[name = tensor<string, []>("op_35413_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35413_cast_fp16 = einsum(equation = var_35413_equation_0, values = (var_35243_cast_fp16, var_35387_cast_fp16))[name = tensor<string, []>("op_35413_cast_fp16")];
            tensor<string, []> var_35415_equation_0 = const()[name = tensor<string, []>("op_35415_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35415_cast_fp16 = einsum(equation = var_35415_equation_0, values = (var_35247_cast_fp16, var_35388_cast_fp16))[name = tensor<string, []>("op_35415_cast_fp16")];
            tensor<string, []> var_35417_equation_0 = const()[name = tensor<string, []>("op_35417_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35417_cast_fp16 = einsum(equation = var_35417_equation_0, values = (var_35251_cast_fp16, var_35389_cast_fp16))[name = tensor<string, []>("op_35417_cast_fp16")];
            tensor<string, []> var_35419_equation_0 = const()[name = tensor<string, []>("op_35419_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35419_cast_fp16 = einsum(equation = var_35419_equation_0, values = (var_35255_cast_fp16, var_35390_cast_fp16))[name = tensor<string, []>("op_35419_cast_fp16")];
            tensor<string, []> var_35421_equation_0 = const()[name = tensor<string, []>("op_35421_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35421_cast_fp16 = einsum(equation = var_35421_equation_0, values = (var_35259_cast_fp16, var_35391_cast_fp16))[name = tensor<string, []>("op_35421_cast_fp16")];
            tensor<string, []> var_35423_equation_0 = const()[name = tensor<string, []>("op_35423_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35423_cast_fp16 = einsum(equation = var_35423_equation_0, values = (var_35263_cast_fp16, var_35392_cast_fp16))[name = tensor<string, []>("op_35423_cast_fp16")];
            tensor<string, []> var_35425_equation_0 = const()[name = tensor<string, []>("op_35425_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35425_cast_fp16 = einsum(equation = var_35425_equation_0, values = (var_35267_cast_fp16, var_35393_cast_fp16))[name = tensor<string, []>("op_35425_cast_fp16")];
            tensor<string, []> var_35427_equation_0 = const()[name = tensor<string, []>("op_35427_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35427_cast_fp16 = einsum(equation = var_35427_equation_0, values = (var_35271_cast_fp16, var_35394_cast_fp16))[name = tensor<string, []>("op_35427_cast_fp16")];
            tensor<string, []> var_35429_equation_0 = const()[name = tensor<string, []>("op_35429_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35429_cast_fp16 = einsum(equation = var_35429_equation_0, values = (var_35275_cast_fp16, var_35395_cast_fp16))[name = tensor<string, []>("op_35429_cast_fp16")];
            tensor<string, []> var_35431_equation_0 = const()[name = tensor<string, []>("op_35431_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35431_cast_fp16 = einsum(equation = var_35431_equation_0, values = (var_35279_cast_fp16, var_35396_cast_fp16))[name = tensor<string, []>("op_35431_cast_fp16")];
            tensor<string, []> var_35433_equation_0 = const()[name = tensor<string, []>("op_35433_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35433_cast_fp16 = einsum(equation = var_35433_equation_0, values = (var_35283_cast_fp16, var_35397_cast_fp16))[name = tensor<string, []>("op_35433_cast_fp16")];
            tensor<string, []> var_35435_equation_0 = const()[name = tensor<string, []>("op_35435_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35435_cast_fp16 = einsum(equation = var_35435_equation_0, values = (var_35287_cast_fp16, var_35398_cast_fp16))[name = tensor<string, []>("op_35435_cast_fp16")];
            tensor<string, []> var_35437_equation_0 = const()[name = tensor<string, []>("op_35437_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35437_cast_fp16 = einsum(equation = var_35437_equation_0, values = (var_35291_cast_fp16, var_35399_cast_fp16))[name = tensor<string, []>("op_35437_cast_fp16")];
            tensor<string, []> var_35439_equation_0 = const()[name = tensor<string, []>("op_35439_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35439_cast_fp16 = einsum(equation = var_35439_equation_0, values = (var_35295_cast_fp16, var_35400_cast_fp16))[name = tensor<string, []>("op_35439_cast_fp16")];
            tensor<string, []> var_35441_equation_0 = const()[name = tensor<string, []>("op_35441_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35441_cast_fp16 = einsum(equation = var_35441_equation_0, values = (var_35299_cast_fp16, var_35401_cast_fp16))[name = tensor<string, []>("op_35441_cast_fp16")];
            tensor<bool, []> input_479_interleave_0 = const()[name = tensor<string, []>("input_479_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_479_cast_fp16 = concat(axis = var_30355, interleave = input_479_interleave_0, values = (var_35403_cast_fp16, var_35405_cast_fp16, var_35407_cast_fp16, var_35409_cast_fp16, var_35411_cast_fp16, var_35413_cast_fp16, var_35415_cast_fp16, var_35417_cast_fp16, var_35419_cast_fp16, var_35421_cast_fp16, var_35423_cast_fp16, var_35425_cast_fp16, var_35427_cast_fp16, var_35429_cast_fp16, var_35431_cast_fp16, var_35433_cast_fp16, var_35435_cast_fp16, var_35437_cast_fp16, var_35439_cast_fp16, var_35441_cast_fp16))[name = tensor<string, []>("input_479_cast_fp16")];
            tensor<string, []> var_35451_pad_type_0 = const()[name = tensor<string, []>("op_35451_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_35451_strides_0 = const()[name = tensor<string, []>("op_35451_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_35451_pad_0 = const()[name = tensor<string, []>("op_35451_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_35451_dilations_0 = const()[name = tensor<string, []>("op_35451_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_35451_groups_0 = const()[name = tensor<string, []>("op_35451_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_5_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(137896000))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(139124864))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_5_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_5_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_5_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(139125056)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_35451_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_5_attn1_to_out_0_bias_to_fp16, dilations = var_35451_dilations_0, groups = var_35451_groups_0, pad = var_35451_pad_0, pad_type = var_35451_pad_type_0, strides = var_35451_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_5_attn1_to_out_0_weight_to_fp16_palettized, x = input_479_cast_fp16)[name = tensor<string, []>("op_35451_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_237_cast_fp16 = add(x = var_35451_cast_fp16, y = inputs_235_cast_fp16)[name = tensor<string, []>("inputs_237_cast_fp16")];
            tensor<int32, [1]> hidden_states_319_axes_0 = const()[name = tensor<string, []>("hidden_states_319_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_319_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_319_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(139127680)))];
            tensor<fp16, [1280]> hidden_states_319_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_319_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(139130304)))];
            tensor<fp16, []> var_35461_to_fp16 = const()[name = tensor<string, []>("op_35461_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_319_cast_fp16 = layer_norm(axes = hidden_states_319_axes_0, beta = hidden_states_319_beta_0_to_fp16, epsilon = var_35461_to_fp16, gamma = hidden_states_319_gamma_0_to_fp16, x = inputs_237_cast_fp16)[name = tensor<string, []>("hidden_states_319_cast_fp16")];
            tensor<string, []> q_159_pad_type_0 = const()[name = tensor<string, []>("q_159_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_159_strides_0 = const()[name = tensor<string, []>("q_159_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_159_pad_0 = const()[name = tensor<string, []>("q_159_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_159_dilations_0 = const()[name = tensor<string, []>("q_159_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_159_groups_0 = const()[name = tensor<string, []>("q_159_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_5_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(139132928))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(140361792))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_5_attn2_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_159_cast_fp16 = conv(dilations = q_159_dilations_0, groups = q_159_groups_0, pad = q_159_pad_0, pad_type = q_159_pad_type_0, strides = q_159_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_5_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_319_cast_fp16)[name = tensor<string, []>("q_159_cast_fp16")];
            tensor<string, []> k_317_pad_type_0 = const()[name = tensor<string, []>("k_317_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_317_strides_0 = const()[name = tensor<string, []>("k_317_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_317_pad_0 = const()[name = tensor<string, []>("k_317_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_317_dilations_0 = const()[name = tensor<string, []>("k_317_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_317_groups_0 = const()[name = tensor<string, []>("k_317_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_5_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(140361984))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(142328128))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_5_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> k_317_cast_fp16 = conv(dilations = k_317_dilations_0, groups = k_317_groups_0, pad = k_317_pad_0, pad_type = k_317_pad_type_0, strides = k_317_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_5_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("k_317_cast_fp16")];
            tensor<string, []> v_159_pad_type_0 = const()[name = tensor<string, []>("v_159_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_159_strides_0 = const()[name = tensor<string, []>("v_159_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_159_pad_0 = const()[name = tensor<string, []>("v_159_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_159_dilations_0 = const()[name = tensor<string, []>("v_159_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_159_groups_0 = const()[name = tensor<string, []>("v_159_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_5_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(142328320))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(144294464))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_5_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> v_159_cast_fp16 = conv(dilations = v_159_dilations_0, groups = v_159_groups_0, pad = v_159_pad_0, pad_type = v_159_pad_type_0, strides = v_159_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_5_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("v_159_cast_fp16")];
            tensor<int32, [4]> var_35494_begin_0 = const()[name = tensor<string, []>("op_35494_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_35494_end_0 = const()[name = tensor<string, []>("op_35494_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_35494_end_mask_0 = const()[name = tensor<string, []>("op_35494_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35494_cast_fp16 = slice_by_index(begin = var_35494_begin_0, end = var_35494_end_0, end_mask = var_35494_end_mask_0, x = q_159_cast_fp16)[name = tensor<string, []>("op_35494_cast_fp16")];
            tensor<int32, [4]> var_35498_begin_0 = const()[name = tensor<string, []>("op_35498_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_35498_end_0 = const()[name = tensor<string, []>("op_35498_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_35498_end_mask_0 = const()[name = tensor<string, []>("op_35498_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35498_cast_fp16 = slice_by_index(begin = var_35498_begin_0, end = var_35498_end_0, end_mask = var_35498_end_mask_0, x = q_159_cast_fp16)[name = tensor<string, []>("op_35498_cast_fp16")];
            tensor<int32, [4]> var_35502_begin_0 = const()[name = tensor<string, []>("op_35502_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_35502_end_0 = const()[name = tensor<string, []>("op_35502_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_35502_end_mask_0 = const()[name = tensor<string, []>("op_35502_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35502_cast_fp16 = slice_by_index(begin = var_35502_begin_0, end = var_35502_end_0, end_mask = var_35502_end_mask_0, x = q_159_cast_fp16)[name = tensor<string, []>("op_35502_cast_fp16")];
            tensor<int32, [4]> var_35506_begin_0 = const()[name = tensor<string, []>("op_35506_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_35506_end_0 = const()[name = tensor<string, []>("op_35506_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_35506_end_mask_0 = const()[name = tensor<string, []>("op_35506_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35506_cast_fp16 = slice_by_index(begin = var_35506_begin_0, end = var_35506_end_0, end_mask = var_35506_end_mask_0, x = q_159_cast_fp16)[name = tensor<string, []>("op_35506_cast_fp16")];
            tensor<int32, [4]> var_35510_begin_0 = const()[name = tensor<string, []>("op_35510_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_35510_end_0 = const()[name = tensor<string, []>("op_35510_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_35510_end_mask_0 = const()[name = tensor<string, []>("op_35510_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35510_cast_fp16 = slice_by_index(begin = var_35510_begin_0, end = var_35510_end_0, end_mask = var_35510_end_mask_0, x = q_159_cast_fp16)[name = tensor<string, []>("op_35510_cast_fp16")];
            tensor<int32, [4]> var_35514_begin_0 = const()[name = tensor<string, []>("op_35514_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_35514_end_0 = const()[name = tensor<string, []>("op_35514_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_35514_end_mask_0 = const()[name = tensor<string, []>("op_35514_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35514_cast_fp16 = slice_by_index(begin = var_35514_begin_0, end = var_35514_end_0, end_mask = var_35514_end_mask_0, x = q_159_cast_fp16)[name = tensor<string, []>("op_35514_cast_fp16")];
            tensor<int32, [4]> var_35518_begin_0 = const()[name = tensor<string, []>("op_35518_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_35518_end_0 = const()[name = tensor<string, []>("op_35518_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_35518_end_mask_0 = const()[name = tensor<string, []>("op_35518_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35518_cast_fp16 = slice_by_index(begin = var_35518_begin_0, end = var_35518_end_0, end_mask = var_35518_end_mask_0, x = q_159_cast_fp16)[name = tensor<string, []>("op_35518_cast_fp16")];
            tensor<int32, [4]> var_35522_begin_0 = const()[name = tensor<string, []>("op_35522_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_35522_end_0 = const()[name = tensor<string, []>("op_35522_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_35522_end_mask_0 = const()[name = tensor<string, []>("op_35522_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35522_cast_fp16 = slice_by_index(begin = var_35522_begin_0, end = var_35522_end_0, end_mask = var_35522_end_mask_0, x = q_159_cast_fp16)[name = tensor<string, []>("op_35522_cast_fp16")];
            tensor<int32, [4]> var_35526_begin_0 = const()[name = tensor<string, []>("op_35526_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_35526_end_0 = const()[name = tensor<string, []>("op_35526_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_35526_end_mask_0 = const()[name = tensor<string, []>("op_35526_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35526_cast_fp16 = slice_by_index(begin = var_35526_begin_0, end = var_35526_end_0, end_mask = var_35526_end_mask_0, x = q_159_cast_fp16)[name = tensor<string, []>("op_35526_cast_fp16")];
            tensor<int32, [4]> var_35530_begin_0 = const()[name = tensor<string, []>("op_35530_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_35530_end_0 = const()[name = tensor<string, []>("op_35530_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_35530_end_mask_0 = const()[name = tensor<string, []>("op_35530_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35530_cast_fp16 = slice_by_index(begin = var_35530_begin_0, end = var_35530_end_0, end_mask = var_35530_end_mask_0, x = q_159_cast_fp16)[name = tensor<string, []>("op_35530_cast_fp16")];
            tensor<int32, [4]> var_35534_begin_0 = const()[name = tensor<string, []>("op_35534_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_35534_end_0 = const()[name = tensor<string, []>("op_35534_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_35534_end_mask_0 = const()[name = tensor<string, []>("op_35534_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35534_cast_fp16 = slice_by_index(begin = var_35534_begin_0, end = var_35534_end_0, end_mask = var_35534_end_mask_0, x = q_159_cast_fp16)[name = tensor<string, []>("op_35534_cast_fp16")];
            tensor<int32, [4]> var_35538_begin_0 = const()[name = tensor<string, []>("op_35538_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_35538_end_0 = const()[name = tensor<string, []>("op_35538_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_35538_end_mask_0 = const()[name = tensor<string, []>("op_35538_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35538_cast_fp16 = slice_by_index(begin = var_35538_begin_0, end = var_35538_end_0, end_mask = var_35538_end_mask_0, x = q_159_cast_fp16)[name = tensor<string, []>("op_35538_cast_fp16")];
            tensor<int32, [4]> var_35542_begin_0 = const()[name = tensor<string, []>("op_35542_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_35542_end_0 = const()[name = tensor<string, []>("op_35542_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_35542_end_mask_0 = const()[name = tensor<string, []>("op_35542_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35542_cast_fp16 = slice_by_index(begin = var_35542_begin_0, end = var_35542_end_0, end_mask = var_35542_end_mask_0, x = q_159_cast_fp16)[name = tensor<string, []>("op_35542_cast_fp16")];
            tensor<int32, [4]> var_35546_begin_0 = const()[name = tensor<string, []>("op_35546_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_35546_end_0 = const()[name = tensor<string, []>("op_35546_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_35546_end_mask_0 = const()[name = tensor<string, []>("op_35546_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35546_cast_fp16 = slice_by_index(begin = var_35546_begin_0, end = var_35546_end_0, end_mask = var_35546_end_mask_0, x = q_159_cast_fp16)[name = tensor<string, []>("op_35546_cast_fp16")];
            tensor<int32, [4]> var_35550_begin_0 = const()[name = tensor<string, []>("op_35550_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_35550_end_0 = const()[name = tensor<string, []>("op_35550_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_35550_end_mask_0 = const()[name = tensor<string, []>("op_35550_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35550_cast_fp16 = slice_by_index(begin = var_35550_begin_0, end = var_35550_end_0, end_mask = var_35550_end_mask_0, x = q_159_cast_fp16)[name = tensor<string, []>("op_35550_cast_fp16")];
            tensor<int32, [4]> var_35554_begin_0 = const()[name = tensor<string, []>("op_35554_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_35554_end_0 = const()[name = tensor<string, []>("op_35554_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_35554_end_mask_0 = const()[name = tensor<string, []>("op_35554_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35554_cast_fp16 = slice_by_index(begin = var_35554_begin_0, end = var_35554_end_0, end_mask = var_35554_end_mask_0, x = q_159_cast_fp16)[name = tensor<string, []>("op_35554_cast_fp16")];
            tensor<int32, [4]> var_35558_begin_0 = const()[name = tensor<string, []>("op_35558_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_35558_end_0 = const()[name = tensor<string, []>("op_35558_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_35558_end_mask_0 = const()[name = tensor<string, []>("op_35558_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35558_cast_fp16 = slice_by_index(begin = var_35558_begin_0, end = var_35558_end_0, end_mask = var_35558_end_mask_0, x = q_159_cast_fp16)[name = tensor<string, []>("op_35558_cast_fp16")];
            tensor<int32, [4]> var_35562_begin_0 = const()[name = tensor<string, []>("op_35562_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_35562_end_0 = const()[name = tensor<string, []>("op_35562_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_35562_end_mask_0 = const()[name = tensor<string, []>("op_35562_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35562_cast_fp16 = slice_by_index(begin = var_35562_begin_0, end = var_35562_end_0, end_mask = var_35562_end_mask_0, x = q_159_cast_fp16)[name = tensor<string, []>("op_35562_cast_fp16")];
            tensor<int32, [4]> var_35566_begin_0 = const()[name = tensor<string, []>("op_35566_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_35566_end_0 = const()[name = tensor<string, []>("op_35566_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_35566_end_mask_0 = const()[name = tensor<string, []>("op_35566_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35566_cast_fp16 = slice_by_index(begin = var_35566_begin_0, end = var_35566_end_0, end_mask = var_35566_end_mask_0, x = q_159_cast_fp16)[name = tensor<string, []>("op_35566_cast_fp16")];
            tensor<int32, [4]> var_35570_begin_0 = const()[name = tensor<string, []>("op_35570_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_35570_end_0 = const()[name = tensor<string, []>("op_35570_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_35570_end_mask_0 = const()[name = tensor<string, []>("op_35570_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35570_cast_fp16 = slice_by_index(begin = var_35570_begin_0, end = var_35570_end_0, end_mask = var_35570_end_mask_0, x = q_159_cast_fp16)[name = tensor<string, []>("op_35570_cast_fp16")];
            tensor<int32, [4]> k_319_perm_0 = const()[name = tensor<string, []>("k_319_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_35577_begin_0 = const()[name = tensor<string, []>("op_35577_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_35577_end_0 = const()[name = tensor<string, []>("op_35577_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_35577_end_mask_0 = const()[name = tensor<string, []>("op_35577_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> k_319_cast_fp16 = transpose(perm = k_319_perm_0, x = k_317_cast_fp16)[name = tensor<string, []>("transpose_60")];
            tensor<fp16, [2, 77, 1, 64]> var_35577_cast_fp16 = slice_by_index(begin = var_35577_begin_0, end = var_35577_end_0, end_mask = var_35577_end_mask_0, x = k_319_cast_fp16)[name = tensor<string, []>("op_35577_cast_fp16")];
            tensor<int32, [4]> var_35581_begin_0 = const()[name = tensor<string, []>("op_35581_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_35581_end_0 = const()[name = tensor<string, []>("op_35581_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_35581_end_mask_0 = const()[name = tensor<string, []>("op_35581_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_35581_cast_fp16 = slice_by_index(begin = var_35581_begin_0, end = var_35581_end_0, end_mask = var_35581_end_mask_0, x = k_319_cast_fp16)[name = tensor<string, []>("op_35581_cast_fp16")];
            tensor<int32, [4]> var_35585_begin_0 = const()[name = tensor<string, []>("op_35585_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_35585_end_0 = const()[name = tensor<string, []>("op_35585_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_35585_end_mask_0 = const()[name = tensor<string, []>("op_35585_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_35585_cast_fp16 = slice_by_index(begin = var_35585_begin_0, end = var_35585_end_0, end_mask = var_35585_end_mask_0, x = k_319_cast_fp16)[name = tensor<string, []>("op_35585_cast_fp16")];
            tensor<int32, [4]> var_35589_begin_0 = const()[name = tensor<string, []>("op_35589_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_35589_end_0 = const()[name = tensor<string, []>("op_35589_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_35589_end_mask_0 = const()[name = tensor<string, []>("op_35589_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_35589_cast_fp16 = slice_by_index(begin = var_35589_begin_0, end = var_35589_end_0, end_mask = var_35589_end_mask_0, x = k_319_cast_fp16)[name = tensor<string, []>("op_35589_cast_fp16")];
            tensor<int32, [4]> var_35593_begin_0 = const()[name = tensor<string, []>("op_35593_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_35593_end_0 = const()[name = tensor<string, []>("op_35593_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_35593_end_mask_0 = const()[name = tensor<string, []>("op_35593_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_35593_cast_fp16 = slice_by_index(begin = var_35593_begin_0, end = var_35593_end_0, end_mask = var_35593_end_mask_0, x = k_319_cast_fp16)[name = tensor<string, []>("op_35593_cast_fp16")];
            tensor<int32, [4]> var_35597_begin_0 = const()[name = tensor<string, []>("op_35597_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_35597_end_0 = const()[name = tensor<string, []>("op_35597_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_35597_end_mask_0 = const()[name = tensor<string, []>("op_35597_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_35597_cast_fp16 = slice_by_index(begin = var_35597_begin_0, end = var_35597_end_0, end_mask = var_35597_end_mask_0, x = k_319_cast_fp16)[name = tensor<string, []>("op_35597_cast_fp16")];
            tensor<int32, [4]> var_35601_begin_0 = const()[name = tensor<string, []>("op_35601_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_35601_end_0 = const()[name = tensor<string, []>("op_35601_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_35601_end_mask_0 = const()[name = tensor<string, []>("op_35601_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_35601_cast_fp16 = slice_by_index(begin = var_35601_begin_0, end = var_35601_end_0, end_mask = var_35601_end_mask_0, x = k_319_cast_fp16)[name = tensor<string, []>("op_35601_cast_fp16")];
            tensor<int32, [4]> var_35605_begin_0 = const()[name = tensor<string, []>("op_35605_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_35605_end_0 = const()[name = tensor<string, []>("op_35605_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_35605_end_mask_0 = const()[name = tensor<string, []>("op_35605_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_35605_cast_fp16 = slice_by_index(begin = var_35605_begin_0, end = var_35605_end_0, end_mask = var_35605_end_mask_0, x = k_319_cast_fp16)[name = tensor<string, []>("op_35605_cast_fp16")];
            tensor<int32, [4]> var_35609_begin_0 = const()[name = tensor<string, []>("op_35609_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_35609_end_0 = const()[name = tensor<string, []>("op_35609_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_35609_end_mask_0 = const()[name = tensor<string, []>("op_35609_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_35609_cast_fp16 = slice_by_index(begin = var_35609_begin_0, end = var_35609_end_0, end_mask = var_35609_end_mask_0, x = k_319_cast_fp16)[name = tensor<string, []>("op_35609_cast_fp16")];
            tensor<int32, [4]> var_35613_begin_0 = const()[name = tensor<string, []>("op_35613_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_35613_end_0 = const()[name = tensor<string, []>("op_35613_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_35613_end_mask_0 = const()[name = tensor<string, []>("op_35613_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_35613_cast_fp16 = slice_by_index(begin = var_35613_begin_0, end = var_35613_end_0, end_mask = var_35613_end_mask_0, x = k_319_cast_fp16)[name = tensor<string, []>("op_35613_cast_fp16")];
            tensor<int32, [4]> var_35617_begin_0 = const()[name = tensor<string, []>("op_35617_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_35617_end_0 = const()[name = tensor<string, []>("op_35617_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_35617_end_mask_0 = const()[name = tensor<string, []>("op_35617_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_35617_cast_fp16 = slice_by_index(begin = var_35617_begin_0, end = var_35617_end_0, end_mask = var_35617_end_mask_0, x = k_319_cast_fp16)[name = tensor<string, []>("op_35617_cast_fp16")];
            tensor<int32, [4]> var_35621_begin_0 = const()[name = tensor<string, []>("op_35621_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_35621_end_0 = const()[name = tensor<string, []>("op_35621_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_35621_end_mask_0 = const()[name = tensor<string, []>("op_35621_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_35621_cast_fp16 = slice_by_index(begin = var_35621_begin_0, end = var_35621_end_0, end_mask = var_35621_end_mask_0, x = k_319_cast_fp16)[name = tensor<string, []>("op_35621_cast_fp16")];
            tensor<int32, [4]> var_35625_begin_0 = const()[name = tensor<string, []>("op_35625_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_35625_end_0 = const()[name = tensor<string, []>("op_35625_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_35625_end_mask_0 = const()[name = tensor<string, []>("op_35625_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_35625_cast_fp16 = slice_by_index(begin = var_35625_begin_0, end = var_35625_end_0, end_mask = var_35625_end_mask_0, x = k_319_cast_fp16)[name = tensor<string, []>("op_35625_cast_fp16")];
            tensor<int32, [4]> var_35629_begin_0 = const()[name = tensor<string, []>("op_35629_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_35629_end_0 = const()[name = tensor<string, []>("op_35629_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_35629_end_mask_0 = const()[name = tensor<string, []>("op_35629_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_35629_cast_fp16 = slice_by_index(begin = var_35629_begin_0, end = var_35629_end_0, end_mask = var_35629_end_mask_0, x = k_319_cast_fp16)[name = tensor<string, []>("op_35629_cast_fp16")];
            tensor<int32, [4]> var_35633_begin_0 = const()[name = tensor<string, []>("op_35633_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_35633_end_0 = const()[name = tensor<string, []>("op_35633_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_35633_end_mask_0 = const()[name = tensor<string, []>("op_35633_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_35633_cast_fp16 = slice_by_index(begin = var_35633_begin_0, end = var_35633_end_0, end_mask = var_35633_end_mask_0, x = k_319_cast_fp16)[name = tensor<string, []>("op_35633_cast_fp16")];
            tensor<int32, [4]> var_35637_begin_0 = const()[name = tensor<string, []>("op_35637_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_35637_end_0 = const()[name = tensor<string, []>("op_35637_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_35637_end_mask_0 = const()[name = tensor<string, []>("op_35637_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_35637_cast_fp16 = slice_by_index(begin = var_35637_begin_0, end = var_35637_end_0, end_mask = var_35637_end_mask_0, x = k_319_cast_fp16)[name = tensor<string, []>("op_35637_cast_fp16")];
            tensor<int32, [4]> var_35641_begin_0 = const()[name = tensor<string, []>("op_35641_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_35641_end_0 = const()[name = tensor<string, []>("op_35641_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_35641_end_mask_0 = const()[name = tensor<string, []>("op_35641_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_35641_cast_fp16 = slice_by_index(begin = var_35641_begin_0, end = var_35641_end_0, end_mask = var_35641_end_mask_0, x = k_319_cast_fp16)[name = tensor<string, []>("op_35641_cast_fp16")];
            tensor<int32, [4]> var_35645_begin_0 = const()[name = tensor<string, []>("op_35645_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_35645_end_0 = const()[name = tensor<string, []>("op_35645_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_35645_end_mask_0 = const()[name = tensor<string, []>("op_35645_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_35645_cast_fp16 = slice_by_index(begin = var_35645_begin_0, end = var_35645_end_0, end_mask = var_35645_end_mask_0, x = k_319_cast_fp16)[name = tensor<string, []>("op_35645_cast_fp16")];
            tensor<int32, [4]> var_35649_begin_0 = const()[name = tensor<string, []>("op_35649_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_35649_end_0 = const()[name = tensor<string, []>("op_35649_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_35649_end_mask_0 = const()[name = tensor<string, []>("op_35649_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_35649_cast_fp16 = slice_by_index(begin = var_35649_begin_0, end = var_35649_end_0, end_mask = var_35649_end_mask_0, x = k_319_cast_fp16)[name = tensor<string, []>("op_35649_cast_fp16")];
            tensor<int32, [4]> var_35653_begin_0 = const()[name = tensor<string, []>("op_35653_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_35653_end_0 = const()[name = tensor<string, []>("op_35653_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_35653_end_mask_0 = const()[name = tensor<string, []>("op_35653_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_35653_cast_fp16 = slice_by_index(begin = var_35653_begin_0, end = var_35653_end_0, end_mask = var_35653_end_mask_0, x = k_319_cast_fp16)[name = tensor<string, []>("op_35653_cast_fp16")];
            tensor<int32, [4]> var_35655_begin_0 = const()[name = tensor<string, []>("op_35655_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_35655_end_0 = const()[name = tensor<string, []>("op_35655_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_35655_end_mask_0 = const()[name = tensor<string, []>("op_35655_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_35655_cast_fp16 = slice_by_index(begin = var_35655_begin_0, end = var_35655_end_0, end_mask = var_35655_end_mask_0, x = v_159_cast_fp16)[name = tensor<string, []>("op_35655_cast_fp16")];
            tensor<int32, [4]> var_35659_begin_0 = const()[name = tensor<string, []>("op_35659_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_35659_end_0 = const()[name = tensor<string, []>("op_35659_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_35659_end_mask_0 = const()[name = tensor<string, []>("op_35659_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_35659_cast_fp16 = slice_by_index(begin = var_35659_begin_0, end = var_35659_end_0, end_mask = var_35659_end_mask_0, x = v_159_cast_fp16)[name = tensor<string, []>("op_35659_cast_fp16")];
            tensor<int32, [4]> var_35663_begin_0 = const()[name = tensor<string, []>("op_35663_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_35663_end_0 = const()[name = tensor<string, []>("op_35663_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_35663_end_mask_0 = const()[name = tensor<string, []>("op_35663_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_35663_cast_fp16 = slice_by_index(begin = var_35663_begin_0, end = var_35663_end_0, end_mask = var_35663_end_mask_0, x = v_159_cast_fp16)[name = tensor<string, []>("op_35663_cast_fp16")];
            tensor<int32, [4]> var_35667_begin_0 = const()[name = tensor<string, []>("op_35667_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_35667_end_0 = const()[name = tensor<string, []>("op_35667_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_35667_end_mask_0 = const()[name = tensor<string, []>("op_35667_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_35667_cast_fp16 = slice_by_index(begin = var_35667_begin_0, end = var_35667_end_0, end_mask = var_35667_end_mask_0, x = v_159_cast_fp16)[name = tensor<string, []>("op_35667_cast_fp16")];
            tensor<int32, [4]> var_35671_begin_0 = const()[name = tensor<string, []>("op_35671_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_35671_end_0 = const()[name = tensor<string, []>("op_35671_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_35671_end_mask_0 = const()[name = tensor<string, []>("op_35671_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_35671_cast_fp16 = slice_by_index(begin = var_35671_begin_0, end = var_35671_end_0, end_mask = var_35671_end_mask_0, x = v_159_cast_fp16)[name = tensor<string, []>("op_35671_cast_fp16")];
            tensor<int32, [4]> var_35675_begin_0 = const()[name = tensor<string, []>("op_35675_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_35675_end_0 = const()[name = tensor<string, []>("op_35675_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_35675_end_mask_0 = const()[name = tensor<string, []>("op_35675_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_35675_cast_fp16 = slice_by_index(begin = var_35675_begin_0, end = var_35675_end_0, end_mask = var_35675_end_mask_0, x = v_159_cast_fp16)[name = tensor<string, []>("op_35675_cast_fp16")];
            tensor<int32, [4]> var_35679_begin_0 = const()[name = tensor<string, []>("op_35679_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_35679_end_0 = const()[name = tensor<string, []>("op_35679_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_35679_end_mask_0 = const()[name = tensor<string, []>("op_35679_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_35679_cast_fp16 = slice_by_index(begin = var_35679_begin_0, end = var_35679_end_0, end_mask = var_35679_end_mask_0, x = v_159_cast_fp16)[name = tensor<string, []>("op_35679_cast_fp16")];
            tensor<int32, [4]> var_35683_begin_0 = const()[name = tensor<string, []>("op_35683_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_35683_end_0 = const()[name = tensor<string, []>("op_35683_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_35683_end_mask_0 = const()[name = tensor<string, []>("op_35683_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_35683_cast_fp16 = slice_by_index(begin = var_35683_begin_0, end = var_35683_end_0, end_mask = var_35683_end_mask_0, x = v_159_cast_fp16)[name = tensor<string, []>("op_35683_cast_fp16")];
            tensor<int32, [4]> var_35687_begin_0 = const()[name = tensor<string, []>("op_35687_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_35687_end_0 = const()[name = tensor<string, []>("op_35687_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_35687_end_mask_0 = const()[name = tensor<string, []>("op_35687_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_35687_cast_fp16 = slice_by_index(begin = var_35687_begin_0, end = var_35687_end_0, end_mask = var_35687_end_mask_0, x = v_159_cast_fp16)[name = tensor<string, []>("op_35687_cast_fp16")];
            tensor<int32, [4]> var_35691_begin_0 = const()[name = tensor<string, []>("op_35691_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_35691_end_0 = const()[name = tensor<string, []>("op_35691_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_35691_end_mask_0 = const()[name = tensor<string, []>("op_35691_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_35691_cast_fp16 = slice_by_index(begin = var_35691_begin_0, end = var_35691_end_0, end_mask = var_35691_end_mask_0, x = v_159_cast_fp16)[name = tensor<string, []>("op_35691_cast_fp16")];
            tensor<int32, [4]> var_35695_begin_0 = const()[name = tensor<string, []>("op_35695_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_35695_end_0 = const()[name = tensor<string, []>("op_35695_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_35695_end_mask_0 = const()[name = tensor<string, []>("op_35695_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_35695_cast_fp16 = slice_by_index(begin = var_35695_begin_0, end = var_35695_end_0, end_mask = var_35695_end_mask_0, x = v_159_cast_fp16)[name = tensor<string, []>("op_35695_cast_fp16")];
            tensor<int32, [4]> var_35699_begin_0 = const()[name = tensor<string, []>("op_35699_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_35699_end_0 = const()[name = tensor<string, []>("op_35699_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_35699_end_mask_0 = const()[name = tensor<string, []>("op_35699_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_35699_cast_fp16 = slice_by_index(begin = var_35699_begin_0, end = var_35699_end_0, end_mask = var_35699_end_mask_0, x = v_159_cast_fp16)[name = tensor<string, []>("op_35699_cast_fp16")];
            tensor<int32, [4]> var_35703_begin_0 = const()[name = tensor<string, []>("op_35703_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_35703_end_0 = const()[name = tensor<string, []>("op_35703_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_35703_end_mask_0 = const()[name = tensor<string, []>("op_35703_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_35703_cast_fp16 = slice_by_index(begin = var_35703_begin_0, end = var_35703_end_0, end_mask = var_35703_end_mask_0, x = v_159_cast_fp16)[name = tensor<string, []>("op_35703_cast_fp16")];
            tensor<int32, [4]> var_35707_begin_0 = const()[name = tensor<string, []>("op_35707_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_35707_end_0 = const()[name = tensor<string, []>("op_35707_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_35707_end_mask_0 = const()[name = tensor<string, []>("op_35707_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_35707_cast_fp16 = slice_by_index(begin = var_35707_begin_0, end = var_35707_end_0, end_mask = var_35707_end_mask_0, x = v_159_cast_fp16)[name = tensor<string, []>("op_35707_cast_fp16")];
            tensor<int32, [4]> var_35711_begin_0 = const()[name = tensor<string, []>("op_35711_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_35711_end_0 = const()[name = tensor<string, []>("op_35711_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_35711_end_mask_0 = const()[name = tensor<string, []>("op_35711_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_35711_cast_fp16 = slice_by_index(begin = var_35711_begin_0, end = var_35711_end_0, end_mask = var_35711_end_mask_0, x = v_159_cast_fp16)[name = tensor<string, []>("op_35711_cast_fp16")];
            tensor<int32, [4]> var_35715_begin_0 = const()[name = tensor<string, []>("op_35715_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_35715_end_0 = const()[name = tensor<string, []>("op_35715_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_35715_end_mask_0 = const()[name = tensor<string, []>("op_35715_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_35715_cast_fp16 = slice_by_index(begin = var_35715_begin_0, end = var_35715_end_0, end_mask = var_35715_end_mask_0, x = v_159_cast_fp16)[name = tensor<string, []>("op_35715_cast_fp16")];
            tensor<int32, [4]> var_35719_begin_0 = const()[name = tensor<string, []>("op_35719_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_35719_end_0 = const()[name = tensor<string, []>("op_35719_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_35719_end_mask_0 = const()[name = tensor<string, []>("op_35719_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_35719_cast_fp16 = slice_by_index(begin = var_35719_begin_0, end = var_35719_end_0, end_mask = var_35719_end_mask_0, x = v_159_cast_fp16)[name = tensor<string, []>("op_35719_cast_fp16")];
            tensor<int32, [4]> var_35723_begin_0 = const()[name = tensor<string, []>("op_35723_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_35723_end_0 = const()[name = tensor<string, []>("op_35723_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_35723_end_mask_0 = const()[name = tensor<string, []>("op_35723_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_35723_cast_fp16 = slice_by_index(begin = var_35723_begin_0, end = var_35723_end_0, end_mask = var_35723_end_mask_0, x = v_159_cast_fp16)[name = tensor<string, []>("op_35723_cast_fp16")];
            tensor<int32, [4]> var_35727_begin_0 = const()[name = tensor<string, []>("op_35727_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_35727_end_0 = const()[name = tensor<string, []>("op_35727_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_35727_end_mask_0 = const()[name = tensor<string, []>("op_35727_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_35727_cast_fp16 = slice_by_index(begin = var_35727_begin_0, end = var_35727_end_0, end_mask = var_35727_end_mask_0, x = v_159_cast_fp16)[name = tensor<string, []>("op_35727_cast_fp16")];
            tensor<int32, [4]> var_35731_begin_0 = const()[name = tensor<string, []>("op_35731_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_35731_end_0 = const()[name = tensor<string, []>("op_35731_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_35731_end_mask_0 = const()[name = tensor<string, []>("op_35731_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_35731_cast_fp16 = slice_by_index(begin = var_35731_begin_0, end = var_35731_end_0, end_mask = var_35731_end_mask_0, x = v_159_cast_fp16)[name = tensor<string, []>("op_35731_cast_fp16")];
            tensor<string, []> var_35735_equation_0 = const()[name = tensor<string, []>("op_35735_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_35735_cast_fp16 = einsum(equation = var_35735_equation_0, values = (var_35577_cast_fp16, var_35494_cast_fp16))[name = tensor<string, []>("op_35735_cast_fp16")];
            tensor<fp16, []> var_35736_to_fp16 = const()[name = tensor<string, []>("op_35736_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3001_cast_fp16 = mul(x = var_35735_cast_fp16, y = var_35736_to_fp16)[name = tensor<string, []>("aw_3001_cast_fp16")];
            tensor<string, []> var_35739_equation_0 = const()[name = tensor<string, []>("op_35739_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_35739_cast_fp16 = einsum(equation = var_35739_equation_0, values = (var_35581_cast_fp16, var_35498_cast_fp16))[name = tensor<string, []>("op_35739_cast_fp16")];
            tensor<fp16, []> var_35740_to_fp16 = const()[name = tensor<string, []>("op_35740_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3003_cast_fp16 = mul(x = var_35739_cast_fp16, y = var_35740_to_fp16)[name = tensor<string, []>("aw_3003_cast_fp16")];
            tensor<string, []> var_35743_equation_0 = const()[name = tensor<string, []>("op_35743_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_35743_cast_fp16 = einsum(equation = var_35743_equation_0, values = (var_35585_cast_fp16, var_35502_cast_fp16))[name = tensor<string, []>("op_35743_cast_fp16")];
            tensor<fp16, []> var_35744_to_fp16 = const()[name = tensor<string, []>("op_35744_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3005_cast_fp16 = mul(x = var_35743_cast_fp16, y = var_35744_to_fp16)[name = tensor<string, []>("aw_3005_cast_fp16")];
            tensor<string, []> var_35747_equation_0 = const()[name = tensor<string, []>("op_35747_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_35747_cast_fp16 = einsum(equation = var_35747_equation_0, values = (var_35589_cast_fp16, var_35506_cast_fp16))[name = tensor<string, []>("op_35747_cast_fp16")];
            tensor<fp16, []> var_35748_to_fp16 = const()[name = tensor<string, []>("op_35748_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3007_cast_fp16 = mul(x = var_35747_cast_fp16, y = var_35748_to_fp16)[name = tensor<string, []>("aw_3007_cast_fp16")];
            tensor<string, []> var_35751_equation_0 = const()[name = tensor<string, []>("op_35751_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_35751_cast_fp16 = einsum(equation = var_35751_equation_0, values = (var_35593_cast_fp16, var_35510_cast_fp16))[name = tensor<string, []>("op_35751_cast_fp16")];
            tensor<fp16, []> var_35752_to_fp16 = const()[name = tensor<string, []>("op_35752_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3009_cast_fp16 = mul(x = var_35751_cast_fp16, y = var_35752_to_fp16)[name = tensor<string, []>("aw_3009_cast_fp16")];
            tensor<string, []> var_35755_equation_0 = const()[name = tensor<string, []>("op_35755_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_35755_cast_fp16 = einsum(equation = var_35755_equation_0, values = (var_35597_cast_fp16, var_35514_cast_fp16))[name = tensor<string, []>("op_35755_cast_fp16")];
            tensor<fp16, []> var_35756_to_fp16 = const()[name = tensor<string, []>("op_35756_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3011_cast_fp16 = mul(x = var_35755_cast_fp16, y = var_35756_to_fp16)[name = tensor<string, []>("aw_3011_cast_fp16")];
            tensor<string, []> var_35759_equation_0 = const()[name = tensor<string, []>("op_35759_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_35759_cast_fp16 = einsum(equation = var_35759_equation_0, values = (var_35601_cast_fp16, var_35518_cast_fp16))[name = tensor<string, []>("op_35759_cast_fp16")];
            tensor<fp16, []> var_35760_to_fp16 = const()[name = tensor<string, []>("op_35760_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3013_cast_fp16 = mul(x = var_35759_cast_fp16, y = var_35760_to_fp16)[name = tensor<string, []>("aw_3013_cast_fp16")];
            tensor<string, []> var_35763_equation_0 = const()[name = tensor<string, []>("op_35763_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_35763_cast_fp16 = einsum(equation = var_35763_equation_0, values = (var_35605_cast_fp16, var_35522_cast_fp16))[name = tensor<string, []>("op_35763_cast_fp16")];
            tensor<fp16, []> var_35764_to_fp16 = const()[name = tensor<string, []>("op_35764_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3015_cast_fp16 = mul(x = var_35763_cast_fp16, y = var_35764_to_fp16)[name = tensor<string, []>("aw_3015_cast_fp16")];
            tensor<string, []> var_35767_equation_0 = const()[name = tensor<string, []>("op_35767_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_35767_cast_fp16 = einsum(equation = var_35767_equation_0, values = (var_35609_cast_fp16, var_35526_cast_fp16))[name = tensor<string, []>("op_35767_cast_fp16")];
            tensor<fp16, []> var_35768_to_fp16 = const()[name = tensor<string, []>("op_35768_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3017_cast_fp16 = mul(x = var_35767_cast_fp16, y = var_35768_to_fp16)[name = tensor<string, []>("aw_3017_cast_fp16")];
            tensor<string, []> var_35771_equation_0 = const()[name = tensor<string, []>("op_35771_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_35771_cast_fp16 = einsum(equation = var_35771_equation_0, values = (var_35613_cast_fp16, var_35530_cast_fp16))[name = tensor<string, []>("op_35771_cast_fp16")];
            tensor<fp16, []> var_35772_to_fp16 = const()[name = tensor<string, []>("op_35772_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3019_cast_fp16 = mul(x = var_35771_cast_fp16, y = var_35772_to_fp16)[name = tensor<string, []>("aw_3019_cast_fp16")];
            tensor<string, []> var_35775_equation_0 = const()[name = tensor<string, []>("op_35775_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_35775_cast_fp16 = einsum(equation = var_35775_equation_0, values = (var_35617_cast_fp16, var_35534_cast_fp16))[name = tensor<string, []>("op_35775_cast_fp16")];
            tensor<fp16, []> var_35776_to_fp16 = const()[name = tensor<string, []>("op_35776_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3021_cast_fp16 = mul(x = var_35775_cast_fp16, y = var_35776_to_fp16)[name = tensor<string, []>("aw_3021_cast_fp16")];
            tensor<string, []> var_35779_equation_0 = const()[name = tensor<string, []>("op_35779_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_35779_cast_fp16 = einsum(equation = var_35779_equation_0, values = (var_35621_cast_fp16, var_35538_cast_fp16))[name = tensor<string, []>("op_35779_cast_fp16")];
            tensor<fp16, []> var_35780_to_fp16 = const()[name = tensor<string, []>("op_35780_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3023_cast_fp16 = mul(x = var_35779_cast_fp16, y = var_35780_to_fp16)[name = tensor<string, []>("aw_3023_cast_fp16")];
            tensor<string, []> var_35783_equation_0 = const()[name = tensor<string, []>("op_35783_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_35783_cast_fp16 = einsum(equation = var_35783_equation_0, values = (var_35625_cast_fp16, var_35542_cast_fp16))[name = tensor<string, []>("op_35783_cast_fp16")];
            tensor<fp16, []> var_35784_to_fp16 = const()[name = tensor<string, []>("op_35784_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3025_cast_fp16 = mul(x = var_35783_cast_fp16, y = var_35784_to_fp16)[name = tensor<string, []>("aw_3025_cast_fp16")];
            tensor<string, []> var_35787_equation_0 = const()[name = tensor<string, []>("op_35787_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_35787_cast_fp16 = einsum(equation = var_35787_equation_0, values = (var_35629_cast_fp16, var_35546_cast_fp16))[name = tensor<string, []>("op_35787_cast_fp16")];
            tensor<fp16, []> var_35788_to_fp16 = const()[name = tensor<string, []>("op_35788_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3027_cast_fp16 = mul(x = var_35787_cast_fp16, y = var_35788_to_fp16)[name = tensor<string, []>("aw_3027_cast_fp16")];
            tensor<string, []> var_35791_equation_0 = const()[name = tensor<string, []>("op_35791_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_35791_cast_fp16 = einsum(equation = var_35791_equation_0, values = (var_35633_cast_fp16, var_35550_cast_fp16))[name = tensor<string, []>("op_35791_cast_fp16")];
            tensor<fp16, []> var_35792_to_fp16 = const()[name = tensor<string, []>("op_35792_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3029_cast_fp16 = mul(x = var_35791_cast_fp16, y = var_35792_to_fp16)[name = tensor<string, []>("aw_3029_cast_fp16")];
            tensor<string, []> var_35795_equation_0 = const()[name = tensor<string, []>("op_35795_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_35795_cast_fp16 = einsum(equation = var_35795_equation_0, values = (var_35637_cast_fp16, var_35554_cast_fp16))[name = tensor<string, []>("op_35795_cast_fp16")];
            tensor<fp16, []> var_35796_to_fp16 = const()[name = tensor<string, []>("op_35796_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3031_cast_fp16 = mul(x = var_35795_cast_fp16, y = var_35796_to_fp16)[name = tensor<string, []>("aw_3031_cast_fp16")];
            tensor<string, []> var_35799_equation_0 = const()[name = tensor<string, []>("op_35799_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_35799_cast_fp16 = einsum(equation = var_35799_equation_0, values = (var_35641_cast_fp16, var_35558_cast_fp16))[name = tensor<string, []>("op_35799_cast_fp16")];
            tensor<fp16, []> var_35800_to_fp16 = const()[name = tensor<string, []>("op_35800_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3033_cast_fp16 = mul(x = var_35799_cast_fp16, y = var_35800_to_fp16)[name = tensor<string, []>("aw_3033_cast_fp16")];
            tensor<string, []> var_35803_equation_0 = const()[name = tensor<string, []>("op_35803_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_35803_cast_fp16 = einsum(equation = var_35803_equation_0, values = (var_35645_cast_fp16, var_35562_cast_fp16))[name = tensor<string, []>("op_35803_cast_fp16")];
            tensor<fp16, []> var_35804_to_fp16 = const()[name = tensor<string, []>("op_35804_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3035_cast_fp16 = mul(x = var_35803_cast_fp16, y = var_35804_to_fp16)[name = tensor<string, []>("aw_3035_cast_fp16")];
            tensor<string, []> var_35807_equation_0 = const()[name = tensor<string, []>("op_35807_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_35807_cast_fp16 = einsum(equation = var_35807_equation_0, values = (var_35649_cast_fp16, var_35566_cast_fp16))[name = tensor<string, []>("op_35807_cast_fp16")];
            tensor<fp16, []> var_35808_to_fp16 = const()[name = tensor<string, []>("op_35808_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3037_cast_fp16 = mul(x = var_35807_cast_fp16, y = var_35808_to_fp16)[name = tensor<string, []>("aw_3037_cast_fp16")];
            tensor<string, []> var_35811_equation_0 = const()[name = tensor<string, []>("op_35811_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_35811_cast_fp16 = einsum(equation = var_35811_equation_0, values = (var_35653_cast_fp16, var_35570_cast_fp16))[name = tensor<string, []>("op_35811_cast_fp16")];
            tensor<fp16, []> var_35812_to_fp16 = const()[name = tensor<string, []>("op_35812_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3039_cast_fp16 = mul(x = var_35811_cast_fp16, y = var_35812_to_fp16)[name = tensor<string, []>("aw_3039_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_35814_cast_fp16 = softmax(axis = var_30355, x = aw_3001_cast_fp16)[name = tensor<string, []>("op_35814_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_35815_cast_fp16 = softmax(axis = var_30355, x = aw_3003_cast_fp16)[name = tensor<string, []>("op_35815_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_35816_cast_fp16 = softmax(axis = var_30355, x = aw_3005_cast_fp16)[name = tensor<string, []>("op_35816_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_35817_cast_fp16 = softmax(axis = var_30355, x = aw_3007_cast_fp16)[name = tensor<string, []>("op_35817_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_35818_cast_fp16 = softmax(axis = var_30355, x = aw_3009_cast_fp16)[name = tensor<string, []>("op_35818_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_35819_cast_fp16 = softmax(axis = var_30355, x = aw_3011_cast_fp16)[name = tensor<string, []>("op_35819_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_35820_cast_fp16 = softmax(axis = var_30355, x = aw_3013_cast_fp16)[name = tensor<string, []>("op_35820_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_35821_cast_fp16 = softmax(axis = var_30355, x = aw_3015_cast_fp16)[name = tensor<string, []>("op_35821_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_35822_cast_fp16 = softmax(axis = var_30355, x = aw_3017_cast_fp16)[name = tensor<string, []>("op_35822_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_35823_cast_fp16 = softmax(axis = var_30355, x = aw_3019_cast_fp16)[name = tensor<string, []>("op_35823_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_35824_cast_fp16 = softmax(axis = var_30355, x = aw_3021_cast_fp16)[name = tensor<string, []>("op_35824_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_35825_cast_fp16 = softmax(axis = var_30355, x = aw_3023_cast_fp16)[name = tensor<string, []>("op_35825_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_35826_cast_fp16 = softmax(axis = var_30355, x = aw_3025_cast_fp16)[name = tensor<string, []>("op_35826_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_35827_cast_fp16 = softmax(axis = var_30355, x = aw_3027_cast_fp16)[name = tensor<string, []>("op_35827_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_35828_cast_fp16 = softmax(axis = var_30355, x = aw_3029_cast_fp16)[name = tensor<string, []>("op_35828_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_35829_cast_fp16 = softmax(axis = var_30355, x = aw_3031_cast_fp16)[name = tensor<string, []>("op_35829_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_35830_cast_fp16 = softmax(axis = var_30355, x = aw_3033_cast_fp16)[name = tensor<string, []>("op_35830_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_35831_cast_fp16 = softmax(axis = var_30355, x = aw_3035_cast_fp16)[name = tensor<string, []>("op_35831_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_35832_cast_fp16 = softmax(axis = var_30355, x = aw_3037_cast_fp16)[name = tensor<string, []>("op_35832_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_35833_cast_fp16 = softmax(axis = var_30355, x = aw_3039_cast_fp16)[name = tensor<string, []>("op_35833_cast_fp16")];
            tensor<string, []> var_35835_equation_0 = const()[name = tensor<string, []>("op_35835_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35835_cast_fp16 = einsum(equation = var_35835_equation_0, values = (var_35655_cast_fp16, var_35814_cast_fp16))[name = tensor<string, []>("op_35835_cast_fp16")];
            tensor<string, []> var_35837_equation_0 = const()[name = tensor<string, []>("op_35837_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35837_cast_fp16 = einsum(equation = var_35837_equation_0, values = (var_35659_cast_fp16, var_35815_cast_fp16))[name = tensor<string, []>("op_35837_cast_fp16")];
            tensor<string, []> var_35839_equation_0 = const()[name = tensor<string, []>("op_35839_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35839_cast_fp16 = einsum(equation = var_35839_equation_0, values = (var_35663_cast_fp16, var_35816_cast_fp16))[name = tensor<string, []>("op_35839_cast_fp16")];
            tensor<string, []> var_35841_equation_0 = const()[name = tensor<string, []>("op_35841_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35841_cast_fp16 = einsum(equation = var_35841_equation_0, values = (var_35667_cast_fp16, var_35817_cast_fp16))[name = tensor<string, []>("op_35841_cast_fp16")];
            tensor<string, []> var_35843_equation_0 = const()[name = tensor<string, []>("op_35843_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35843_cast_fp16 = einsum(equation = var_35843_equation_0, values = (var_35671_cast_fp16, var_35818_cast_fp16))[name = tensor<string, []>("op_35843_cast_fp16")];
            tensor<string, []> var_35845_equation_0 = const()[name = tensor<string, []>("op_35845_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35845_cast_fp16 = einsum(equation = var_35845_equation_0, values = (var_35675_cast_fp16, var_35819_cast_fp16))[name = tensor<string, []>("op_35845_cast_fp16")];
            tensor<string, []> var_35847_equation_0 = const()[name = tensor<string, []>("op_35847_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35847_cast_fp16 = einsum(equation = var_35847_equation_0, values = (var_35679_cast_fp16, var_35820_cast_fp16))[name = tensor<string, []>("op_35847_cast_fp16")];
            tensor<string, []> var_35849_equation_0 = const()[name = tensor<string, []>("op_35849_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35849_cast_fp16 = einsum(equation = var_35849_equation_0, values = (var_35683_cast_fp16, var_35821_cast_fp16))[name = tensor<string, []>("op_35849_cast_fp16")];
            tensor<string, []> var_35851_equation_0 = const()[name = tensor<string, []>("op_35851_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35851_cast_fp16 = einsum(equation = var_35851_equation_0, values = (var_35687_cast_fp16, var_35822_cast_fp16))[name = tensor<string, []>("op_35851_cast_fp16")];
            tensor<string, []> var_35853_equation_0 = const()[name = tensor<string, []>("op_35853_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35853_cast_fp16 = einsum(equation = var_35853_equation_0, values = (var_35691_cast_fp16, var_35823_cast_fp16))[name = tensor<string, []>("op_35853_cast_fp16")];
            tensor<string, []> var_35855_equation_0 = const()[name = tensor<string, []>("op_35855_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35855_cast_fp16 = einsum(equation = var_35855_equation_0, values = (var_35695_cast_fp16, var_35824_cast_fp16))[name = tensor<string, []>("op_35855_cast_fp16")];
            tensor<string, []> var_35857_equation_0 = const()[name = tensor<string, []>("op_35857_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35857_cast_fp16 = einsum(equation = var_35857_equation_0, values = (var_35699_cast_fp16, var_35825_cast_fp16))[name = tensor<string, []>("op_35857_cast_fp16")];
            tensor<string, []> var_35859_equation_0 = const()[name = tensor<string, []>("op_35859_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35859_cast_fp16 = einsum(equation = var_35859_equation_0, values = (var_35703_cast_fp16, var_35826_cast_fp16))[name = tensor<string, []>("op_35859_cast_fp16")];
            tensor<string, []> var_35861_equation_0 = const()[name = tensor<string, []>("op_35861_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35861_cast_fp16 = einsum(equation = var_35861_equation_0, values = (var_35707_cast_fp16, var_35827_cast_fp16))[name = tensor<string, []>("op_35861_cast_fp16")];
            tensor<string, []> var_35863_equation_0 = const()[name = tensor<string, []>("op_35863_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35863_cast_fp16 = einsum(equation = var_35863_equation_0, values = (var_35711_cast_fp16, var_35828_cast_fp16))[name = tensor<string, []>("op_35863_cast_fp16")];
            tensor<string, []> var_35865_equation_0 = const()[name = tensor<string, []>("op_35865_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35865_cast_fp16 = einsum(equation = var_35865_equation_0, values = (var_35715_cast_fp16, var_35829_cast_fp16))[name = tensor<string, []>("op_35865_cast_fp16")];
            tensor<string, []> var_35867_equation_0 = const()[name = tensor<string, []>("op_35867_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35867_cast_fp16 = einsum(equation = var_35867_equation_0, values = (var_35719_cast_fp16, var_35830_cast_fp16))[name = tensor<string, []>("op_35867_cast_fp16")];
            tensor<string, []> var_35869_equation_0 = const()[name = tensor<string, []>("op_35869_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35869_cast_fp16 = einsum(equation = var_35869_equation_0, values = (var_35723_cast_fp16, var_35831_cast_fp16))[name = tensor<string, []>("op_35869_cast_fp16")];
            tensor<string, []> var_35871_equation_0 = const()[name = tensor<string, []>("op_35871_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35871_cast_fp16 = einsum(equation = var_35871_equation_0, values = (var_35727_cast_fp16, var_35832_cast_fp16))[name = tensor<string, []>("op_35871_cast_fp16")];
            tensor<string, []> var_35873_equation_0 = const()[name = tensor<string, []>("op_35873_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35873_cast_fp16 = einsum(equation = var_35873_equation_0, values = (var_35731_cast_fp16, var_35833_cast_fp16))[name = tensor<string, []>("op_35873_cast_fp16")];
            tensor<bool, []> input_481_interleave_0 = const()[name = tensor<string, []>("input_481_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_481_cast_fp16 = concat(axis = var_30355, interleave = input_481_interleave_0, values = (var_35835_cast_fp16, var_35837_cast_fp16, var_35839_cast_fp16, var_35841_cast_fp16, var_35843_cast_fp16, var_35845_cast_fp16, var_35847_cast_fp16, var_35849_cast_fp16, var_35851_cast_fp16, var_35853_cast_fp16, var_35855_cast_fp16, var_35857_cast_fp16, var_35859_cast_fp16, var_35861_cast_fp16, var_35863_cast_fp16, var_35865_cast_fp16, var_35867_cast_fp16, var_35869_cast_fp16, var_35871_cast_fp16, var_35873_cast_fp16))[name = tensor<string, []>("input_481_cast_fp16")];
            tensor<string, []> var_35883_pad_type_0 = const()[name = tensor<string, []>("op_35883_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_35883_strides_0 = const()[name = tensor<string, []>("op_35883_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_35883_pad_0 = const()[name = tensor<string, []>("op_35883_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_35883_dilations_0 = const()[name = tensor<string, []>("op_35883_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_35883_groups_0 = const()[name = tensor<string, []>("op_35883_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_5_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(144294656))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145523520))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_5_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_5_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_5_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145523712)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_35883_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_5_attn2_to_out_0_bias_to_fp16, dilations = var_35883_dilations_0, groups = var_35883_groups_0, pad = var_35883_pad_0, pad_type = var_35883_pad_type_0, strides = var_35883_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_5_attn2_to_out_0_weight_to_fp16_palettized, x = input_481_cast_fp16)[name = tensor<string, []>("op_35883_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_239_cast_fp16 = add(x = var_35883_cast_fp16, y = inputs_237_cast_fp16)[name = tensor<string, []>("inputs_239_cast_fp16")];
            tensor<int32, [1]> input_483_axes_0 = const()[name = tensor<string, []>("input_483_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> input_483_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_483_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145526336)))];
            tensor<fp16, [1280]> input_483_beta_0_to_fp16 = const()[name = tensor<string, []>("input_483_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145528960)))];
            tensor<fp16, []> var_35893_to_fp16 = const()[name = tensor<string, []>("op_35893_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> input_483_cast_fp16 = layer_norm(axes = input_483_axes_0, beta = input_483_beta_0_to_fp16, epsilon = var_35893_to_fp16, gamma = input_483_gamma_0_to_fp16, x = inputs_239_cast_fp16)[name = tensor<string, []>("input_483_cast_fp16")];
            tensor<string, []> var_35913_pad_type_0 = const()[name = tensor<string, []>("op_35913_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_35913_strides_0 = const()[name = tensor<string, []>("op_35913_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_35913_pad_0 = const()[name = tensor<string, []>("op_35913_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_35913_dilations_0 = const()[name = tensor<string, []>("op_35913_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_35913_groups_0 = const()[name = tensor<string, []>("op_35913_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_5_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [9830400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145531584))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(155362048))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_5_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([10240, 1280, 1, 1])];
            tensor<fp16, [10240]> up_blocks_0_attentions_0_transformer_blocks_5_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_5_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(155362240)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_35913_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_5_ff_net_0_proj_bias_to_fp16, dilations = var_35913_dilations_0, groups = var_35913_groups_0, pad = var_35913_pad_0, pad_type = var_35913_pad_type_0, strides = var_35913_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_5_ff_net_0_proj_weight_to_fp16_palettized, x = input_483_cast_fp16)[name = tensor<string, []>("op_35913_cast_fp16")];
            tensor<int32, [2]> var_35914_split_sizes_0 = const()[name = tensor<string, []>("op_35914_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_35914_axis_0 = const()[name = tensor<string, []>("op_35914_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_35914_cast_fp16_0, tensor<fp16, [2, 5120, 1, 1024]> var_35914_cast_fp16_1 = split(axis = var_35914_axis_0, split_sizes = var_35914_split_sizes_0, x = var_35913_cast_fp16)[name = tensor<string, []>("op_35914_cast_fp16")];
            tensor<string, []> var_35916_mode_0 = const()[name = tensor<string, []>("op_35916_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_35916_cast_fp16 = gelu(mode = var_35916_mode_0, x = var_35914_cast_fp16_1)[name = tensor<string, []>("op_35916_cast_fp16")];
            tensor<fp16, [2, 5120, 1, 1024]> input_485_cast_fp16 = mul(x = var_35914_cast_fp16_0, y = var_35916_cast_fp16)[name = tensor<string, []>("input_485_cast_fp16")];
            tensor<string, []> var_35924_pad_type_0 = const()[name = tensor<string, []>("op_35924_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_35924_strides_0 = const()[name = tensor<string, []>("op_35924_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_35924_pad_0 = const()[name = tensor<string, []>("op_35924_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_35924_dilations_0 = const()[name = tensor<string, []>("op_35924_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_35924_groups_0 = const()[name = tensor<string, []>("op_35924_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_5_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [4915200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(155382784))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(160298048))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_5_ff_net_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 5120, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_5_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_5_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(160298240)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_35924_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_5_ff_net_2_bias_to_fp16, dilations = var_35924_dilations_0, groups = var_35924_groups_0, pad = var_35924_pad_0, pad_type = var_35924_pad_type_0, strides = var_35924_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_5_ff_net_2_weight_to_fp16_palettized, x = input_485_cast_fp16)[name = tensor<string, []>("op_35924_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_241_cast_fp16 = add(x = var_35924_cast_fp16, y = inputs_239_cast_fp16)[name = tensor<string, []>("inputs_241_cast_fp16")];
            tensor<int32, [1]> hidden_states_323_axes_0 = const()[name = tensor<string, []>("hidden_states_323_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_323_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_323_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(160300864)))];
            tensor<fp16, [1280]> hidden_states_323_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_323_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(160303488)))];
            tensor<fp16, []> var_35940_to_fp16 = const()[name = tensor<string, []>("op_35940_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_323_cast_fp16 = layer_norm(axes = hidden_states_323_axes_0, beta = hidden_states_323_beta_0_to_fp16, epsilon = var_35940_to_fp16, gamma = hidden_states_323_gamma_0_to_fp16, x = inputs_241_cast_fp16)[name = tensor<string, []>("hidden_states_323_cast_fp16")];
            tensor<string, []> q_161_pad_type_0 = const()[name = tensor<string, []>("q_161_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_161_strides_0 = const()[name = tensor<string, []>("q_161_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_161_pad_0 = const()[name = tensor<string, []>("q_161_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_161_dilations_0 = const()[name = tensor<string, []>("q_161_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_161_groups_0 = const()[name = tensor<string, []>("q_161_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_6_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(160306112))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(161534976))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_6_attn1_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_161_cast_fp16 = conv(dilations = q_161_dilations_0, groups = q_161_groups_0, pad = q_161_pad_0, pad_type = q_161_pad_type_0, strides = q_161_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_6_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_323_cast_fp16)[name = tensor<string, []>("q_161_cast_fp16")];
            tensor<string, []> k_321_pad_type_0 = const()[name = tensor<string, []>("k_321_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_321_strides_0 = const()[name = tensor<string, []>("k_321_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_321_pad_0 = const()[name = tensor<string, []>("k_321_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_321_dilations_0 = const()[name = tensor<string, []>("k_321_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_321_groups_0 = const()[name = tensor<string, []>("k_321_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_6_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(161535168))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(162764032))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_6_attn1_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> k_321_cast_fp16 = conv(dilations = k_321_dilations_0, groups = k_321_groups_0, pad = k_321_pad_0, pad_type = k_321_pad_type_0, strides = k_321_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_6_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_323_cast_fp16)[name = tensor<string, []>("k_321_cast_fp16")];
            tensor<string, []> v_161_pad_type_0 = const()[name = tensor<string, []>("v_161_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_161_strides_0 = const()[name = tensor<string, []>("v_161_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_161_pad_0 = const()[name = tensor<string, []>("v_161_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_161_dilations_0 = const()[name = tensor<string, []>("v_161_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_161_groups_0 = const()[name = tensor<string, []>("v_161_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_6_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(162764224))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(163993088))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_6_attn1_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> v_161_cast_fp16 = conv(dilations = v_161_dilations_0, groups = v_161_groups_0, pad = v_161_pad_0, pad_type = v_161_pad_type_0, strides = v_161_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_6_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_323_cast_fp16)[name = tensor<string, []>("v_161_cast_fp16")];
            tensor<int32, [4]> var_35973_begin_0 = const()[name = tensor<string, []>("op_35973_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_35973_end_0 = const()[name = tensor<string, []>("op_35973_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_35973_end_mask_0 = const()[name = tensor<string, []>("op_35973_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35973_cast_fp16 = slice_by_index(begin = var_35973_begin_0, end = var_35973_end_0, end_mask = var_35973_end_mask_0, x = q_161_cast_fp16)[name = tensor<string, []>("op_35973_cast_fp16")];
            tensor<int32, [4]> var_35977_begin_0 = const()[name = tensor<string, []>("op_35977_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_35977_end_0 = const()[name = tensor<string, []>("op_35977_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_35977_end_mask_0 = const()[name = tensor<string, []>("op_35977_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35977_cast_fp16 = slice_by_index(begin = var_35977_begin_0, end = var_35977_end_0, end_mask = var_35977_end_mask_0, x = q_161_cast_fp16)[name = tensor<string, []>("op_35977_cast_fp16")];
            tensor<int32, [4]> var_35981_begin_0 = const()[name = tensor<string, []>("op_35981_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_35981_end_0 = const()[name = tensor<string, []>("op_35981_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_35981_end_mask_0 = const()[name = tensor<string, []>("op_35981_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35981_cast_fp16 = slice_by_index(begin = var_35981_begin_0, end = var_35981_end_0, end_mask = var_35981_end_mask_0, x = q_161_cast_fp16)[name = tensor<string, []>("op_35981_cast_fp16")];
            tensor<int32, [4]> var_35985_begin_0 = const()[name = tensor<string, []>("op_35985_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_35985_end_0 = const()[name = tensor<string, []>("op_35985_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_35985_end_mask_0 = const()[name = tensor<string, []>("op_35985_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35985_cast_fp16 = slice_by_index(begin = var_35985_begin_0, end = var_35985_end_0, end_mask = var_35985_end_mask_0, x = q_161_cast_fp16)[name = tensor<string, []>("op_35985_cast_fp16")];
            tensor<int32, [4]> var_35989_begin_0 = const()[name = tensor<string, []>("op_35989_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_35989_end_0 = const()[name = tensor<string, []>("op_35989_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_35989_end_mask_0 = const()[name = tensor<string, []>("op_35989_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35989_cast_fp16 = slice_by_index(begin = var_35989_begin_0, end = var_35989_end_0, end_mask = var_35989_end_mask_0, x = q_161_cast_fp16)[name = tensor<string, []>("op_35989_cast_fp16")];
            tensor<int32, [4]> var_35993_begin_0 = const()[name = tensor<string, []>("op_35993_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_35993_end_0 = const()[name = tensor<string, []>("op_35993_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_35993_end_mask_0 = const()[name = tensor<string, []>("op_35993_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35993_cast_fp16 = slice_by_index(begin = var_35993_begin_0, end = var_35993_end_0, end_mask = var_35993_end_mask_0, x = q_161_cast_fp16)[name = tensor<string, []>("op_35993_cast_fp16")];
            tensor<int32, [4]> var_35997_begin_0 = const()[name = tensor<string, []>("op_35997_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_35997_end_0 = const()[name = tensor<string, []>("op_35997_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_35997_end_mask_0 = const()[name = tensor<string, []>("op_35997_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35997_cast_fp16 = slice_by_index(begin = var_35997_begin_0, end = var_35997_end_0, end_mask = var_35997_end_mask_0, x = q_161_cast_fp16)[name = tensor<string, []>("op_35997_cast_fp16")];
            tensor<int32, [4]> var_36001_begin_0 = const()[name = tensor<string, []>("op_36001_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_36001_end_0 = const()[name = tensor<string, []>("op_36001_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_36001_end_mask_0 = const()[name = tensor<string, []>("op_36001_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36001_cast_fp16 = slice_by_index(begin = var_36001_begin_0, end = var_36001_end_0, end_mask = var_36001_end_mask_0, x = q_161_cast_fp16)[name = tensor<string, []>("op_36001_cast_fp16")];
            tensor<int32, [4]> var_36005_begin_0 = const()[name = tensor<string, []>("op_36005_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_36005_end_0 = const()[name = tensor<string, []>("op_36005_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_36005_end_mask_0 = const()[name = tensor<string, []>("op_36005_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36005_cast_fp16 = slice_by_index(begin = var_36005_begin_0, end = var_36005_end_0, end_mask = var_36005_end_mask_0, x = q_161_cast_fp16)[name = tensor<string, []>("op_36005_cast_fp16")];
            tensor<int32, [4]> var_36009_begin_0 = const()[name = tensor<string, []>("op_36009_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_36009_end_0 = const()[name = tensor<string, []>("op_36009_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_36009_end_mask_0 = const()[name = tensor<string, []>("op_36009_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36009_cast_fp16 = slice_by_index(begin = var_36009_begin_0, end = var_36009_end_0, end_mask = var_36009_end_mask_0, x = q_161_cast_fp16)[name = tensor<string, []>("op_36009_cast_fp16")];
            tensor<int32, [4]> var_36013_begin_0 = const()[name = tensor<string, []>("op_36013_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_36013_end_0 = const()[name = tensor<string, []>("op_36013_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_36013_end_mask_0 = const()[name = tensor<string, []>("op_36013_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36013_cast_fp16 = slice_by_index(begin = var_36013_begin_0, end = var_36013_end_0, end_mask = var_36013_end_mask_0, x = q_161_cast_fp16)[name = tensor<string, []>("op_36013_cast_fp16")];
            tensor<int32, [4]> var_36017_begin_0 = const()[name = tensor<string, []>("op_36017_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_36017_end_0 = const()[name = tensor<string, []>("op_36017_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_36017_end_mask_0 = const()[name = tensor<string, []>("op_36017_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36017_cast_fp16 = slice_by_index(begin = var_36017_begin_0, end = var_36017_end_0, end_mask = var_36017_end_mask_0, x = q_161_cast_fp16)[name = tensor<string, []>("op_36017_cast_fp16")];
            tensor<int32, [4]> var_36021_begin_0 = const()[name = tensor<string, []>("op_36021_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_36021_end_0 = const()[name = tensor<string, []>("op_36021_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_36021_end_mask_0 = const()[name = tensor<string, []>("op_36021_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36021_cast_fp16 = slice_by_index(begin = var_36021_begin_0, end = var_36021_end_0, end_mask = var_36021_end_mask_0, x = q_161_cast_fp16)[name = tensor<string, []>("op_36021_cast_fp16")];
            tensor<int32, [4]> var_36025_begin_0 = const()[name = tensor<string, []>("op_36025_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_36025_end_0 = const()[name = tensor<string, []>("op_36025_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_36025_end_mask_0 = const()[name = tensor<string, []>("op_36025_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36025_cast_fp16 = slice_by_index(begin = var_36025_begin_0, end = var_36025_end_0, end_mask = var_36025_end_mask_0, x = q_161_cast_fp16)[name = tensor<string, []>("op_36025_cast_fp16")];
            tensor<int32, [4]> var_36029_begin_0 = const()[name = tensor<string, []>("op_36029_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_36029_end_0 = const()[name = tensor<string, []>("op_36029_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_36029_end_mask_0 = const()[name = tensor<string, []>("op_36029_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36029_cast_fp16 = slice_by_index(begin = var_36029_begin_0, end = var_36029_end_0, end_mask = var_36029_end_mask_0, x = q_161_cast_fp16)[name = tensor<string, []>("op_36029_cast_fp16")];
            tensor<int32, [4]> var_36033_begin_0 = const()[name = tensor<string, []>("op_36033_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_36033_end_0 = const()[name = tensor<string, []>("op_36033_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_36033_end_mask_0 = const()[name = tensor<string, []>("op_36033_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36033_cast_fp16 = slice_by_index(begin = var_36033_begin_0, end = var_36033_end_0, end_mask = var_36033_end_mask_0, x = q_161_cast_fp16)[name = tensor<string, []>("op_36033_cast_fp16")];
            tensor<int32, [4]> var_36037_begin_0 = const()[name = tensor<string, []>("op_36037_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_36037_end_0 = const()[name = tensor<string, []>("op_36037_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_36037_end_mask_0 = const()[name = tensor<string, []>("op_36037_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36037_cast_fp16 = slice_by_index(begin = var_36037_begin_0, end = var_36037_end_0, end_mask = var_36037_end_mask_0, x = q_161_cast_fp16)[name = tensor<string, []>("op_36037_cast_fp16")];
            tensor<int32, [4]> var_36041_begin_0 = const()[name = tensor<string, []>("op_36041_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_36041_end_0 = const()[name = tensor<string, []>("op_36041_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_36041_end_mask_0 = const()[name = tensor<string, []>("op_36041_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36041_cast_fp16 = slice_by_index(begin = var_36041_begin_0, end = var_36041_end_0, end_mask = var_36041_end_mask_0, x = q_161_cast_fp16)[name = tensor<string, []>("op_36041_cast_fp16")];
            tensor<int32, [4]> var_36045_begin_0 = const()[name = tensor<string, []>("op_36045_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_36045_end_0 = const()[name = tensor<string, []>("op_36045_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_36045_end_mask_0 = const()[name = tensor<string, []>("op_36045_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36045_cast_fp16 = slice_by_index(begin = var_36045_begin_0, end = var_36045_end_0, end_mask = var_36045_end_mask_0, x = q_161_cast_fp16)[name = tensor<string, []>("op_36045_cast_fp16")];
            tensor<int32, [4]> var_36049_begin_0 = const()[name = tensor<string, []>("op_36049_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_36049_end_0 = const()[name = tensor<string, []>("op_36049_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_36049_end_mask_0 = const()[name = tensor<string, []>("op_36049_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36049_cast_fp16 = slice_by_index(begin = var_36049_begin_0, end = var_36049_end_0, end_mask = var_36049_end_mask_0, x = q_161_cast_fp16)[name = tensor<string, []>("op_36049_cast_fp16")];
            tensor<int32, [4]> k_323_perm_0 = const()[name = tensor<string, []>("k_323_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_36056_begin_0 = const()[name = tensor<string, []>("op_36056_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_36056_end_0 = const()[name = tensor<string, []>("op_36056_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_36056_end_mask_0 = const()[name = tensor<string, []>("op_36056_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> k_323_cast_fp16 = transpose(perm = k_323_perm_0, x = k_321_cast_fp16)[name = tensor<string, []>("transpose_59")];
            tensor<fp16, [2, 1024, 1, 64]> var_36056_cast_fp16 = slice_by_index(begin = var_36056_begin_0, end = var_36056_end_0, end_mask = var_36056_end_mask_0, x = k_323_cast_fp16)[name = tensor<string, []>("op_36056_cast_fp16")];
            tensor<int32, [4]> var_36060_begin_0 = const()[name = tensor<string, []>("op_36060_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_36060_end_0 = const()[name = tensor<string, []>("op_36060_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_36060_end_mask_0 = const()[name = tensor<string, []>("op_36060_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_36060_cast_fp16 = slice_by_index(begin = var_36060_begin_0, end = var_36060_end_0, end_mask = var_36060_end_mask_0, x = k_323_cast_fp16)[name = tensor<string, []>("op_36060_cast_fp16")];
            tensor<int32, [4]> var_36064_begin_0 = const()[name = tensor<string, []>("op_36064_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_36064_end_0 = const()[name = tensor<string, []>("op_36064_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_36064_end_mask_0 = const()[name = tensor<string, []>("op_36064_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_36064_cast_fp16 = slice_by_index(begin = var_36064_begin_0, end = var_36064_end_0, end_mask = var_36064_end_mask_0, x = k_323_cast_fp16)[name = tensor<string, []>("op_36064_cast_fp16")];
            tensor<int32, [4]> var_36068_begin_0 = const()[name = tensor<string, []>("op_36068_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_36068_end_0 = const()[name = tensor<string, []>("op_36068_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_36068_end_mask_0 = const()[name = tensor<string, []>("op_36068_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_36068_cast_fp16 = slice_by_index(begin = var_36068_begin_0, end = var_36068_end_0, end_mask = var_36068_end_mask_0, x = k_323_cast_fp16)[name = tensor<string, []>("op_36068_cast_fp16")];
            tensor<int32, [4]> var_36072_begin_0 = const()[name = tensor<string, []>("op_36072_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_36072_end_0 = const()[name = tensor<string, []>("op_36072_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_36072_end_mask_0 = const()[name = tensor<string, []>("op_36072_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_36072_cast_fp16 = slice_by_index(begin = var_36072_begin_0, end = var_36072_end_0, end_mask = var_36072_end_mask_0, x = k_323_cast_fp16)[name = tensor<string, []>("op_36072_cast_fp16")];
            tensor<int32, [4]> var_36076_begin_0 = const()[name = tensor<string, []>("op_36076_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_36076_end_0 = const()[name = tensor<string, []>("op_36076_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_36076_end_mask_0 = const()[name = tensor<string, []>("op_36076_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_36076_cast_fp16 = slice_by_index(begin = var_36076_begin_0, end = var_36076_end_0, end_mask = var_36076_end_mask_0, x = k_323_cast_fp16)[name = tensor<string, []>("op_36076_cast_fp16")];
            tensor<int32, [4]> var_36080_begin_0 = const()[name = tensor<string, []>("op_36080_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_36080_end_0 = const()[name = tensor<string, []>("op_36080_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_36080_end_mask_0 = const()[name = tensor<string, []>("op_36080_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_36080_cast_fp16 = slice_by_index(begin = var_36080_begin_0, end = var_36080_end_0, end_mask = var_36080_end_mask_0, x = k_323_cast_fp16)[name = tensor<string, []>("op_36080_cast_fp16")];
            tensor<int32, [4]> var_36084_begin_0 = const()[name = tensor<string, []>("op_36084_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_36084_end_0 = const()[name = tensor<string, []>("op_36084_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_36084_end_mask_0 = const()[name = tensor<string, []>("op_36084_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_36084_cast_fp16 = slice_by_index(begin = var_36084_begin_0, end = var_36084_end_0, end_mask = var_36084_end_mask_0, x = k_323_cast_fp16)[name = tensor<string, []>("op_36084_cast_fp16")];
            tensor<int32, [4]> var_36088_begin_0 = const()[name = tensor<string, []>("op_36088_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_36088_end_0 = const()[name = tensor<string, []>("op_36088_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_36088_end_mask_0 = const()[name = tensor<string, []>("op_36088_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_36088_cast_fp16 = slice_by_index(begin = var_36088_begin_0, end = var_36088_end_0, end_mask = var_36088_end_mask_0, x = k_323_cast_fp16)[name = tensor<string, []>("op_36088_cast_fp16")];
            tensor<int32, [4]> var_36092_begin_0 = const()[name = tensor<string, []>("op_36092_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_36092_end_0 = const()[name = tensor<string, []>("op_36092_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_36092_end_mask_0 = const()[name = tensor<string, []>("op_36092_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_36092_cast_fp16 = slice_by_index(begin = var_36092_begin_0, end = var_36092_end_0, end_mask = var_36092_end_mask_0, x = k_323_cast_fp16)[name = tensor<string, []>("op_36092_cast_fp16")];
            tensor<int32, [4]> var_36096_begin_0 = const()[name = tensor<string, []>("op_36096_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_36096_end_0 = const()[name = tensor<string, []>("op_36096_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_36096_end_mask_0 = const()[name = tensor<string, []>("op_36096_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_36096_cast_fp16 = slice_by_index(begin = var_36096_begin_0, end = var_36096_end_0, end_mask = var_36096_end_mask_0, x = k_323_cast_fp16)[name = tensor<string, []>("op_36096_cast_fp16")];
            tensor<int32, [4]> var_36100_begin_0 = const()[name = tensor<string, []>("op_36100_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_36100_end_0 = const()[name = tensor<string, []>("op_36100_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_36100_end_mask_0 = const()[name = tensor<string, []>("op_36100_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_36100_cast_fp16 = slice_by_index(begin = var_36100_begin_0, end = var_36100_end_0, end_mask = var_36100_end_mask_0, x = k_323_cast_fp16)[name = tensor<string, []>("op_36100_cast_fp16")];
            tensor<int32, [4]> var_36104_begin_0 = const()[name = tensor<string, []>("op_36104_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_36104_end_0 = const()[name = tensor<string, []>("op_36104_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_36104_end_mask_0 = const()[name = tensor<string, []>("op_36104_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_36104_cast_fp16 = slice_by_index(begin = var_36104_begin_0, end = var_36104_end_0, end_mask = var_36104_end_mask_0, x = k_323_cast_fp16)[name = tensor<string, []>("op_36104_cast_fp16")];
            tensor<int32, [4]> var_36108_begin_0 = const()[name = tensor<string, []>("op_36108_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_36108_end_0 = const()[name = tensor<string, []>("op_36108_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_36108_end_mask_0 = const()[name = tensor<string, []>("op_36108_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_36108_cast_fp16 = slice_by_index(begin = var_36108_begin_0, end = var_36108_end_0, end_mask = var_36108_end_mask_0, x = k_323_cast_fp16)[name = tensor<string, []>("op_36108_cast_fp16")];
            tensor<int32, [4]> var_36112_begin_0 = const()[name = tensor<string, []>("op_36112_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_36112_end_0 = const()[name = tensor<string, []>("op_36112_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_36112_end_mask_0 = const()[name = tensor<string, []>("op_36112_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_36112_cast_fp16 = slice_by_index(begin = var_36112_begin_0, end = var_36112_end_0, end_mask = var_36112_end_mask_0, x = k_323_cast_fp16)[name = tensor<string, []>("op_36112_cast_fp16")];
            tensor<int32, [4]> var_36116_begin_0 = const()[name = tensor<string, []>("op_36116_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_36116_end_0 = const()[name = tensor<string, []>("op_36116_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_36116_end_mask_0 = const()[name = tensor<string, []>("op_36116_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_36116_cast_fp16 = slice_by_index(begin = var_36116_begin_0, end = var_36116_end_0, end_mask = var_36116_end_mask_0, x = k_323_cast_fp16)[name = tensor<string, []>("op_36116_cast_fp16")];
            tensor<int32, [4]> var_36120_begin_0 = const()[name = tensor<string, []>("op_36120_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_36120_end_0 = const()[name = tensor<string, []>("op_36120_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_36120_end_mask_0 = const()[name = tensor<string, []>("op_36120_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_36120_cast_fp16 = slice_by_index(begin = var_36120_begin_0, end = var_36120_end_0, end_mask = var_36120_end_mask_0, x = k_323_cast_fp16)[name = tensor<string, []>("op_36120_cast_fp16")];
            tensor<int32, [4]> var_36124_begin_0 = const()[name = tensor<string, []>("op_36124_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_36124_end_0 = const()[name = tensor<string, []>("op_36124_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_36124_end_mask_0 = const()[name = tensor<string, []>("op_36124_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_36124_cast_fp16 = slice_by_index(begin = var_36124_begin_0, end = var_36124_end_0, end_mask = var_36124_end_mask_0, x = k_323_cast_fp16)[name = tensor<string, []>("op_36124_cast_fp16")];
            tensor<int32, [4]> var_36128_begin_0 = const()[name = tensor<string, []>("op_36128_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_36128_end_0 = const()[name = tensor<string, []>("op_36128_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_36128_end_mask_0 = const()[name = tensor<string, []>("op_36128_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_36128_cast_fp16 = slice_by_index(begin = var_36128_begin_0, end = var_36128_end_0, end_mask = var_36128_end_mask_0, x = k_323_cast_fp16)[name = tensor<string, []>("op_36128_cast_fp16")];
            tensor<int32, [4]> var_36132_begin_0 = const()[name = tensor<string, []>("op_36132_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_36132_end_0 = const()[name = tensor<string, []>("op_36132_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_36132_end_mask_0 = const()[name = tensor<string, []>("op_36132_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_36132_cast_fp16 = slice_by_index(begin = var_36132_begin_0, end = var_36132_end_0, end_mask = var_36132_end_mask_0, x = k_323_cast_fp16)[name = tensor<string, []>("op_36132_cast_fp16")];
            tensor<int32, [4]> var_36134_begin_0 = const()[name = tensor<string, []>("op_36134_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_36134_end_0 = const()[name = tensor<string, []>("op_36134_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_36134_end_mask_0 = const()[name = tensor<string, []>("op_36134_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36134_cast_fp16 = slice_by_index(begin = var_36134_begin_0, end = var_36134_end_0, end_mask = var_36134_end_mask_0, x = v_161_cast_fp16)[name = tensor<string, []>("op_36134_cast_fp16")];
            tensor<int32, [4]> var_36138_begin_0 = const()[name = tensor<string, []>("op_36138_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_36138_end_0 = const()[name = tensor<string, []>("op_36138_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_36138_end_mask_0 = const()[name = tensor<string, []>("op_36138_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36138_cast_fp16 = slice_by_index(begin = var_36138_begin_0, end = var_36138_end_0, end_mask = var_36138_end_mask_0, x = v_161_cast_fp16)[name = tensor<string, []>("op_36138_cast_fp16")];
            tensor<int32, [4]> var_36142_begin_0 = const()[name = tensor<string, []>("op_36142_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_36142_end_0 = const()[name = tensor<string, []>("op_36142_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_36142_end_mask_0 = const()[name = tensor<string, []>("op_36142_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36142_cast_fp16 = slice_by_index(begin = var_36142_begin_0, end = var_36142_end_0, end_mask = var_36142_end_mask_0, x = v_161_cast_fp16)[name = tensor<string, []>("op_36142_cast_fp16")];
            tensor<int32, [4]> var_36146_begin_0 = const()[name = tensor<string, []>("op_36146_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_36146_end_0 = const()[name = tensor<string, []>("op_36146_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_36146_end_mask_0 = const()[name = tensor<string, []>("op_36146_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36146_cast_fp16 = slice_by_index(begin = var_36146_begin_0, end = var_36146_end_0, end_mask = var_36146_end_mask_0, x = v_161_cast_fp16)[name = tensor<string, []>("op_36146_cast_fp16")];
            tensor<int32, [4]> var_36150_begin_0 = const()[name = tensor<string, []>("op_36150_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_36150_end_0 = const()[name = tensor<string, []>("op_36150_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_36150_end_mask_0 = const()[name = tensor<string, []>("op_36150_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36150_cast_fp16 = slice_by_index(begin = var_36150_begin_0, end = var_36150_end_0, end_mask = var_36150_end_mask_0, x = v_161_cast_fp16)[name = tensor<string, []>("op_36150_cast_fp16")];
            tensor<int32, [4]> var_36154_begin_0 = const()[name = tensor<string, []>("op_36154_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_36154_end_0 = const()[name = tensor<string, []>("op_36154_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_36154_end_mask_0 = const()[name = tensor<string, []>("op_36154_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36154_cast_fp16 = slice_by_index(begin = var_36154_begin_0, end = var_36154_end_0, end_mask = var_36154_end_mask_0, x = v_161_cast_fp16)[name = tensor<string, []>("op_36154_cast_fp16")];
            tensor<int32, [4]> var_36158_begin_0 = const()[name = tensor<string, []>("op_36158_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_36158_end_0 = const()[name = tensor<string, []>("op_36158_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_36158_end_mask_0 = const()[name = tensor<string, []>("op_36158_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36158_cast_fp16 = slice_by_index(begin = var_36158_begin_0, end = var_36158_end_0, end_mask = var_36158_end_mask_0, x = v_161_cast_fp16)[name = tensor<string, []>("op_36158_cast_fp16")];
            tensor<int32, [4]> var_36162_begin_0 = const()[name = tensor<string, []>("op_36162_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_36162_end_0 = const()[name = tensor<string, []>("op_36162_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_36162_end_mask_0 = const()[name = tensor<string, []>("op_36162_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36162_cast_fp16 = slice_by_index(begin = var_36162_begin_0, end = var_36162_end_0, end_mask = var_36162_end_mask_0, x = v_161_cast_fp16)[name = tensor<string, []>("op_36162_cast_fp16")];
            tensor<int32, [4]> var_36166_begin_0 = const()[name = tensor<string, []>("op_36166_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_36166_end_0 = const()[name = tensor<string, []>("op_36166_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_36166_end_mask_0 = const()[name = tensor<string, []>("op_36166_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36166_cast_fp16 = slice_by_index(begin = var_36166_begin_0, end = var_36166_end_0, end_mask = var_36166_end_mask_0, x = v_161_cast_fp16)[name = tensor<string, []>("op_36166_cast_fp16")];
            tensor<int32, [4]> var_36170_begin_0 = const()[name = tensor<string, []>("op_36170_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_36170_end_0 = const()[name = tensor<string, []>("op_36170_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_36170_end_mask_0 = const()[name = tensor<string, []>("op_36170_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36170_cast_fp16 = slice_by_index(begin = var_36170_begin_0, end = var_36170_end_0, end_mask = var_36170_end_mask_0, x = v_161_cast_fp16)[name = tensor<string, []>("op_36170_cast_fp16")];
            tensor<int32, [4]> var_36174_begin_0 = const()[name = tensor<string, []>("op_36174_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_36174_end_0 = const()[name = tensor<string, []>("op_36174_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_36174_end_mask_0 = const()[name = tensor<string, []>("op_36174_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36174_cast_fp16 = slice_by_index(begin = var_36174_begin_0, end = var_36174_end_0, end_mask = var_36174_end_mask_0, x = v_161_cast_fp16)[name = tensor<string, []>("op_36174_cast_fp16")];
            tensor<int32, [4]> var_36178_begin_0 = const()[name = tensor<string, []>("op_36178_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_36178_end_0 = const()[name = tensor<string, []>("op_36178_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_36178_end_mask_0 = const()[name = tensor<string, []>("op_36178_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36178_cast_fp16 = slice_by_index(begin = var_36178_begin_0, end = var_36178_end_0, end_mask = var_36178_end_mask_0, x = v_161_cast_fp16)[name = tensor<string, []>("op_36178_cast_fp16")];
            tensor<int32, [4]> var_36182_begin_0 = const()[name = tensor<string, []>("op_36182_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_36182_end_0 = const()[name = tensor<string, []>("op_36182_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_36182_end_mask_0 = const()[name = tensor<string, []>("op_36182_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36182_cast_fp16 = slice_by_index(begin = var_36182_begin_0, end = var_36182_end_0, end_mask = var_36182_end_mask_0, x = v_161_cast_fp16)[name = tensor<string, []>("op_36182_cast_fp16")];
            tensor<int32, [4]> var_36186_begin_0 = const()[name = tensor<string, []>("op_36186_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_36186_end_0 = const()[name = tensor<string, []>("op_36186_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_36186_end_mask_0 = const()[name = tensor<string, []>("op_36186_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36186_cast_fp16 = slice_by_index(begin = var_36186_begin_0, end = var_36186_end_0, end_mask = var_36186_end_mask_0, x = v_161_cast_fp16)[name = tensor<string, []>("op_36186_cast_fp16")];
            tensor<int32, [4]> var_36190_begin_0 = const()[name = tensor<string, []>("op_36190_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_36190_end_0 = const()[name = tensor<string, []>("op_36190_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_36190_end_mask_0 = const()[name = tensor<string, []>("op_36190_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36190_cast_fp16 = slice_by_index(begin = var_36190_begin_0, end = var_36190_end_0, end_mask = var_36190_end_mask_0, x = v_161_cast_fp16)[name = tensor<string, []>("op_36190_cast_fp16")];
            tensor<int32, [4]> var_36194_begin_0 = const()[name = tensor<string, []>("op_36194_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_36194_end_0 = const()[name = tensor<string, []>("op_36194_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_36194_end_mask_0 = const()[name = tensor<string, []>("op_36194_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36194_cast_fp16 = slice_by_index(begin = var_36194_begin_0, end = var_36194_end_0, end_mask = var_36194_end_mask_0, x = v_161_cast_fp16)[name = tensor<string, []>("op_36194_cast_fp16")];
            tensor<int32, [4]> var_36198_begin_0 = const()[name = tensor<string, []>("op_36198_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_36198_end_0 = const()[name = tensor<string, []>("op_36198_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_36198_end_mask_0 = const()[name = tensor<string, []>("op_36198_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36198_cast_fp16 = slice_by_index(begin = var_36198_begin_0, end = var_36198_end_0, end_mask = var_36198_end_mask_0, x = v_161_cast_fp16)[name = tensor<string, []>("op_36198_cast_fp16")];
            tensor<int32, [4]> var_36202_begin_0 = const()[name = tensor<string, []>("op_36202_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_36202_end_0 = const()[name = tensor<string, []>("op_36202_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_36202_end_mask_0 = const()[name = tensor<string, []>("op_36202_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36202_cast_fp16 = slice_by_index(begin = var_36202_begin_0, end = var_36202_end_0, end_mask = var_36202_end_mask_0, x = v_161_cast_fp16)[name = tensor<string, []>("op_36202_cast_fp16")];
            tensor<int32, [4]> var_36206_begin_0 = const()[name = tensor<string, []>("op_36206_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_36206_end_0 = const()[name = tensor<string, []>("op_36206_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_36206_end_mask_0 = const()[name = tensor<string, []>("op_36206_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36206_cast_fp16 = slice_by_index(begin = var_36206_begin_0, end = var_36206_end_0, end_mask = var_36206_end_mask_0, x = v_161_cast_fp16)[name = tensor<string, []>("op_36206_cast_fp16")];
            tensor<int32, [4]> var_36210_begin_0 = const()[name = tensor<string, []>("op_36210_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_36210_end_0 = const()[name = tensor<string, []>("op_36210_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_36210_end_mask_0 = const()[name = tensor<string, []>("op_36210_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36210_cast_fp16 = slice_by_index(begin = var_36210_begin_0, end = var_36210_end_0, end_mask = var_36210_end_mask_0, x = v_161_cast_fp16)[name = tensor<string, []>("op_36210_cast_fp16")];
            tensor<string, []> var_36214_equation_0 = const()[name = tensor<string, []>("op_36214_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36214_cast_fp16 = einsum(equation = var_36214_equation_0, values = (var_36056_cast_fp16, var_35973_cast_fp16))[name = tensor<string, []>("op_36214_cast_fp16")];
            tensor<fp16, []> var_36215_to_fp16 = const()[name = tensor<string, []>("op_36215_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3041_cast_fp16 = mul(x = var_36214_cast_fp16, y = var_36215_to_fp16)[name = tensor<string, []>("aw_3041_cast_fp16")];
            tensor<string, []> var_36218_equation_0 = const()[name = tensor<string, []>("op_36218_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36218_cast_fp16 = einsum(equation = var_36218_equation_0, values = (var_36060_cast_fp16, var_35977_cast_fp16))[name = tensor<string, []>("op_36218_cast_fp16")];
            tensor<fp16, []> var_36219_to_fp16 = const()[name = tensor<string, []>("op_36219_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3043_cast_fp16 = mul(x = var_36218_cast_fp16, y = var_36219_to_fp16)[name = tensor<string, []>("aw_3043_cast_fp16")];
            tensor<string, []> var_36222_equation_0 = const()[name = tensor<string, []>("op_36222_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36222_cast_fp16 = einsum(equation = var_36222_equation_0, values = (var_36064_cast_fp16, var_35981_cast_fp16))[name = tensor<string, []>("op_36222_cast_fp16")];
            tensor<fp16, []> var_36223_to_fp16 = const()[name = tensor<string, []>("op_36223_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3045_cast_fp16 = mul(x = var_36222_cast_fp16, y = var_36223_to_fp16)[name = tensor<string, []>("aw_3045_cast_fp16")];
            tensor<string, []> var_36226_equation_0 = const()[name = tensor<string, []>("op_36226_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36226_cast_fp16 = einsum(equation = var_36226_equation_0, values = (var_36068_cast_fp16, var_35985_cast_fp16))[name = tensor<string, []>("op_36226_cast_fp16")];
            tensor<fp16, []> var_36227_to_fp16 = const()[name = tensor<string, []>("op_36227_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3047_cast_fp16 = mul(x = var_36226_cast_fp16, y = var_36227_to_fp16)[name = tensor<string, []>("aw_3047_cast_fp16")];
            tensor<string, []> var_36230_equation_0 = const()[name = tensor<string, []>("op_36230_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36230_cast_fp16 = einsum(equation = var_36230_equation_0, values = (var_36072_cast_fp16, var_35989_cast_fp16))[name = tensor<string, []>("op_36230_cast_fp16")];
            tensor<fp16, []> var_36231_to_fp16 = const()[name = tensor<string, []>("op_36231_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3049_cast_fp16 = mul(x = var_36230_cast_fp16, y = var_36231_to_fp16)[name = tensor<string, []>("aw_3049_cast_fp16")];
            tensor<string, []> var_36234_equation_0 = const()[name = tensor<string, []>("op_36234_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36234_cast_fp16 = einsum(equation = var_36234_equation_0, values = (var_36076_cast_fp16, var_35993_cast_fp16))[name = tensor<string, []>("op_36234_cast_fp16")];
            tensor<fp16, []> var_36235_to_fp16 = const()[name = tensor<string, []>("op_36235_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3051_cast_fp16 = mul(x = var_36234_cast_fp16, y = var_36235_to_fp16)[name = tensor<string, []>("aw_3051_cast_fp16")];
            tensor<string, []> var_36238_equation_0 = const()[name = tensor<string, []>("op_36238_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36238_cast_fp16 = einsum(equation = var_36238_equation_0, values = (var_36080_cast_fp16, var_35997_cast_fp16))[name = tensor<string, []>("op_36238_cast_fp16")];
            tensor<fp16, []> var_36239_to_fp16 = const()[name = tensor<string, []>("op_36239_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3053_cast_fp16 = mul(x = var_36238_cast_fp16, y = var_36239_to_fp16)[name = tensor<string, []>("aw_3053_cast_fp16")];
            tensor<string, []> var_36242_equation_0 = const()[name = tensor<string, []>("op_36242_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36242_cast_fp16 = einsum(equation = var_36242_equation_0, values = (var_36084_cast_fp16, var_36001_cast_fp16))[name = tensor<string, []>("op_36242_cast_fp16")];
            tensor<fp16, []> var_36243_to_fp16 = const()[name = tensor<string, []>("op_36243_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3055_cast_fp16 = mul(x = var_36242_cast_fp16, y = var_36243_to_fp16)[name = tensor<string, []>("aw_3055_cast_fp16")];
            tensor<string, []> var_36246_equation_0 = const()[name = tensor<string, []>("op_36246_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36246_cast_fp16 = einsum(equation = var_36246_equation_0, values = (var_36088_cast_fp16, var_36005_cast_fp16))[name = tensor<string, []>("op_36246_cast_fp16")];
            tensor<fp16, []> var_36247_to_fp16 = const()[name = tensor<string, []>("op_36247_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3057_cast_fp16 = mul(x = var_36246_cast_fp16, y = var_36247_to_fp16)[name = tensor<string, []>("aw_3057_cast_fp16")];
            tensor<string, []> var_36250_equation_0 = const()[name = tensor<string, []>("op_36250_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36250_cast_fp16 = einsum(equation = var_36250_equation_0, values = (var_36092_cast_fp16, var_36009_cast_fp16))[name = tensor<string, []>("op_36250_cast_fp16")];
            tensor<fp16, []> var_36251_to_fp16 = const()[name = tensor<string, []>("op_36251_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3059_cast_fp16 = mul(x = var_36250_cast_fp16, y = var_36251_to_fp16)[name = tensor<string, []>("aw_3059_cast_fp16")];
            tensor<string, []> var_36254_equation_0 = const()[name = tensor<string, []>("op_36254_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36254_cast_fp16 = einsum(equation = var_36254_equation_0, values = (var_36096_cast_fp16, var_36013_cast_fp16))[name = tensor<string, []>("op_36254_cast_fp16")];
            tensor<fp16, []> var_36255_to_fp16 = const()[name = tensor<string, []>("op_36255_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3061_cast_fp16 = mul(x = var_36254_cast_fp16, y = var_36255_to_fp16)[name = tensor<string, []>("aw_3061_cast_fp16")];
            tensor<string, []> var_36258_equation_0 = const()[name = tensor<string, []>("op_36258_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36258_cast_fp16 = einsum(equation = var_36258_equation_0, values = (var_36100_cast_fp16, var_36017_cast_fp16))[name = tensor<string, []>("op_36258_cast_fp16")];
            tensor<fp16, []> var_36259_to_fp16 = const()[name = tensor<string, []>("op_36259_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3063_cast_fp16 = mul(x = var_36258_cast_fp16, y = var_36259_to_fp16)[name = tensor<string, []>("aw_3063_cast_fp16")];
            tensor<string, []> var_36262_equation_0 = const()[name = tensor<string, []>("op_36262_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36262_cast_fp16 = einsum(equation = var_36262_equation_0, values = (var_36104_cast_fp16, var_36021_cast_fp16))[name = tensor<string, []>("op_36262_cast_fp16")];
            tensor<fp16, []> var_36263_to_fp16 = const()[name = tensor<string, []>("op_36263_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3065_cast_fp16 = mul(x = var_36262_cast_fp16, y = var_36263_to_fp16)[name = tensor<string, []>("aw_3065_cast_fp16")];
            tensor<string, []> var_36266_equation_0 = const()[name = tensor<string, []>("op_36266_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36266_cast_fp16 = einsum(equation = var_36266_equation_0, values = (var_36108_cast_fp16, var_36025_cast_fp16))[name = tensor<string, []>("op_36266_cast_fp16")];
            tensor<fp16, []> var_36267_to_fp16 = const()[name = tensor<string, []>("op_36267_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3067_cast_fp16 = mul(x = var_36266_cast_fp16, y = var_36267_to_fp16)[name = tensor<string, []>("aw_3067_cast_fp16")];
            tensor<string, []> var_36270_equation_0 = const()[name = tensor<string, []>("op_36270_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36270_cast_fp16 = einsum(equation = var_36270_equation_0, values = (var_36112_cast_fp16, var_36029_cast_fp16))[name = tensor<string, []>("op_36270_cast_fp16")];
            tensor<fp16, []> var_36271_to_fp16 = const()[name = tensor<string, []>("op_36271_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3069_cast_fp16 = mul(x = var_36270_cast_fp16, y = var_36271_to_fp16)[name = tensor<string, []>("aw_3069_cast_fp16")];
            tensor<string, []> var_36274_equation_0 = const()[name = tensor<string, []>("op_36274_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36274_cast_fp16 = einsum(equation = var_36274_equation_0, values = (var_36116_cast_fp16, var_36033_cast_fp16))[name = tensor<string, []>("op_36274_cast_fp16")];
            tensor<fp16, []> var_36275_to_fp16 = const()[name = tensor<string, []>("op_36275_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3071_cast_fp16 = mul(x = var_36274_cast_fp16, y = var_36275_to_fp16)[name = tensor<string, []>("aw_3071_cast_fp16")];
            tensor<string, []> var_36278_equation_0 = const()[name = tensor<string, []>("op_36278_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36278_cast_fp16 = einsum(equation = var_36278_equation_0, values = (var_36120_cast_fp16, var_36037_cast_fp16))[name = tensor<string, []>("op_36278_cast_fp16")];
            tensor<fp16, []> var_36279_to_fp16 = const()[name = tensor<string, []>("op_36279_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3073_cast_fp16 = mul(x = var_36278_cast_fp16, y = var_36279_to_fp16)[name = tensor<string, []>("aw_3073_cast_fp16")];
            tensor<string, []> var_36282_equation_0 = const()[name = tensor<string, []>("op_36282_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36282_cast_fp16 = einsum(equation = var_36282_equation_0, values = (var_36124_cast_fp16, var_36041_cast_fp16))[name = tensor<string, []>("op_36282_cast_fp16")];
            tensor<fp16, []> var_36283_to_fp16 = const()[name = tensor<string, []>("op_36283_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3075_cast_fp16 = mul(x = var_36282_cast_fp16, y = var_36283_to_fp16)[name = tensor<string, []>("aw_3075_cast_fp16")];
            tensor<string, []> var_36286_equation_0 = const()[name = tensor<string, []>("op_36286_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36286_cast_fp16 = einsum(equation = var_36286_equation_0, values = (var_36128_cast_fp16, var_36045_cast_fp16))[name = tensor<string, []>("op_36286_cast_fp16")];
            tensor<fp16, []> var_36287_to_fp16 = const()[name = tensor<string, []>("op_36287_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3077_cast_fp16 = mul(x = var_36286_cast_fp16, y = var_36287_to_fp16)[name = tensor<string, []>("aw_3077_cast_fp16")];
            tensor<string, []> var_36290_equation_0 = const()[name = tensor<string, []>("op_36290_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36290_cast_fp16 = einsum(equation = var_36290_equation_0, values = (var_36132_cast_fp16, var_36049_cast_fp16))[name = tensor<string, []>("op_36290_cast_fp16")];
            tensor<fp16, []> var_36291_to_fp16 = const()[name = tensor<string, []>("op_36291_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3079_cast_fp16 = mul(x = var_36290_cast_fp16, y = var_36291_to_fp16)[name = tensor<string, []>("aw_3079_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36293_cast_fp16 = softmax(axis = var_30355, x = aw_3041_cast_fp16)[name = tensor<string, []>("op_36293_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36294_cast_fp16 = softmax(axis = var_30355, x = aw_3043_cast_fp16)[name = tensor<string, []>("op_36294_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36295_cast_fp16 = softmax(axis = var_30355, x = aw_3045_cast_fp16)[name = tensor<string, []>("op_36295_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36296_cast_fp16 = softmax(axis = var_30355, x = aw_3047_cast_fp16)[name = tensor<string, []>("op_36296_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36297_cast_fp16 = softmax(axis = var_30355, x = aw_3049_cast_fp16)[name = tensor<string, []>("op_36297_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36298_cast_fp16 = softmax(axis = var_30355, x = aw_3051_cast_fp16)[name = tensor<string, []>("op_36298_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36299_cast_fp16 = softmax(axis = var_30355, x = aw_3053_cast_fp16)[name = tensor<string, []>("op_36299_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36300_cast_fp16 = softmax(axis = var_30355, x = aw_3055_cast_fp16)[name = tensor<string, []>("op_36300_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36301_cast_fp16 = softmax(axis = var_30355, x = aw_3057_cast_fp16)[name = tensor<string, []>("op_36301_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36302_cast_fp16 = softmax(axis = var_30355, x = aw_3059_cast_fp16)[name = tensor<string, []>("op_36302_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36303_cast_fp16 = softmax(axis = var_30355, x = aw_3061_cast_fp16)[name = tensor<string, []>("op_36303_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36304_cast_fp16 = softmax(axis = var_30355, x = aw_3063_cast_fp16)[name = tensor<string, []>("op_36304_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36305_cast_fp16 = softmax(axis = var_30355, x = aw_3065_cast_fp16)[name = tensor<string, []>("op_36305_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36306_cast_fp16 = softmax(axis = var_30355, x = aw_3067_cast_fp16)[name = tensor<string, []>("op_36306_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36307_cast_fp16 = softmax(axis = var_30355, x = aw_3069_cast_fp16)[name = tensor<string, []>("op_36307_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36308_cast_fp16 = softmax(axis = var_30355, x = aw_3071_cast_fp16)[name = tensor<string, []>("op_36308_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36309_cast_fp16 = softmax(axis = var_30355, x = aw_3073_cast_fp16)[name = tensor<string, []>("op_36309_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36310_cast_fp16 = softmax(axis = var_30355, x = aw_3075_cast_fp16)[name = tensor<string, []>("op_36310_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36311_cast_fp16 = softmax(axis = var_30355, x = aw_3077_cast_fp16)[name = tensor<string, []>("op_36311_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36312_cast_fp16 = softmax(axis = var_30355, x = aw_3079_cast_fp16)[name = tensor<string, []>("op_36312_cast_fp16")];
            tensor<string, []> var_36314_equation_0 = const()[name = tensor<string, []>("op_36314_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36314_cast_fp16 = einsum(equation = var_36314_equation_0, values = (var_36134_cast_fp16, var_36293_cast_fp16))[name = tensor<string, []>("op_36314_cast_fp16")];
            tensor<string, []> var_36316_equation_0 = const()[name = tensor<string, []>("op_36316_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36316_cast_fp16 = einsum(equation = var_36316_equation_0, values = (var_36138_cast_fp16, var_36294_cast_fp16))[name = tensor<string, []>("op_36316_cast_fp16")];
            tensor<string, []> var_36318_equation_0 = const()[name = tensor<string, []>("op_36318_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36318_cast_fp16 = einsum(equation = var_36318_equation_0, values = (var_36142_cast_fp16, var_36295_cast_fp16))[name = tensor<string, []>("op_36318_cast_fp16")];
            tensor<string, []> var_36320_equation_0 = const()[name = tensor<string, []>("op_36320_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36320_cast_fp16 = einsum(equation = var_36320_equation_0, values = (var_36146_cast_fp16, var_36296_cast_fp16))[name = tensor<string, []>("op_36320_cast_fp16")];
            tensor<string, []> var_36322_equation_0 = const()[name = tensor<string, []>("op_36322_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36322_cast_fp16 = einsum(equation = var_36322_equation_0, values = (var_36150_cast_fp16, var_36297_cast_fp16))[name = tensor<string, []>("op_36322_cast_fp16")];
            tensor<string, []> var_36324_equation_0 = const()[name = tensor<string, []>("op_36324_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36324_cast_fp16 = einsum(equation = var_36324_equation_0, values = (var_36154_cast_fp16, var_36298_cast_fp16))[name = tensor<string, []>("op_36324_cast_fp16")];
            tensor<string, []> var_36326_equation_0 = const()[name = tensor<string, []>("op_36326_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36326_cast_fp16 = einsum(equation = var_36326_equation_0, values = (var_36158_cast_fp16, var_36299_cast_fp16))[name = tensor<string, []>("op_36326_cast_fp16")];
            tensor<string, []> var_36328_equation_0 = const()[name = tensor<string, []>("op_36328_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36328_cast_fp16 = einsum(equation = var_36328_equation_0, values = (var_36162_cast_fp16, var_36300_cast_fp16))[name = tensor<string, []>("op_36328_cast_fp16")];
            tensor<string, []> var_36330_equation_0 = const()[name = tensor<string, []>("op_36330_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36330_cast_fp16 = einsum(equation = var_36330_equation_0, values = (var_36166_cast_fp16, var_36301_cast_fp16))[name = tensor<string, []>("op_36330_cast_fp16")];
            tensor<string, []> var_36332_equation_0 = const()[name = tensor<string, []>("op_36332_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36332_cast_fp16 = einsum(equation = var_36332_equation_0, values = (var_36170_cast_fp16, var_36302_cast_fp16))[name = tensor<string, []>("op_36332_cast_fp16")];
            tensor<string, []> var_36334_equation_0 = const()[name = tensor<string, []>("op_36334_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36334_cast_fp16 = einsum(equation = var_36334_equation_0, values = (var_36174_cast_fp16, var_36303_cast_fp16))[name = tensor<string, []>("op_36334_cast_fp16")];
            tensor<string, []> var_36336_equation_0 = const()[name = tensor<string, []>("op_36336_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36336_cast_fp16 = einsum(equation = var_36336_equation_0, values = (var_36178_cast_fp16, var_36304_cast_fp16))[name = tensor<string, []>("op_36336_cast_fp16")];
            tensor<string, []> var_36338_equation_0 = const()[name = tensor<string, []>("op_36338_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36338_cast_fp16 = einsum(equation = var_36338_equation_0, values = (var_36182_cast_fp16, var_36305_cast_fp16))[name = tensor<string, []>("op_36338_cast_fp16")];
            tensor<string, []> var_36340_equation_0 = const()[name = tensor<string, []>("op_36340_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36340_cast_fp16 = einsum(equation = var_36340_equation_0, values = (var_36186_cast_fp16, var_36306_cast_fp16))[name = tensor<string, []>("op_36340_cast_fp16")];
            tensor<string, []> var_36342_equation_0 = const()[name = tensor<string, []>("op_36342_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36342_cast_fp16 = einsum(equation = var_36342_equation_0, values = (var_36190_cast_fp16, var_36307_cast_fp16))[name = tensor<string, []>("op_36342_cast_fp16")];
            tensor<string, []> var_36344_equation_0 = const()[name = tensor<string, []>("op_36344_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36344_cast_fp16 = einsum(equation = var_36344_equation_0, values = (var_36194_cast_fp16, var_36308_cast_fp16))[name = tensor<string, []>("op_36344_cast_fp16")];
            tensor<string, []> var_36346_equation_0 = const()[name = tensor<string, []>("op_36346_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36346_cast_fp16 = einsum(equation = var_36346_equation_0, values = (var_36198_cast_fp16, var_36309_cast_fp16))[name = tensor<string, []>("op_36346_cast_fp16")];
            tensor<string, []> var_36348_equation_0 = const()[name = tensor<string, []>("op_36348_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36348_cast_fp16 = einsum(equation = var_36348_equation_0, values = (var_36202_cast_fp16, var_36310_cast_fp16))[name = tensor<string, []>("op_36348_cast_fp16")];
            tensor<string, []> var_36350_equation_0 = const()[name = tensor<string, []>("op_36350_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36350_cast_fp16 = einsum(equation = var_36350_equation_0, values = (var_36206_cast_fp16, var_36311_cast_fp16))[name = tensor<string, []>("op_36350_cast_fp16")];
            tensor<string, []> var_36352_equation_0 = const()[name = tensor<string, []>("op_36352_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36352_cast_fp16 = einsum(equation = var_36352_equation_0, values = (var_36210_cast_fp16, var_36312_cast_fp16))[name = tensor<string, []>("op_36352_cast_fp16")];
            tensor<bool, []> input_487_interleave_0 = const()[name = tensor<string, []>("input_487_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_487_cast_fp16 = concat(axis = var_30355, interleave = input_487_interleave_0, values = (var_36314_cast_fp16, var_36316_cast_fp16, var_36318_cast_fp16, var_36320_cast_fp16, var_36322_cast_fp16, var_36324_cast_fp16, var_36326_cast_fp16, var_36328_cast_fp16, var_36330_cast_fp16, var_36332_cast_fp16, var_36334_cast_fp16, var_36336_cast_fp16, var_36338_cast_fp16, var_36340_cast_fp16, var_36342_cast_fp16, var_36344_cast_fp16, var_36346_cast_fp16, var_36348_cast_fp16, var_36350_cast_fp16, var_36352_cast_fp16))[name = tensor<string, []>("input_487_cast_fp16")];
            tensor<string, []> var_36362_pad_type_0 = const()[name = tensor<string, []>("op_36362_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_36362_strides_0 = const()[name = tensor<string, []>("op_36362_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_36362_pad_0 = const()[name = tensor<string, []>("op_36362_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_36362_dilations_0 = const()[name = tensor<string, []>("op_36362_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_36362_groups_0 = const()[name = tensor<string, []>("op_36362_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_6_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(163993280))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(165222144))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_6_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_6_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_6_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(165222336)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_36362_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_6_attn1_to_out_0_bias_to_fp16, dilations = var_36362_dilations_0, groups = var_36362_groups_0, pad = var_36362_pad_0, pad_type = var_36362_pad_type_0, strides = var_36362_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_6_attn1_to_out_0_weight_to_fp16_palettized, x = input_487_cast_fp16)[name = tensor<string, []>("op_36362_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_243_cast_fp16 = add(x = var_36362_cast_fp16, y = inputs_241_cast_fp16)[name = tensor<string, []>("inputs_243_cast_fp16")];
            tensor<int32, [1]> hidden_states_325_axes_0 = const()[name = tensor<string, []>("hidden_states_325_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_325_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_325_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(165224960)))];
            tensor<fp16, [1280]> hidden_states_325_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_325_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(165227584)))];
            tensor<fp16, []> var_36372_to_fp16 = const()[name = tensor<string, []>("op_36372_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_325_cast_fp16 = layer_norm(axes = hidden_states_325_axes_0, beta = hidden_states_325_beta_0_to_fp16, epsilon = var_36372_to_fp16, gamma = hidden_states_325_gamma_0_to_fp16, x = inputs_243_cast_fp16)[name = tensor<string, []>("hidden_states_325_cast_fp16")];
            tensor<string, []> q_163_pad_type_0 = const()[name = tensor<string, []>("q_163_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_163_strides_0 = const()[name = tensor<string, []>("q_163_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_163_pad_0 = const()[name = tensor<string, []>("q_163_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_163_dilations_0 = const()[name = tensor<string, []>("q_163_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_163_groups_0 = const()[name = tensor<string, []>("q_163_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_6_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(165230208))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(166459072))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_6_attn2_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_163_cast_fp16 = conv(dilations = q_163_dilations_0, groups = q_163_groups_0, pad = q_163_pad_0, pad_type = q_163_pad_type_0, strides = q_163_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_6_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_325_cast_fp16)[name = tensor<string, []>("q_163_cast_fp16")];
            tensor<string, []> k_325_pad_type_0 = const()[name = tensor<string, []>("k_325_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_325_strides_0 = const()[name = tensor<string, []>("k_325_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_325_pad_0 = const()[name = tensor<string, []>("k_325_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_325_dilations_0 = const()[name = tensor<string, []>("k_325_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_325_groups_0 = const()[name = tensor<string, []>("k_325_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_6_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(166459264))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(168425408))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_6_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> k_325_cast_fp16 = conv(dilations = k_325_dilations_0, groups = k_325_groups_0, pad = k_325_pad_0, pad_type = k_325_pad_type_0, strides = k_325_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_6_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("k_325_cast_fp16")];
            tensor<string, []> v_163_pad_type_0 = const()[name = tensor<string, []>("v_163_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_163_strides_0 = const()[name = tensor<string, []>("v_163_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_163_pad_0 = const()[name = tensor<string, []>("v_163_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_163_dilations_0 = const()[name = tensor<string, []>("v_163_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_163_groups_0 = const()[name = tensor<string, []>("v_163_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_6_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(168425600))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(170391744))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_6_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> v_163_cast_fp16 = conv(dilations = v_163_dilations_0, groups = v_163_groups_0, pad = v_163_pad_0, pad_type = v_163_pad_type_0, strides = v_163_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_6_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("v_163_cast_fp16")];
            tensor<int32, [4]> var_36405_begin_0 = const()[name = tensor<string, []>("op_36405_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_36405_end_0 = const()[name = tensor<string, []>("op_36405_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_36405_end_mask_0 = const()[name = tensor<string, []>("op_36405_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36405_cast_fp16 = slice_by_index(begin = var_36405_begin_0, end = var_36405_end_0, end_mask = var_36405_end_mask_0, x = q_163_cast_fp16)[name = tensor<string, []>("op_36405_cast_fp16")];
            tensor<int32, [4]> var_36409_begin_0 = const()[name = tensor<string, []>("op_36409_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_36409_end_0 = const()[name = tensor<string, []>("op_36409_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_36409_end_mask_0 = const()[name = tensor<string, []>("op_36409_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36409_cast_fp16 = slice_by_index(begin = var_36409_begin_0, end = var_36409_end_0, end_mask = var_36409_end_mask_0, x = q_163_cast_fp16)[name = tensor<string, []>("op_36409_cast_fp16")];
            tensor<int32, [4]> var_36413_begin_0 = const()[name = tensor<string, []>("op_36413_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_36413_end_0 = const()[name = tensor<string, []>("op_36413_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_36413_end_mask_0 = const()[name = tensor<string, []>("op_36413_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36413_cast_fp16 = slice_by_index(begin = var_36413_begin_0, end = var_36413_end_0, end_mask = var_36413_end_mask_0, x = q_163_cast_fp16)[name = tensor<string, []>("op_36413_cast_fp16")];
            tensor<int32, [4]> var_36417_begin_0 = const()[name = tensor<string, []>("op_36417_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_36417_end_0 = const()[name = tensor<string, []>("op_36417_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_36417_end_mask_0 = const()[name = tensor<string, []>("op_36417_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36417_cast_fp16 = slice_by_index(begin = var_36417_begin_0, end = var_36417_end_0, end_mask = var_36417_end_mask_0, x = q_163_cast_fp16)[name = tensor<string, []>("op_36417_cast_fp16")];
            tensor<int32, [4]> var_36421_begin_0 = const()[name = tensor<string, []>("op_36421_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_36421_end_0 = const()[name = tensor<string, []>("op_36421_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_36421_end_mask_0 = const()[name = tensor<string, []>("op_36421_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36421_cast_fp16 = slice_by_index(begin = var_36421_begin_0, end = var_36421_end_0, end_mask = var_36421_end_mask_0, x = q_163_cast_fp16)[name = tensor<string, []>("op_36421_cast_fp16")];
            tensor<int32, [4]> var_36425_begin_0 = const()[name = tensor<string, []>("op_36425_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_36425_end_0 = const()[name = tensor<string, []>("op_36425_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_36425_end_mask_0 = const()[name = tensor<string, []>("op_36425_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36425_cast_fp16 = slice_by_index(begin = var_36425_begin_0, end = var_36425_end_0, end_mask = var_36425_end_mask_0, x = q_163_cast_fp16)[name = tensor<string, []>("op_36425_cast_fp16")];
            tensor<int32, [4]> var_36429_begin_0 = const()[name = tensor<string, []>("op_36429_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_36429_end_0 = const()[name = tensor<string, []>("op_36429_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_36429_end_mask_0 = const()[name = tensor<string, []>("op_36429_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36429_cast_fp16 = slice_by_index(begin = var_36429_begin_0, end = var_36429_end_0, end_mask = var_36429_end_mask_0, x = q_163_cast_fp16)[name = tensor<string, []>("op_36429_cast_fp16")];
            tensor<int32, [4]> var_36433_begin_0 = const()[name = tensor<string, []>("op_36433_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_36433_end_0 = const()[name = tensor<string, []>("op_36433_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_36433_end_mask_0 = const()[name = tensor<string, []>("op_36433_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36433_cast_fp16 = slice_by_index(begin = var_36433_begin_0, end = var_36433_end_0, end_mask = var_36433_end_mask_0, x = q_163_cast_fp16)[name = tensor<string, []>("op_36433_cast_fp16")];
            tensor<int32, [4]> var_36437_begin_0 = const()[name = tensor<string, []>("op_36437_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_36437_end_0 = const()[name = tensor<string, []>("op_36437_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_36437_end_mask_0 = const()[name = tensor<string, []>("op_36437_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36437_cast_fp16 = slice_by_index(begin = var_36437_begin_0, end = var_36437_end_0, end_mask = var_36437_end_mask_0, x = q_163_cast_fp16)[name = tensor<string, []>("op_36437_cast_fp16")];
            tensor<int32, [4]> var_36441_begin_0 = const()[name = tensor<string, []>("op_36441_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_36441_end_0 = const()[name = tensor<string, []>("op_36441_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_36441_end_mask_0 = const()[name = tensor<string, []>("op_36441_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36441_cast_fp16 = slice_by_index(begin = var_36441_begin_0, end = var_36441_end_0, end_mask = var_36441_end_mask_0, x = q_163_cast_fp16)[name = tensor<string, []>("op_36441_cast_fp16")];
            tensor<int32, [4]> var_36445_begin_0 = const()[name = tensor<string, []>("op_36445_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_36445_end_0 = const()[name = tensor<string, []>("op_36445_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_36445_end_mask_0 = const()[name = tensor<string, []>("op_36445_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36445_cast_fp16 = slice_by_index(begin = var_36445_begin_0, end = var_36445_end_0, end_mask = var_36445_end_mask_0, x = q_163_cast_fp16)[name = tensor<string, []>("op_36445_cast_fp16")];
            tensor<int32, [4]> var_36449_begin_0 = const()[name = tensor<string, []>("op_36449_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_36449_end_0 = const()[name = tensor<string, []>("op_36449_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_36449_end_mask_0 = const()[name = tensor<string, []>("op_36449_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36449_cast_fp16 = slice_by_index(begin = var_36449_begin_0, end = var_36449_end_0, end_mask = var_36449_end_mask_0, x = q_163_cast_fp16)[name = tensor<string, []>("op_36449_cast_fp16")];
            tensor<int32, [4]> var_36453_begin_0 = const()[name = tensor<string, []>("op_36453_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_36453_end_0 = const()[name = tensor<string, []>("op_36453_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_36453_end_mask_0 = const()[name = tensor<string, []>("op_36453_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36453_cast_fp16 = slice_by_index(begin = var_36453_begin_0, end = var_36453_end_0, end_mask = var_36453_end_mask_0, x = q_163_cast_fp16)[name = tensor<string, []>("op_36453_cast_fp16")];
            tensor<int32, [4]> var_36457_begin_0 = const()[name = tensor<string, []>("op_36457_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_36457_end_0 = const()[name = tensor<string, []>("op_36457_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_36457_end_mask_0 = const()[name = tensor<string, []>("op_36457_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36457_cast_fp16 = slice_by_index(begin = var_36457_begin_0, end = var_36457_end_0, end_mask = var_36457_end_mask_0, x = q_163_cast_fp16)[name = tensor<string, []>("op_36457_cast_fp16")];
            tensor<int32, [4]> var_36461_begin_0 = const()[name = tensor<string, []>("op_36461_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_36461_end_0 = const()[name = tensor<string, []>("op_36461_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_36461_end_mask_0 = const()[name = tensor<string, []>("op_36461_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36461_cast_fp16 = slice_by_index(begin = var_36461_begin_0, end = var_36461_end_0, end_mask = var_36461_end_mask_0, x = q_163_cast_fp16)[name = tensor<string, []>("op_36461_cast_fp16")];
            tensor<int32, [4]> var_36465_begin_0 = const()[name = tensor<string, []>("op_36465_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_36465_end_0 = const()[name = tensor<string, []>("op_36465_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_36465_end_mask_0 = const()[name = tensor<string, []>("op_36465_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36465_cast_fp16 = slice_by_index(begin = var_36465_begin_0, end = var_36465_end_0, end_mask = var_36465_end_mask_0, x = q_163_cast_fp16)[name = tensor<string, []>("op_36465_cast_fp16")];
            tensor<int32, [4]> var_36469_begin_0 = const()[name = tensor<string, []>("op_36469_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_36469_end_0 = const()[name = tensor<string, []>("op_36469_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_36469_end_mask_0 = const()[name = tensor<string, []>("op_36469_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36469_cast_fp16 = slice_by_index(begin = var_36469_begin_0, end = var_36469_end_0, end_mask = var_36469_end_mask_0, x = q_163_cast_fp16)[name = tensor<string, []>("op_36469_cast_fp16")];
            tensor<int32, [4]> var_36473_begin_0 = const()[name = tensor<string, []>("op_36473_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_36473_end_0 = const()[name = tensor<string, []>("op_36473_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_36473_end_mask_0 = const()[name = tensor<string, []>("op_36473_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36473_cast_fp16 = slice_by_index(begin = var_36473_begin_0, end = var_36473_end_0, end_mask = var_36473_end_mask_0, x = q_163_cast_fp16)[name = tensor<string, []>("op_36473_cast_fp16")];
            tensor<int32, [4]> var_36477_begin_0 = const()[name = tensor<string, []>("op_36477_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_36477_end_0 = const()[name = tensor<string, []>("op_36477_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_36477_end_mask_0 = const()[name = tensor<string, []>("op_36477_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36477_cast_fp16 = slice_by_index(begin = var_36477_begin_0, end = var_36477_end_0, end_mask = var_36477_end_mask_0, x = q_163_cast_fp16)[name = tensor<string, []>("op_36477_cast_fp16")];
            tensor<int32, [4]> var_36481_begin_0 = const()[name = tensor<string, []>("op_36481_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_36481_end_0 = const()[name = tensor<string, []>("op_36481_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_36481_end_mask_0 = const()[name = tensor<string, []>("op_36481_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36481_cast_fp16 = slice_by_index(begin = var_36481_begin_0, end = var_36481_end_0, end_mask = var_36481_end_mask_0, x = q_163_cast_fp16)[name = tensor<string, []>("op_36481_cast_fp16")];
            tensor<int32, [4]> k_327_perm_0 = const()[name = tensor<string, []>("k_327_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_36488_begin_0 = const()[name = tensor<string, []>("op_36488_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_36488_end_0 = const()[name = tensor<string, []>("op_36488_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_36488_end_mask_0 = const()[name = tensor<string, []>("op_36488_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> k_327_cast_fp16 = transpose(perm = k_327_perm_0, x = k_325_cast_fp16)[name = tensor<string, []>("transpose_58")];
            tensor<fp16, [2, 77, 1, 64]> var_36488_cast_fp16 = slice_by_index(begin = var_36488_begin_0, end = var_36488_end_0, end_mask = var_36488_end_mask_0, x = k_327_cast_fp16)[name = tensor<string, []>("op_36488_cast_fp16")];
            tensor<int32, [4]> var_36492_begin_0 = const()[name = tensor<string, []>("op_36492_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_36492_end_0 = const()[name = tensor<string, []>("op_36492_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_36492_end_mask_0 = const()[name = tensor<string, []>("op_36492_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_36492_cast_fp16 = slice_by_index(begin = var_36492_begin_0, end = var_36492_end_0, end_mask = var_36492_end_mask_0, x = k_327_cast_fp16)[name = tensor<string, []>("op_36492_cast_fp16")];
            tensor<int32, [4]> var_36496_begin_0 = const()[name = tensor<string, []>("op_36496_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_36496_end_0 = const()[name = tensor<string, []>("op_36496_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_36496_end_mask_0 = const()[name = tensor<string, []>("op_36496_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_36496_cast_fp16 = slice_by_index(begin = var_36496_begin_0, end = var_36496_end_0, end_mask = var_36496_end_mask_0, x = k_327_cast_fp16)[name = tensor<string, []>("op_36496_cast_fp16")];
            tensor<int32, [4]> var_36500_begin_0 = const()[name = tensor<string, []>("op_36500_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_36500_end_0 = const()[name = tensor<string, []>("op_36500_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_36500_end_mask_0 = const()[name = tensor<string, []>("op_36500_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_36500_cast_fp16 = slice_by_index(begin = var_36500_begin_0, end = var_36500_end_0, end_mask = var_36500_end_mask_0, x = k_327_cast_fp16)[name = tensor<string, []>("op_36500_cast_fp16")];
            tensor<int32, [4]> var_36504_begin_0 = const()[name = tensor<string, []>("op_36504_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_36504_end_0 = const()[name = tensor<string, []>("op_36504_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_36504_end_mask_0 = const()[name = tensor<string, []>("op_36504_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_36504_cast_fp16 = slice_by_index(begin = var_36504_begin_0, end = var_36504_end_0, end_mask = var_36504_end_mask_0, x = k_327_cast_fp16)[name = tensor<string, []>("op_36504_cast_fp16")];
            tensor<int32, [4]> var_36508_begin_0 = const()[name = tensor<string, []>("op_36508_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_36508_end_0 = const()[name = tensor<string, []>("op_36508_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_36508_end_mask_0 = const()[name = tensor<string, []>("op_36508_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_36508_cast_fp16 = slice_by_index(begin = var_36508_begin_0, end = var_36508_end_0, end_mask = var_36508_end_mask_0, x = k_327_cast_fp16)[name = tensor<string, []>("op_36508_cast_fp16")];
            tensor<int32, [4]> var_36512_begin_0 = const()[name = tensor<string, []>("op_36512_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_36512_end_0 = const()[name = tensor<string, []>("op_36512_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_36512_end_mask_0 = const()[name = tensor<string, []>("op_36512_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_36512_cast_fp16 = slice_by_index(begin = var_36512_begin_0, end = var_36512_end_0, end_mask = var_36512_end_mask_0, x = k_327_cast_fp16)[name = tensor<string, []>("op_36512_cast_fp16")];
            tensor<int32, [4]> var_36516_begin_0 = const()[name = tensor<string, []>("op_36516_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_36516_end_0 = const()[name = tensor<string, []>("op_36516_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_36516_end_mask_0 = const()[name = tensor<string, []>("op_36516_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_36516_cast_fp16 = slice_by_index(begin = var_36516_begin_0, end = var_36516_end_0, end_mask = var_36516_end_mask_0, x = k_327_cast_fp16)[name = tensor<string, []>("op_36516_cast_fp16")];
            tensor<int32, [4]> var_36520_begin_0 = const()[name = tensor<string, []>("op_36520_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_36520_end_0 = const()[name = tensor<string, []>("op_36520_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_36520_end_mask_0 = const()[name = tensor<string, []>("op_36520_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_36520_cast_fp16 = slice_by_index(begin = var_36520_begin_0, end = var_36520_end_0, end_mask = var_36520_end_mask_0, x = k_327_cast_fp16)[name = tensor<string, []>("op_36520_cast_fp16")];
            tensor<int32, [4]> var_36524_begin_0 = const()[name = tensor<string, []>("op_36524_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_36524_end_0 = const()[name = tensor<string, []>("op_36524_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_36524_end_mask_0 = const()[name = tensor<string, []>("op_36524_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_36524_cast_fp16 = slice_by_index(begin = var_36524_begin_0, end = var_36524_end_0, end_mask = var_36524_end_mask_0, x = k_327_cast_fp16)[name = tensor<string, []>("op_36524_cast_fp16")];
            tensor<int32, [4]> var_36528_begin_0 = const()[name = tensor<string, []>("op_36528_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_36528_end_0 = const()[name = tensor<string, []>("op_36528_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_36528_end_mask_0 = const()[name = tensor<string, []>("op_36528_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_36528_cast_fp16 = slice_by_index(begin = var_36528_begin_0, end = var_36528_end_0, end_mask = var_36528_end_mask_0, x = k_327_cast_fp16)[name = tensor<string, []>("op_36528_cast_fp16")];
            tensor<int32, [4]> var_36532_begin_0 = const()[name = tensor<string, []>("op_36532_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_36532_end_0 = const()[name = tensor<string, []>("op_36532_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_36532_end_mask_0 = const()[name = tensor<string, []>("op_36532_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_36532_cast_fp16 = slice_by_index(begin = var_36532_begin_0, end = var_36532_end_0, end_mask = var_36532_end_mask_0, x = k_327_cast_fp16)[name = tensor<string, []>("op_36532_cast_fp16")];
            tensor<int32, [4]> var_36536_begin_0 = const()[name = tensor<string, []>("op_36536_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_36536_end_0 = const()[name = tensor<string, []>("op_36536_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_36536_end_mask_0 = const()[name = tensor<string, []>("op_36536_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_36536_cast_fp16 = slice_by_index(begin = var_36536_begin_0, end = var_36536_end_0, end_mask = var_36536_end_mask_0, x = k_327_cast_fp16)[name = tensor<string, []>("op_36536_cast_fp16")];
            tensor<int32, [4]> var_36540_begin_0 = const()[name = tensor<string, []>("op_36540_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_36540_end_0 = const()[name = tensor<string, []>("op_36540_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_36540_end_mask_0 = const()[name = tensor<string, []>("op_36540_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_36540_cast_fp16 = slice_by_index(begin = var_36540_begin_0, end = var_36540_end_0, end_mask = var_36540_end_mask_0, x = k_327_cast_fp16)[name = tensor<string, []>("op_36540_cast_fp16")];
            tensor<int32, [4]> var_36544_begin_0 = const()[name = tensor<string, []>("op_36544_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_36544_end_0 = const()[name = tensor<string, []>("op_36544_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_36544_end_mask_0 = const()[name = tensor<string, []>("op_36544_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_36544_cast_fp16 = slice_by_index(begin = var_36544_begin_0, end = var_36544_end_0, end_mask = var_36544_end_mask_0, x = k_327_cast_fp16)[name = tensor<string, []>("op_36544_cast_fp16")];
            tensor<int32, [4]> var_36548_begin_0 = const()[name = tensor<string, []>("op_36548_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_36548_end_0 = const()[name = tensor<string, []>("op_36548_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_36548_end_mask_0 = const()[name = tensor<string, []>("op_36548_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_36548_cast_fp16 = slice_by_index(begin = var_36548_begin_0, end = var_36548_end_0, end_mask = var_36548_end_mask_0, x = k_327_cast_fp16)[name = tensor<string, []>("op_36548_cast_fp16")];
            tensor<int32, [4]> var_36552_begin_0 = const()[name = tensor<string, []>("op_36552_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_36552_end_0 = const()[name = tensor<string, []>("op_36552_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_36552_end_mask_0 = const()[name = tensor<string, []>("op_36552_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_36552_cast_fp16 = slice_by_index(begin = var_36552_begin_0, end = var_36552_end_0, end_mask = var_36552_end_mask_0, x = k_327_cast_fp16)[name = tensor<string, []>("op_36552_cast_fp16")];
            tensor<int32, [4]> var_36556_begin_0 = const()[name = tensor<string, []>("op_36556_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_36556_end_0 = const()[name = tensor<string, []>("op_36556_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_36556_end_mask_0 = const()[name = tensor<string, []>("op_36556_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_36556_cast_fp16 = slice_by_index(begin = var_36556_begin_0, end = var_36556_end_0, end_mask = var_36556_end_mask_0, x = k_327_cast_fp16)[name = tensor<string, []>("op_36556_cast_fp16")];
            tensor<int32, [4]> var_36560_begin_0 = const()[name = tensor<string, []>("op_36560_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_36560_end_0 = const()[name = tensor<string, []>("op_36560_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_36560_end_mask_0 = const()[name = tensor<string, []>("op_36560_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_36560_cast_fp16 = slice_by_index(begin = var_36560_begin_0, end = var_36560_end_0, end_mask = var_36560_end_mask_0, x = k_327_cast_fp16)[name = tensor<string, []>("op_36560_cast_fp16")];
            tensor<int32, [4]> var_36564_begin_0 = const()[name = tensor<string, []>("op_36564_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_36564_end_0 = const()[name = tensor<string, []>("op_36564_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_36564_end_mask_0 = const()[name = tensor<string, []>("op_36564_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_36564_cast_fp16 = slice_by_index(begin = var_36564_begin_0, end = var_36564_end_0, end_mask = var_36564_end_mask_0, x = k_327_cast_fp16)[name = tensor<string, []>("op_36564_cast_fp16")];
            tensor<int32, [4]> var_36566_begin_0 = const()[name = tensor<string, []>("op_36566_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_36566_end_0 = const()[name = tensor<string, []>("op_36566_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_36566_end_mask_0 = const()[name = tensor<string, []>("op_36566_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_36566_cast_fp16 = slice_by_index(begin = var_36566_begin_0, end = var_36566_end_0, end_mask = var_36566_end_mask_0, x = v_163_cast_fp16)[name = tensor<string, []>("op_36566_cast_fp16")];
            tensor<int32, [4]> var_36570_begin_0 = const()[name = tensor<string, []>("op_36570_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_36570_end_0 = const()[name = tensor<string, []>("op_36570_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_36570_end_mask_0 = const()[name = tensor<string, []>("op_36570_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_36570_cast_fp16 = slice_by_index(begin = var_36570_begin_0, end = var_36570_end_0, end_mask = var_36570_end_mask_0, x = v_163_cast_fp16)[name = tensor<string, []>("op_36570_cast_fp16")];
            tensor<int32, [4]> var_36574_begin_0 = const()[name = tensor<string, []>("op_36574_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_36574_end_0 = const()[name = tensor<string, []>("op_36574_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_36574_end_mask_0 = const()[name = tensor<string, []>("op_36574_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_36574_cast_fp16 = slice_by_index(begin = var_36574_begin_0, end = var_36574_end_0, end_mask = var_36574_end_mask_0, x = v_163_cast_fp16)[name = tensor<string, []>("op_36574_cast_fp16")];
            tensor<int32, [4]> var_36578_begin_0 = const()[name = tensor<string, []>("op_36578_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_36578_end_0 = const()[name = tensor<string, []>("op_36578_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_36578_end_mask_0 = const()[name = tensor<string, []>("op_36578_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_36578_cast_fp16 = slice_by_index(begin = var_36578_begin_0, end = var_36578_end_0, end_mask = var_36578_end_mask_0, x = v_163_cast_fp16)[name = tensor<string, []>("op_36578_cast_fp16")];
            tensor<int32, [4]> var_36582_begin_0 = const()[name = tensor<string, []>("op_36582_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_36582_end_0 = const()[name = tensor<string, []>("op_36582_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_36582_end_mask_0 = const()[name = tensor<string, []>("op_36582_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_36582_cast_fp16 = slice_by_index(begin = var_36582_begin_0, end = var_36582_end_0, end_mask = var_36582_end_mask_0, x = v_163_cast_fp16)[name = tensor<string, []>("op_36582_cast_fp16")];
            tensor<int32, [4]> var_36586_begin_0 = const()[name = tensor<string, []>("op_36586_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_36586_end_0 = const()[name = tensor<string, []>("op_36586_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_36586_end_mask_0 = const()[name = tensor<string, []>("op_36586_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_36586_cast_fp16 = slice_by_index(begin = var_36586_begin_0, end = var_36586_end_0, end_mask = var_36586_end_mask_0, x = v_163_cast_fp16)[name = tensor<string, []>("op_36586_cast_fp16")];
            tensor<int32, [4]> var_36590_begin_0 = const()[name = tensor<string, []>("op_36590_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_36590_end_0 = const()[name = tensor<string, []>("op_36590_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_36590_end_mask_0 = const()[name = tensor<string, []>("op_36590_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_36590_cast_fp16 = slice_by_index(begin = var_36590_begin_0, end = var_36590_end_0, end_mask = var_36590_end_mask_0, x = v_163_cast_fp16)[name = tensor<string, []>("op_36590_cast_fp16")];
            tensor<int32, [4]> var_36594_begin_0 = const()[name = tensor<string, []>("op_36594_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_36594_end_0 = const()[name = tensor<string, []>("op_36594_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_36594_end_mask_0 = const()[name = tensor<string, []>("op_36594_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_36594_cast_fp16 = slice_by_index(begin = var_36594_begin_0, end = var_36594_end_0, end_mask = var_36594_end_mask_0, x = v_163_cast_fp16)[name = tensor<string, []>("op_36594_cast_fp16")];
            tensor<int32, [4]> var_36598_begin_0 = const()[name = tensor<string, []>("op_36598_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_36598_end_0 = const()[name = tensor<string, []>("op_36598_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_36598_end_mask_0 = const()[name = tensor<string, []>("op_36598_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_36598_cast_fp16 = slice_by_index(begin = var_36598_begin_0, end = var_36598_end_0, end_mask = var_36598_end_mask_0, x = v_163_cast_fp16)[name = tensor<string, []>("op_36598_cast_fp16")];
            tensor<int32, [4]> var_36602_begin_0 = const()[name = tensor<string, []>("op_36602_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_36602_end_0 = const()[name = tensor<string, []>("op_36602_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_36602_end_mask_0 = const()[name = tensor<string, []>("op_36602_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_36602_cast_fp16 = slice_by_index(begin = var_36602_begin_0, end = var_36602_end_0, end_mask = var_36602_end_mask_0, x = v_163_cast_fp16)[name = tensor<string, []>("op_36602_cast_fp16")];
            tensor<int32, [4]> var_36606_begin_0 = const()[name = tensor<string, []>("op_36606_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_36606_end_0 = const()[name = tensor<string, []>("op_36606_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_36606_end_mask_0 = const()[name = tensor<string, []>("op_36606_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_36606_cast_fp16 = slice_by_index(begin = var_36606_begin_0, end = var_36606_end_0, end_mask = var_36606_end_mask_0, x = v_163_cast_fp16)[name = tensor<string, []>("op_36606_cast_fp16")];
            tensor<int32, [4]> var_36610_begin_0 = const()[name = tensor<string, []>("op_36610_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_36610_end_0 = const()[name = tensor<string, []>("op_36610_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_36610_end_mask_0 = const()[name = tensor<string, []>("op_36610_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_36610_cast_fp16 = slice_by_index(begin = var_36610_begin_0, end = var_36610_end_0, end_mask = var_36610_end_mask_0, x = v_163_cast_fp16)[name = tensor<string, []>("op_36610_cast_fp16")];
            tensor<int32, [4]> var_36614_begin_0 = const()[name = tensor<string, []>("op_36614_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_36614_end_0 = const()[name = tensor<string, []>("op_36614_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_36614_end_mask_0 = const()[name = tensor<string, []>("op_36614_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_36614_cast_fp16 = slice_by_index(begin = var_36614_begin_0, end = var_36614_end_0, end_mask = var_36614_end_mask_0, x = v_163_cast_fp16)[name = tensor<string, []>("op_36614_cast_fp16")];
            tensor<int32, [4]> var_36618_begin_0 = const()[name = tensor<string, []>("op_36618_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_36618_end_0 = const()[name = tensor<string, []>("op_36618_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_36618_end_mask_0 = const()[name = tensor<string, []>("op_36618_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_36618_cast_fp16 = slice_by_index(begin = var_36618_begin_0, end = var_36618_end_0, end_mask = var_36618_end_mask_0, x = v_163_cast_fp16)[name = tensor<string, []>("op_36618_cast_fp16")];
            tensor<int32, [4]> var_36622_begin_0 = const()[name = tensor<string, []>("op_36622_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_36622_end_0 = const()[name = tensor<string, []>("op_36622_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_36622_end_mask_0 = const()[name = tensor<string, []>("op_36622_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_36622_cast_fp16 = slice_by_index(begin = var_36622_begin_0, end = var_36622_end_0, end_mask = var_36622_end_mask_0, x = v_163_cast_fp16)[name = tensor<string, []>("op_36622_cast_fp16")];
            tensor<int32, [4]> var_36626_begin_0 = const()[name = tensor<string, []>("op_36626_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_36626_end_0 = const()[name = tensor<string, []>("op_36626_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_36626_end_mask_0 = const()[name = tensor<string, []>("op_36626_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_36626_cast_fp16 = slice_by_index(begin = var_36626_begin_0, end = var_36626_end_0, end_mask = var_36626_end_mask_0, x = v_163_cast_fp16)[name = tensor<string, []>("op_36626_cast_fp16")];
            tensor<int32, [4]> var_36630_begin_0 = const()[name = tensor<string, []>("op_36630_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_36630_end_0 = const()[name = tensor<string, []>("op_36630_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_36630_end_mask_0 = const()[name = tensor<string, []>("op_36630_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_36630_cast_fp16 = slice_by_index(begin = var_36630_begin_0, end = var_36630_end_0, end_mask = var_36630_end_mask_0, x = v_163_cast_fp16)[name = tensor<string, []>("op_36630_cast_fp16")];
            tensor<int32, [4]> var_36634_begin_0 = const()[name = tensor<string, []>("op_36634_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_36634_end_0 = const()[name = tensor<string, []>("op_36634_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_36634_end_mask_0 = const()[name = tensor<string, []>("op_36634_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_36634_cast_fp16 = slice_by_index(begin = var_36634_begin_0, end = var_36634_end_0, end_mask = var_36634_end_mask_0, x = v_163_cast_fp16)[name = tensor<string, []>("op_36634_cast_fp16")];
            tensor<int32, [4]> var_36638_begin_0 = const()[name = tensor<string, []>("op_36638_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_36638_end_0 = const()[name = tensor<string, []>("op_36638_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_36638_end_mask_0 = const()[name = tensor<string, []>("op_36638_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_36638_cast_fp16 = slice_by_index(begin = var_36638_begin_0, end = var_36638_end_0, end_mask = var_36638_end_mask_0, x = v_163_cast_fp16)[name = tensor<string, []>("op_36638_cast_fp16")];
            tensor<int32, [4]> var_36642_begin_0 = const()[name = tensor<string, []>("op_36642_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_36642_end_0 = const()[name = tensor<string, []>("op_36642_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_36642_end_mask_0 = const()[name = tensor<string, []>("op_36642_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_36642_cast_fp16 = slice_by_index(begin = var_36642_begin_0, end = var_36642_end_0, end_mask = var_36642_end_mask_0, x = v_163_cast_fp16)[name = tensor<string, []>("op_36642_cast_fp16")];
            tensor<string, []> var_36646_equation_0 = const()[name = tensor<string, []>("op_36646_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_36646_cast_fp16 = einsum(equation = var_36646_equation_0, values = (var_36488_cast_fp16, var_36405_cast_fp16))[name = tensor<string, []>("op_36646_cast_fp16")];
            tensor<fp16, []> var_36647_to_fp16 = const()[name = tensor<string, []>("op_36647_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3081_cast_fp16 = mul(x = var_36646_cast_fp16, y = var_36647_to_fp16)[name = tensor<string, []>("aw_3081_cast_fp16")];
            tensor<string, []> var_36650_equation_0 = const()[name = tensor<string, []>("op_36650_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_36650_cast_fp16 = einsum(equation = var_36650_equation_0, values = (var_36492_cast_fp16, var_36409_cast_fp16))[name = tensor<string, []>("op_36650_cast_fp16")];
            tensor<fp16, []> var_36651_to_fp16 = const()[name = tensor<string, []>("op_36651_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3083_cast_fp16 = mul(x = var_36650_cast_fp16, y = var_36651_to_fp16)[name = tensor<string, []>("aw_3083_cast_fp16")];
            tensor<string, []> var_36654_equation_0 = const()[name = tensor<string, []>("op_36654_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_36654_cast_fp16 = einsum(equation = var_36654_equation_0, values = (var_36496_cast_fp16, var_36413_cast_fp16))[name = tensor<string, []>("op_36654_cast_fp16")];
            tensor<fp16, []> var_36655_to_fp16 = const()[name = tensor<string, []>("op_36655_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3085_cast_fp16 = mul(x = var_36654_cast_fp16, y = var_36655_to_fp16)[name = tensor<string, []>("aw_3085_cast_fp16")];
            tensor<string, []> var_36658_equation_0 = const()[name = tensor<string, []>("op_36658_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_36658_cast_fp16 = einsum(equation = var_36658_equation_0, values = (var_36500_cast_fp16, var_36417_cast_fp16))[name = tensor<string, []>("op_36658_cast_fp16")];
            tensor<fp16, []> var_36659_to_fp16 = const()[name = tensor<string, []>("op_36659_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3087_cast_fp16 = mul(x = var_36658_cast_fp16, y = var_36659_to_fp16)[name = tensor<string, []>("aw_3087_cast_fp16")];
            tensor<string, []> var_36662_equation_0 = const()[name = tensor<string, []>("op_36662_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_36662_cast_fp16 = einsum(equation = var_36662_equation_0, values = (var_36504_cast_fp16, var_36421_cast_fp16))[name = tensor<string, []>("op_36662_cast_fp16")];
            tensor<fp16, []> var_36663_to_fp16 = const()[name = tensor<string, []>("op_36663_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3089_cast_fp16 = mul(x = var_36662_cast_fp16, y = var_36663_to_fp16)[name = tensor<string, []>("aw_3089_cast_fp16")];
            tensor<string, []> var_36666_equation_0 = const()[name = tensor<string, []>("op_36666_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_36666_cast_fp16 = einsum(equation = var_36666_equation_0, values = (var_36508_cast_fp16, var_36425_cast_fp16))[name = tensor<string, []>("op_36666_cast_fp16")];
            tensor<fp16, []> var_36667_to_fp16 = const()[name = tensor<string, []>("op_36667_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3091_cast_fp16 = mul(x = var_36666_cast_fp16, y = var_36667_to_fp16)[name = tensor<string, []>("aw_3091_cast_fp16")];
            tensor<string, []> var_36670_equation_0 = const()[name = tensor<string, []>("op_36670_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_36670_cast_fp16 = einsum(equation = var_36670_equation_0, values = (var_36512_cast_fp16, var_36429_cast_fp16))[name = tensor<string, []>("op_36670_cast_fp16")];
            tensor<fp16, []> var_36671_to_fp16 = const()[name = tensor<string, []>("op_36671_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3093_cast_fp16 = mul(x = var_36670_cast_fp16, y = var_36671_to_fp16)[name = tensor<string, []>("aw_3093_cast_fp16")];
            tensor<string, []> var_36674_equation_0 = const()[name = tensor<string, []>("op_36674_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_36674_cast_fp16 = einsum(equation = var_36674_equation_0, values = (var_36516_cast_fp16, var_36433_cast_fp16))[name = tensor<string, []>("op_36674_cast_fp16")];
            tensor<fp16, []> var_36675_to_fp16 = const()[name = tensor<string, []>("op_36675_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3095_cast_fp16 = mul(x = var_36674_cast_fp16, y = var_36675_to_fp16)[name = tensor<string, []>("aw_3095_cast_fp16")];
            tensor<string, []> var_36678_equation_0 = const()[name = tensor<string, []>("op_36678_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_36678_cast_fp16 = einsum(equation = var_36678_equation_0, values = (var_36520_cast_fp16, var_36437_cast_fp16))[name = tensor<string, []>("op_36678_cast_fp16")];
            tensor<fp16, []> var_36679_to_fp16 = const()[name = tensor<string, []>("op_36679_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3097_cast_fp16 = mul(x = var_36678_cast_fp16, y = var_36679_to_fp16)[name = tensor<string, []>("aw_3097_cast_fp16")];
            tensor<string, []> var_36682_equation_0 = const()[name = tensor<string, []>("op_36682_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_36682_cast_fp16 = einsum(equation = var_36682_equation_0, values = (var_36524_cast_fp16, var_36441_cast_fp16))[name = tensor<string, []>("op_36682_cast_fp16")];
            tensor<fp16, []> var_36683_to_fp16 = const()[name = tensor<string, []>("op_36683_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3099_cast_fp16 = mul(x = var_36682_cast_fp16, y = var_36683_to_fp16)[name = tensor<string, []>("aw_3099_cast_fp16")];
            tensor<string, []> var_36686_equation_0 = const()[name = tensor<string, []>("op_36686_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_36686_cast_fp16 = einsum(equation = var_36686_equation_0, values = (var_36528_cast_fp16, var_36445_cast_fp16))[name = tensor<string, []>("op_36686_cast_fp16")];
            tensor<fp16, []> var_36687_to_fp16 = const()[name = tensor<string, []>("op_36687_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3101_cast_fp16 = mul(x = var_36686_cast_fp16, y = var_36687_to_fp16)[name = tensor<string, []>("aw_3101_cast_fp16")];
            tensor<string, []> var_36690_equation_0 = const()[name = tensor<string, []>("op_36690_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_36690_cast_fp16 = einsum(equation = var_36690_equation_0, values = (var_36532_cast_fp16, var_36449_cast_fp16))[name = tensor<string, []>("op_36690_cast_fp16")];
            tensor<fp16, []> var_36691_to_fp16 = const()[name = tensor<string, []>("op_36691_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3103_cast_fp16 = mul(x = var_36690_cast_fp16, y = var_36691_to_fp16)[name = tensor<string, []>("aw_3103_cast_fp16")];
            tensor<string, []> var_36694_equation_0 = const()[name = tensor<string, []>("op_36694_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_36694_cast_fp16 = einsum(equation = var_36694_equation_0, values = (var_36536_cast_fp16, var_36453_cast_fp16))[name = tensor<string, []>("op_36694_cast_fp16")];
            tensor<fp16, []> var_36695_to_fp16 = const()[name = tensor<string, []>("op_36695_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3105_cast_fp16 = mul(x = var_36694_cast_fp16, y = var_36695_to_fp16)[name = tensor<string, []>("aw_3105_cast_fp16")];
            tensor<string, []> var_36698_equation_0 = const()[name = tensor<string, []>("op_36698_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_36698_cast_fp16 = einsum(equation = var_36698_equation_0, values = (var_36540_cast_fp16, var_36457_cast_fp16))[name = tensor<string, []>("op_36698_cast_fp16")];
            tensor<fp16, []> var_36699_to_fp16 = const()[name = tensor<string, []>("op_36699_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3107_cast_fp16 = mul(x = var_36698_cast_fp16, y = var_36699_to_fp16)[name = tensor<string, []>("aw_3107_cast_fp16")];
            tensor<string, []> var_36702_equation_0 = const()[name = tensor<string, []>("op_36702_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_36702_cast_fp16 = einsum(equation = var_36702_equation_0, values = (var_36544_cast_fp16, var_36461_cast_fp16))[name = tensor<string, []>("op_36702_cast_fp16")];
            tensor<fp16, []> var_36703_to_fp16 = const()[name = tensor<string, []>("op_36703_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3109_cast_fp16 = mul(x = var_36702_cast_fp16, y = var_36703_to_fp16)[name = tensor<string, []>("aw_3109_cast_fp16")];
            tensor<string, []> var_36706_equation_0 = const()[name = tensor<string, []>("op_36706_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_36706_cast_fp16 = einsum(equation = var_36706_equation_0, values = (var_36548_cast_fp16, var_36465_cast_fp16))[name = tensor<string, []>("op_36706_cast_fp16")];
            tensor<fp16, []> var_36707_to_fp16 = const()[name = tensor<string, []>("op_36707_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3111_cast_fp16 = mul(x = var_36706_cast_fp16, y = var_36707_to_fp16)[name = tensor<string, []>("aw_3111_cast_fp16")];
            tensor<string, []> var_36710_equation_0 = const()[name = tensor<string, []>("op_36710_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_36710_cast_fp16 = einsum(equation = var_36710_equation_0, values = (var_36552_cast_fp16, var_36469_cast_fp16))[name = tensor<string, []>("op_36710_cast_fp16")];
            tensor<fp16, []> var_36711_to_fp16 = const()[name = tensor<string, []>("op_36711_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3113_cast_fp16 = mul(x = var_36710_cast_fp16, y = var_36711_to_fp16)[name = tensor<string, []>("aw_3113_cast_fp16")];
            tensor<string, []> var_36714_equation_0 = const()[name = tensor<string, []>("op_36714_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_36714_cast_fp16 = einsum(equation = var_36714_equation_0, values = (var_36556_cast_fp16, var_36473_cast_fp16))[name = tensor<string, []>("op_36714_cast_fp16")];
            tensor<fp16, []> var_36715_to_fp16 = const()[name = tensor<string, []>("op_36715_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3115_cast_fp16 = mul(x = var_36714_cast_fp16, y = var_36715_to_fp16)[name = tensor<string, []>("aw_3115_cast_fp16")];
            tensor<string, []> var_36718_equation_0 = const()[name = tensor<string, []>("op_36718_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_36718_cast_fp16 = einsum(equation = var_36718_equation_0, values = (var_36560_cast_fp16, var_36477_cast_fp16))[name = tensor<string, []>("op_36718_cast_fp16")];
            tensor<fp16, []> var_36719_to_fp16 = const()[name = tensor<string, []>("op_36719_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3117_cast_fp16 = mul(x = var_36718_cast_fp16, y = var_36719_to_fp16)[name = tensor<string, []>("aw_3117_cast_fp16")];
            tensor<string, []> var_36722_equation_0 = const()[name = tensor<string, []>("op_36722_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_36722_cast_fp16 = einsum(equation = var_36722_equation_0, values = (var_36564_cast_fp16, var_36481_cast_fp16))[name = tensor<string, []>("op_36722_cast_fp16")];
            tensor<fp16, []> var_36723_to_fp16 = const()[name = tensor<string, []>("op_36723_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3119_cast_fp16 = mul(x = var_36722_cast_fp16, y = var_36723_to_fp16)[name = tensor<string, []>("aw_3119_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_36725_cast_fp16 = softmax(axis = var_30355, x = aw_3081_cast_fp16)[name = tensor<string, []>("op_36725_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_36726_cast_fp16 = softmax(axis = var_30355, x = aw_3083_cast_fp16)[name = tensor<string, []>("op_36726_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_36727_cast_fp16 = softmax(axis = var_30355, x = aw_3085_cast_fp16)[name = tensor<string, []>("op_36727_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_36728_cast_fp16 = softmax(axis = var_30355, x = aw_3087_cast_fp16)[name = tensor<string, []>("op_36728_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_36729_cast_fp16 = softmax(axis = var_30355, x = aw_3089_cast_fp16)[name = tensor<string, []>("op_36729_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_36730_cast_fp16 = softmax(axis = var_30355, x = aw_3091_cast_fp16)[name = tensor<string, []>("op_36730_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_36731_cast_fp16 = softmax(axis = var_30355, x = aw_3093_cast_fp16)[name = tensor<string, []>("op_36731_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_36732_cast_fp16 = softmax(axis = var_30355, x = aw_3095_cast_fp16)[name = tensor<string, []>("op_36732_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_36733_cast_fp16 = softmax(axis = var_30355, x = aw_3097_cast_fp16)[name = tensor<string, []>("op_36733_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_36734_cast_fp16 = softmax(axis = var_30355, x = aw_3099_cast_fp16)[name = tensor<string, []>("op_36734_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_36735_cast_fp16 = softmax(axis = var_30355, x = aw_3101_cast_fp16)[name = tensor<string, []>("op_36735_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_36736_cast_fp16 = softmax(axis = var_30355, x = aw_3103_cast_fp16)[name = tensor<string, []>("op_36736_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_36737_cast_fp16 = softmax(axis = var_30355, x = aw_3105_cast_fp16)[name = tensor<string, []>("op_36737_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_36738_cast_fp16 = softmax(axis = var_30355, x = aw_3107_cast_fp16)[name = tensor<string, []>("op_36738_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_36739_cast_fp16 = softmax(axis = var_30355, x = aw_3109_cast_fp16)[name = tensor<string, []>("op_36739_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_36740_cast_fp16 = softmax(axis = var_30355, x = aw_3111_cast_fp16)[name = tensor<string, []>("op_36740_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_36741_cast_fp16 = softmax(axis = var_30355, x = aw_3113_cast_fp16)[name = tensor<string, []>("op_36741_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_36742_cast_fp16 = softmax(axis = var_30355, x = aw_3115_cast_fp16)[name = tensor<string, []>("op_36742_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_36743_cast_fp16 = softmax(axis = var_30355, x = aw_3117_cast_fp16)[name = tensor<string, []>("op_36743_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_36744_cast_fp16 = softmax(axis = var_30355, x = aw_3119_cast_fp16)[name = tensor<string, []>("op_36744_cast_fp16")];
            tensor<string, []> var_36746_equation_0 = const()[name = tensor<string, []>("op_36746_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36746_cast_fp16 = einsum(equation = var_36746_equation_0, values = (var_36566_cast_fp16, var_36725_cast_fp16))[name = tensor<string, []>("op_36746_cast_fp16")];
            tensor<string, []> var_36748_equation_0 = const()[name = tensor<string, []>("op_36748_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36748_cast_fp16 = einsum(equation = var_36748_equation_0, values = (var_36570_cast_fp16, var_36726_cast_fp16))[name = tensor<string, []>("op_36748_cast_fp16")];
            tensor<string, []> var_36750_equation_0 = const()[name = tensor<string, []>("op_36750_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36750_cast_fp16 = einsum(equation = var_36750_equation_0, values = (var_36574_cast_fp16, var_36727_cast_fp16))[name = tensor<string, []>("op_36750_cast_fp16")];
            tensor<string, []> var_36752_equation_0 = const()[name = tensor<string, []>("op_36752_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36752_cast_fp16 = einsum(equation = var_36752_equation_0, values = (var_36578_cast_fp16, var_36728_cast_fp16))[name = tensor<string, []>("op_36752_cast_fp16")];
            tensor<string, []> var_36754_equation_0 = const()[name = tensor<string, []>("op_36754_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36754_cast_fp16 = einsum(equation = var_36754_equation_0, values = (var_36582_cast_fp16, var_36729_cast_fp16))[name = tensor<string, []>("op_36754_cast_fp16")];
            tensor<string, []> var_36756_equation_0 = const()[name = tensor<string, []>("op_36756_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36756_cast_fp16 = einsum(equation = var_36756_equation_0, values = (var_36586_cast_fp16, var_36730_cast_fp16))[name = tensor<string, []>("op_36756_cast_fp16")];
            tensor<string, []> var_36758_equation_0 = const()[name = tensor<string, []>("op_36758_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36758_cast_fp16 = einsum(equation = var_36758_equation_0, values = (var_36590_cast_fp16, var_36731_cast_fp16))[name = tensor<string, []>("op_36758_cast_fp16")];
            tensor<string, []> var_36760_equation_0 = const()[name = tensor<string, []>("op_36760_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36760_cast_fp16 = einsum(equation = var_36760_equation_0, values = (var_36594_cast_fp16, var_36732_cast_fp16))[name = tensor<string, []>("op_36760_cast_fp16")];
            tensor<string, []> var_36762_equation_0 = const()[name = tensor<string, []>("op_36762_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36762_cast_fp16 = einsum(equation = var_36762_equation_0, values = (var_36598_cast_fp16, var_36733_cast_fp16))[name = tensor<string, []>("op_36762_cast_fp16")];
            tensor<string, []> var_36764_equation_0 = const()[name = tensor<string, []>("op_36764_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36764_cast_fp16 = einsum(equation = var_36764_equation_0, values = (var_36602_cast_fp16, var_36734_cast_fp16))[name = tensor<string, []>("op_36764_cast_fp16")];
            tensor<string, []> var_36766_equation_0 = const()[name = tensor<string, []>("op_36766_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36766_cast_fp16 = einsum(equation = var_36766_equation_0, values = (var_36606_cast_fp16, var_36735_cast_fp16))[name = tensor<string, []>("op_36766_cast_fp16")];
            tensor<string, []> var_36768_equation_0 = const()[name = tensor<string, []>("op_36768_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36768_cast_fp16 = einsum(equation = var_36768_equation_0, values = (var_36610_cast_fp16, var_36736_cast_fp16))[name = tensor<string, []>("op_36768_cast_fp16")];
            tensor<string, []> var_36770_equation_0 = const()[name = tensor<string, []>("op_36770_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36770_cast_fp16 = einsum(equation = var_36770_equation_0, values = (var_36614_cast_fp16, var_36737_cast_fp16))[name = tensor<string, []>("op_36770_cast_fp16")];
            tensor<string, []> var_36772_equation_0 = const()[name = tensor<string, []>("op_36772_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36772_cast_fp16 = einsum(equation = var_36772_equation_0, values = (var_36618_cast_fp16, var_36738_cast_fp16))[name = tensor<string, []>("op_36772_cast_fp16")];
            tensor<string, []> var_36774_equation_0 = const()[name = tensor<string, []>("op_36774_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36774_cast_fp16 = einsum(equation = var_36774_equation_0, values = (var_36622_cast_fp16, var_36739_cast_fp16))[name = tensor<string, []>("op_36774_cast_fp16")];
            tensor<string, []> var_36776_equation_0 = const()[name = tensor<string, []>("op_36776_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36776_cast_fp16 = einsum(equation = var_36776_equation_0, values = (var_36626_cast_fp16, var_36740_cast_fp16))[name = tensor<string, []>("op_36776_cast_fp16")];
            tensor<string, []> var_36778_equation_0 = const()[name = tensor<string, []>("op_36778_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36778_cast_fp16 = einsum(equation = var_36778_equation_0, values = (var_36630_cast_fp16, var_36741_cast_fp16))[name = tensor<string, []>("op_36778_cast_fp16")];
            tensor<string, []> var_36780_equation_0 = const()[name = tensor<string, []>("op_36780_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36780_cast_fp16 = einsum(equation = var_36780_equation_0, values = (var_36634_cast_fp16, var_36742_cast_fp16))[name = tensor<string, []>("op_36780_cast_fp16")];
            tensor<string, []> var_36782_equation_0 = const()[name = tensor<string, []>("op_36782_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36782_cast_fp16 = einsum(equation = var_36782_equation_0, values = (var_36638_cast_fp16, var_36743_cast_fp16))[name = tensor<string, []>("op_36782_cast_fp16")];
            tensor<string, []> var_36784_equation_0 = const()[name = tensor<string, []>("op_36784_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36784_cast_fp16 = einsum(equation = var_36784_equation_0, values = (var_36642_cast_fp16, var_36744_cast_fp16))[name = tensor<string, []>("op_36784_cast_fp16")];
            tensor<bool, []> input_489_interleave_0 = const()[name = tensor<string, []>("input_489_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_489_cast_fp16 = concat(axis = var_30355, interleave = input_489_interleave_0, values = (var_36746_cast_fp16, var_36748_cast_fp16, var_36750_cast_fp16, var_36752_cast_fp16, var_36754_cast_fp16, var_36756_cast_fp16, var_36758_cast_fp16, var_36760_cast_fp16, var_36762_cast_fp16, var_36764_cast_fp16, var_36766_cast_fp16, var_36768_cast_fp16, var_36770_cast_fp16, var_36772_cast_fp16, var_36774_cast_fp16, var_36776_cast_fp16, var_36778_cast_fp16, var_36780_cast_fp16, var_36782_cast_fp16, var_36784_cast_fp16))[name = tensor<string, []>("input_489_cast_fp16")];
            tensor<string, []> var_36794_pad_type_0 = const()[name = tensor<string, []>("op_36794_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_36794_strides_0 = const()[name = tensor<string, []>("op_36794_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_36794_pad_0 = const()[name = tensor<string, []>("op_36794_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_36794_dilations_0 = const()[name = tensor<string, []>("op_36794_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_36794_groups_0 = const()[name = tensor<string, []>("op_36794_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_6_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(170391936))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(171620800))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_6_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_6_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_6_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(171620992)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_36794_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_6_attn2_to_out_0_bias_to_fp16, dilations = var_36794_dilations_0, groups = var_36794_groups_0, pad = var_36794_pad_0, pad_type = var_36794_pad_type_0, strides = var_36794_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_6_attn2_to_out_0_weight_to_fp16_palettized, x = input_489_cast_fp16)[name = tensor<string, []>("op_36794_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_245_cast_fp16 = add(x = var_36794_cast_fp16, y = inputs_243_cast_fp16)[name = tensor<string, []>("inputs_245_cast_fp16")];
            tensor<int32, [1]> input_491_axes_0 = const()[name = tensor<string, []>("input_491_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> input_491_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_491_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(171623616)))];
            tensor<fp16, [1280]> input_491_beta_0_to_fp16 = const()[name = tensor<string, []>("input_491_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(171626240)))];
            tensor<fp16, []> var_36804_to_fp16 = const()[name = tensor<string, []>("op_36804_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> input_491_cast_fp16 = layer_norm(axes = input_491_axes_0, beta = input_491_beta_0_to_fp16, epsilon = var_36804_to_fp16, gamma = input_491_gamma_0_to_fp16, x = inputs_245_cast_fp16)[name = tensor<string, []>("input_491_cast_fp16")];
            tensor<string, []> var_36824_pad_type_0 = const()[name = tensor<string, []>("op_36824_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_36824_strides_0 = const()[name = tensor<string, []>("op_36824_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_36824_pad_0 = const()[name = tensor<string, []>("op_36824_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_36824_dilations_0 = const()[name = tensor<string, []>("op_36824_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_36824_groups_0 = const()[name = tensor<string, []>("op_36824_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_6_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [9830400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(171628864))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(181459328))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_6_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([10240, 1280, 1, 1])];
            tensor<fp16, [10240]> up_blocks_0_attentions_0_transformer_blocks_6_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_6_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(181459520)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_36824_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_6_ff_net_0_proj_bias_to_fp16, dilations = var_36824_dilations_0, groups = var_36824_groups_0, pad = var_36824_pad_0, pad_type = var_36824_pad_type_0, strides = var_36824_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_6_ff_net_0_proj_weight_to_fp16_palettized, x = input_491_cast_fp16)[name = tensor<string, []>("op_36824_cast_fp16")];
            tensor<int32, [2]> var_36825_split_sizes_0 = const()[name = tensor<string, []>("op_36825_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_36825_axis_0 = const()[name = tensor<string, []>("op_36825_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_36825_cast_fp16_0, tensor<fp16, [2, 5120, 1, 1024]> var_36825_cast_fp16_1 = split(axis = var_36825_axis_0, split_sizes = var_36825_split_sizes_0, x = var_36824_cast_fp16)[name = tensor<string, []>("op_36825_cast_fp16")];
            tensor<string, []> var_36827_mode_0 = const()[name = tensor<string, []>("op_36827_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_36827_cast_fp16 = gelu(mode = var_36827_mode_0, x = var_36825_cast_fp16_1)[name = tensor<string, []>("op_36827_cast_fp16")];
            tensor<fp16, [2, 5120, 1, 1024]> input_493_cast_fp16 = mul(x = var_36825_cast_fp16_0, y = var_36827_cast_fp16)[name = tensor<string, []>("input_493_cast_fp16")];
            tensor<string, []> var_36835_pad_type_0 = const()[name = tensor<string, []>("op_36835_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_36835_strides_0 = const()[name = tensor<string, []>("op_36835_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_36835_pad_0 = const()[name = tensor<string, []>("op_36835_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_36835_dilations_0 = const()[name = tensor<string, []>("op_36835_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_36835_groups_0 = const()[name = tensor<string, []>("op_36835_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_6_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [4915200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(181480064))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(186395328))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_6_ff_net_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 5120, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_6_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_6_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(186395520)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_36835_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_6_ff_net_2_bias_to_fp16, dilations = var_36835_dilations_0, groups = var_36835_groups_0, pad = var_36835_pad_0, pad_type = var_36835_pad_type_0, strides = var_36835_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_6_ff_net_2_weight_to_fp16_palettized, x = input_493_cast_fp16)[name = tensor<string, []>("op_36835_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_247_cast_fp16 = add(x = var_36835_cast_fp16, y = inputs_245_cast_fp16)[name = tensor<string, []>("inputs_247_cast_fp16")];
            tensor<int32, [1]> hidden_states_329_axes_0 = const()[name = tensor<string, []>("hidden_states_329_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_329_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_329_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(186398144)))];
            tensor<fp16, [1280]> hidden_states_329_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_329_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(186400768)))];
            tensor<fp16, []> var_36851_to_fp16 = const()[name = tensor<string, []>("op_36851_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_329_cast_fp16 = layer_norm(axes = hidden_states_329_axes_0, beta = hidden_states_329_beta_0_to_fp16, epsilon = var_36851_to_fp16, gamma = hidden_states_329_gamma_0_to_fp16, x = inputs_247_cast_fp16)[name = tensor<string, []>("hidden_states_329_cast_fp16")];
            tensor<string, []> q_165_pad_type_0 = const()[name = tensor<string, []>("q_165_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_165_strides_0 = const()[name = tensor<string, []>("q_165_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_165_pad_0 = const()[name = tensor<string, []>("q_165_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_165_dilations_0 = const()[name = tensor<string, []>("q_165_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_165_groups_0 = const()[name = tensor<string, []>("q_165_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_7_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(186403392))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(187632256))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_7_attn1_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_165_cast_fp16 = conv(dilations = q_165_dilations_0, groups = q_165_groups_0, pad = q_165_pad_0, pad_type = q_165_pad_type_0, strides = q_165_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_7_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_329_cast_fp16)[name = tensor<string, []>("q_165_cast_fp16")];
            tensor<string, []> k_329_pad_type_0 = const()[name = tensor<string, []>("k_329_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_329_strides_0 = const()[name = tensor<string, []>("k_329_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_329_pad_0 = const()[name = tensor<string, []>("k_329_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_329_dilations_0 = const()[name = tensor<string, []>("k_329_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_329_groups_0 = const()[name = tensor<string, []>("k_329_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_7_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(187632448))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(188861312))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_7_attn1_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> k_329_cast_fp16 = conv(dilations = k_329_dilations_0, groups = k_329_groups_0, pad = k_329_pad_0, pad_type = k_329_pad_type_0, strides = k_329_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_7_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_329_cast_fp16)[name = tensor<string, []>("k_329_cast_fp16")];
            tensor<string, []> v_165_pad_type_0 = const()[name = tensor<string, []>("v_165_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_165_strides_0 = const()[name = tensor<string, []>("v_165_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_165_pad_0 = const()[name = tensor<string, []>("v_165_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_165_dilations_0 = const()[name = tensor<string, []>("v_165_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_165_groups_0 = const()[name = tensor<string, []>("v_165_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_7_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(188861504))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(190090368))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_7_attn1_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> v_165_cast_fp16 = conv(dilations = v_165_dilations_0, groups = v_165_groups_0, pad = v_165_pad_0, pad_type = v_165_pad_type_0, strides = v_165_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_7_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_329_cast_fp16)[name = tensor<string, []>("v_165_cast_fp16")];
            tensor<int32, [4]> var_36884_begin_0 = const()[name = tensor<string, []>("op_36884_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_36884_end_0 = const()[name = tensor<string, []>("op_36884_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_36884_end_mask_0 = const()[name = tensor<string, []>("op_36884_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36884_cast_fp16 = slice_by_index(begin = var_36884_begin_0, end = var_36884_end_0, end_mask = var_36884_end_mask_0, x = q_165_cast_fp16)[name = tensor<string, []>("op_36884_cast_fp16")];
            tensor<int32, [4]> var_36888_begin_0 = const()[name = tensor<string, []>("op_36888_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_36888_end_0 = const()[name = tensor<string, []>("op_36888_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_36888_end_mask_0 = const()[name = tensor<string, []>("op_36888_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36888_cast_fp16 = slice_by_index(begin = var_36888_begin_0, end = var_36888_end_0, end_mask = var_36888_end_mask_0, x = q_165_cast_fp16)[name = tensor<string, []>("op_36888_cast_fp16")];
            tensor<int32, [4]> var_36892_begin_0 = const()[name = tensor<string, []>("op_36892_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_36892_end_0 = const()[name = tensor<string, []>("op_36892_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_36892_end_mask_0 = const()[name = tensor<string, []>("op_36892_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36892_cast_fp16 = slice_by_index(begin = var_36892_begin_0, end = var_36892_end_0, end_mask = var_36892_end_mask_0, x = q_165_cast_fp16)[name = tensor<string, []>("op_36892_cast_fp16")];
            tensor<int32, [4]> var_36896_begin_0 = const()[name = tensor<string, []>("op_36896_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_36896_end_0 = const()[name = tensor<string, []>("op_36896_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_36896_end_mask_0 = const()[name = tensor<string, []>("op_36896_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36896_cast_fp16 = slice_by_index(begin = var_36896_begin_0, end = var_36896_end_0, end_mask = var_36896_end_mask_0, x = q_165_cast_fp16)[name = tensor<string, []>("op_36896_cast_fp16")];
            tensor<int32, [4]> var_36900_begin_0 = const()[name = tensor<string, []>("op_36900_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_36900_end_0 = const()[name = tensor<string, []>("op_36900_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_36900_end_mask_0 = const()[name = tensor<string, []>("op_36900_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36900_cast_fp16 = slice_by_index(begin = var_36900_begin_0, end = var_36900_end_0, end_mask = var_36900_end_mask_0, x = q_165_cast_fp16)[name = tensor<string, []>("op_36900_cast_fp16")];
            tensor<int32, [4]> var_36904_begin_0 = const()[name = tensor<string, []>("op_36904_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_36904_end_0 = const()[name = tensor<string, []>("op_36904_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_36904_end_mask_0 = const()[name = tensor<string, []>("op_36904_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36904_cast_fp16 = slice_by_index(begin = var_36904_begin_0, end = var_36904_end_0, end_mask = var_36904_end_mask_0, x = q_165_cast_fp16)[name = tensor<string, []>("op_36904_cast_fp16")];
            tensor<int32, [4]> var_36908_begin_0 = const()[name = tensor<string, []>("op_36908_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_36908_end_0 = const()[name = tensor<string, []>("op_36908_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_36908_end_mask_0 = const()[name = tensor<string, []>("op_36908_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36908_cast_fp16 = slice_by_index(begin = var_36908_begin_0, end = var_36908_end_0, end_mask = var_36908_end_mask_0, x = q_165_cast_fp16)[name = tensor<string, []>("op_36908_cast_fp16")];
            tensor<int32, [4]> var_36912_begin_0 = const()[name = tensor<string, []>("op_36912_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_36912_end_0 = const()[name = tensor<string, []>("op_36912_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_36912_end_mask_0 = const()[name = tensor<string, []>("op_36912_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36912_cast_fp16 = slice_by_index(begin = var_36912_begin_0, end = var_36912_end_0, end_mask = var_36912_end_mask_0, x = q_165_cast_fp16)[name = tensor<string, []>("op_36912_cast_fp16")];
            tensor<int32, [4]> var_36916_begin_0 = const()[name = tensor<string, []>("op_36916_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_36916_end_0 = const()[name = tensor<string, []>("op_36916_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_36916_end_mask_0 = const()[name = tensor<string, []>("op_36916_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36916_cast_fp16 = slice_by_index(begin = var_36916_begin_0, end = var_36916_end_0, end_mask = var_36916_end_mask_0, x = q_165_cast_fp16)[name = tensor<string, []>("op_36916_cast_fp16")];
            tensor<int32, [4]> var_36920_begin_0 = const()[name = tensor<string, []>("op_36920_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_36920_end_0 = const()[name = tensor<string, []>("op_36920_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_36920_end_mask_0 = const()[name = tensor<string, []>("op_36920_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36920_cast_fp16 = slice_by_index(begin = var_36920_begin_0, end = var_36920_end_0, end_mask = var_36920_end_mask_0, x = q_165_cast_fp16)[name = tensor<string, []>("op_36920_cast_fp16")];
            tensor<int32, [4]> var_36924_begin_0 = const()[name = tensor<string, []>("op_36924_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_36924_end_0 = const()[name = tensor<string, []>("op_36924_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_36924_end_mask_0 = const()[name = tensor<string, []>("op_36924_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36924_cast_fp16 = slice_by_index(begin = var_36924_begin_0, end = var_36924_end_0, end_mask = var_36924_end_mask_0, x = q_165_cast_fp16)[name = tensor<string, []>("op_36924_cast_fp16")];
            tensor<int32, [4]> var_36928_begin_0 = const()[name = tensor<string, []>("op_36928_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_36928_end_0 = const()[name = tensor<string, []>("op_36928_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_36928_end_mask_0 = const()[name = tensor<string, []>("op_36928_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36928_cast_fp16 = slice_by_index(begin = var_36928_begin_0, end = var_36928_end_0, end_mask = var_36928_end_mask_0, x = q_165_cast_fp16)[name = tensor<string, []>("op_36928_cast_fp16")];
            tensor<int32, [4]> var_36932_begin_0 = const()[name = tensor<string, []>("op_36932_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_36932_end_0 = const()[name = tensor<string, []>("op_36932_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_36932_end_mask_0 = const()[name = tensor<string, []>("op_36932_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36932_cast_fp16 = slice_by_index(begin = var_36932_begin_0, end = var_36932_end_0, end_mask = var_36932_end_mask_0, x = q_165_cast_fp16)[name = tensor<string, []>("op_36932_cast_fp16")];
            tensor<int32, [4]> var_36936_begin_0 = const()[name = tensor<string, []>("op_36936_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_36936_end_0 = const()[name = tensor<string, []>("op_36936_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_36936_end_mask_0 = const()[name = tensor<string, []>("op_36936_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36936_cast_fp16 = slice_by_index(begin = var_36936_begin_0, end = var_36936_end_0, end_mask = var_36936_end_mask_0, x = q_165_cast_fp16)[name = tensor<string, []>("op_36936_cast_fp16")];
            tensor<int32, [4]> var_36940_begin_0 = const()[name = tensor<string, []>("op_36940_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_36940_end_0 = const()[name = tensor<string, []>("op_36940_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_36940_end_mask_0 = const()[name = tensor<string, []>("op_36940_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36940_cast_fp16 = slice_by_index(begin = var_36940_begin_0, end = var_36940_end_0, end_mask = var_36940_end_mask_0, x = q_165_cast_fp16)[name = tensor<string, []>("op_36940_cast_fp16")];
            tensor<int32, [4]> var_36944_begin_0 = const()[name = tensor<string, []>("op_36944_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_36944_end_0 = const()[name = tensor<string, []>("op_36944_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_36944_end_mask_0 = const()[name = tensor<string, []>("op_36944_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36944_cast_fp16 = slice_by_index(begin = var_36944_begin_0, end = var_36944_end_0, end_mask = var_36944_end_mask_0, x = q_165_cast_fp16)[name = tensor<string, []>("op_36944_cast_fp16")];
            tensor<int32, [4]> var_36948_begin_0 = const()[name = tensor<string, []>("op_36948_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_36948_end_0 = const()[name = tensor<string, []>("op_36948_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_36948_end_mask_0 = const()[name = tensor<string, []>("op_36948_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36948_cast_fp16 = slice_by_index(begin = var_36948_begin_0, end = var_36948_end_0, end_mask = var_36948_end_mask_0, x = q_165_cast_fp16)[name = tensor<string, []>("op_36948_cast_fp16")];
            tensor<int32, [4]> var_36952_begin_0 = const()[name = tensor<string, []>("op_36952_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_36952_end_0 = const()[name = tensor<string, []>("op_36952_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_36952_end_mask_0 = const()[name = tensor<string, []>("op_36952_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36952_cast_fp16 = slice_by_index(begin = var_36952_begin_0, end = var_36952_end_0, end_mask = var_36952_end_mask_0, x = q_165_cast_fp16)[name = tensor<string, []>("op_36952_cast_fp16")];
            tensor<int32, [4]> var_36956_begin_0 = const()[name = tensor<string, []>("op_36956_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_36956_end_0 = const()[name = tensor<string, []>("op_36956_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_36956_end_mask_0 = const()[name = tensor<string, []>("op_36956_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36956_cast_fp16 = slice_by_index(begin = var_36956_begin_0, end = var_36956_end_0, end_mask = var_36956_end_mask_0, x = q_165_cast_fp16)[name = tensor<string, []>("op_36956_cast_fp16")];
            tensor<int32, [4]> var_36960_begin_0 = const()[name = tensor<string, []>("op_36960_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_36960_end_0 = const()[name = tensor<string, []>("op_36960_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_36960_end_mask_0 = const()[name = tensor<string, []>("op_36960_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36960_cast_fp16 = slice_by_index(begin = var_36960_begin_0, end = var_36960_end_0, end_mask = var_36960_end_mask_0, x = q_165_cast_fp16)[name = tensor<string, []>("op_36960_cast_fp16")];
            tensor<int32, [4]> k_331_perm_0 = const()[name = tensor<string, []>("k_331_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_36967_begin_0 = const()[name = tensor<string, []>("op_36967_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_36967_end_0 = const()[name = tensor<string, []>("op_36967_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_36967_end_mask_0 = const()[name = tensor<string, []>("op_36967_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> k_331_cast_fp16 = transpose(perm = k_331_perm_0, x = k_329_cast_fp16)[name = tensor<string, []>("transpose_57")];
            tensor<fp16, [2, 1024, 1, 64]> var_36967_cast_fp16 = slice_by_index(begin = var_36967_begin_0, end = var_36967_end_0, end_mask = var_36967_end_mask_0, x = k_331_cast_fp16)[name = tensor<string, []>("op_36967_cast_fp16")];
            tensor<int32, [4]> var_36971_begin_0 = const()[name = tensor<string, []>("op_36971_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_36971_end_0 = const()[name = tensor<string, []>("op_36971_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_36971_end_mask_0 = const()[name = tensor<string, []>("op_36971_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_36971_cast_fp16 = slice_by_index(begin = var_36971_begin_0, end = var_36971_end_0, end_mask = var_36971_end_mask_0, x = k_331_cast_fp16)[name = tensor<string, []>("op_36971_cast_fp16")];
            tensor<int32, [4]> var_36975_begin_0 = const()[name = tensor<string, []>("op_36975_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_36975_end_0 = const()[name = tensor<string, []>("op_36975_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_36975_end_mask_0 = const()[name = tensor<string, []>("op_36975_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_36975_cast_fp16 = slice_by_index(begin = var_36975_begin_0, end = var_36975_end_0, end_mask = var_36975_end_mask_0, x = k_331_cast_fp16)[name = tensor<string, []>("op_36975_cast_fp16")];
            tensor<int32, [4]> var_36979_begin_0 = const()[name = tensor<string, []>("op_36979_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_36979_end_0 = const()[name = tensor<string, []>("op_36979_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_36979_end_mask_0 = const()[name = tensor<string, []>("op_36979_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_36979_cast_fp16 = slice_by_index(begin = var_36979_begin_0, end = var_36979_end_0, end_mask = var_36979_end_mask_0, x = k_331_cast_fp16)[name = tensor<string, []>("op_36979_cast_fp16")];
            tensor<int32, [4]> var_36983_begin_0 = const()[name = tensor<string, []>("op_36983_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_36983_end_0 = const()[name = tensor<string, []>("op_36983_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_36983_end_mask_0 = const()[name = tensor<string, []>("op_36983_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_36983_cast_fp16 = slice_by_index(begin = var_36983_begin_0, end = var_36983_end_0, end_mask = var_36983_end_mask_0, x = k_331_cast_fp16)[name = tensor<string, []>("op_36983_cast_fp16")];
            tensor<int32, [4]> var_36987_begin_0 = const()[name = tensor<string, []>("op_36987_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_36987_end_0 = const()[name = tensor<string, []>("op_36987_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_36987_end_mask_0 = const()[name = tensor<string, []>("op_36987_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_36987_cast_fp16 = slice_by_index(begin = var_36987_begin_0, end = var_36987_end_0, end_mask = var_36987_end_mask_0, x = k_331_cast_fp16)[name = tensor<string, []>("op_36987_cast_fp16")];
            tensor<int32, [4]> var_36991_begin_0 = const()[name = tensor<string, []>("op_36991_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_36991_end_0 = const()[name = tensor<string, []>("op_36991_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_36991_end_mask_0 = const()[name = tensor<string, []>("op_36991_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_36991_cast_fp16 = slice_by_index(begin = var_36991_begin_0, end = var_36991_end_0, end_mask = var_36991_end_mask_0, x = k_331_cast_fp16)[name = tensor<string, []>("op_36991_cast_fp16")];
            tensor<int32, [4]> var_36995_begin_0 = const()[name = tensor<string, []>("op_36995_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_36995_end_0 = const()[name = tensor<string, []>("op_36995_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_36995_end_mask_0 = const()[name = tensor<string, []>("op_36995_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_36995_cast_fp16 = slice_by_index(begin = var_36995_begin_0, end = var_36995_end_0, end_mask = var_36995_end_mask_0, x = k_331_cast_fp16)[name = tensor<string, []>("op_36995_cast_fp16")];
            tensor<int32, [4]> var_36999_begin_0 = const()[name = tensor<string, []>("op_36999_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_36999_end_0 = const()[name = tensor<string, []>("op_36999_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_36999_end_mask_0 = const()[name = tensor<string, []>("op_36999_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_36999_cast_fp16 = slice_by_index(begin = var_36999_begin_0, end = var_36999_end_0, end_mask = var_36999_end_mask_0, x = k_331_cast_fp16)[name = tensor<string, []>("op_36999_cast_fp16")];
            tensor<int32, [4]> var_37003_begin_0 = const()[name = tensor<string, []>("op_37003_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_37003_end_0 = const()[name = tensor<string, []>("op_37003_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_37003_end_mask_0 = const()[name = tensor<string, []>("op_37003_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37003_cast_fp16 = slice_by_index(begin = var_37003_begin_0, end = var_37003_end_0, end_mask = var_37003_end_mask_0, x = k_331_cast_fp16)[name = tensor<string, []>("op_37003_cast_fp16")];
            tensor<int32, [4]> var_37007_begin_0 = const()[name = tensor<string, []>("op_37007_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_37007_end_0 = const()[name = tensor<string, []>("op_37007_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_37007_end_mask_0 = const()[name = tensor<string, []>("op_37007_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37007_cast_fp16 = slice_by_index(begin = var_37007_begin_0, end = var_37007_end_0, end_mask = var_37007_end_mask_0, x = k_331_cast_fp16)[name = tensor<string, []>("op_37007_cast_fp16")];
            tensor<int32, [4]> var_37011_begin_0 = const()[name = tensor<string, []>("op_37011_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_37011_end_0 = const()[name = tensor<string, []>("op_37011_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_37011_end_mask_0 = const()[name = tensor<string, []>("op_37011_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37011_cast_fp16 = slice_by_index(begin = var_37011_begin_0, end = var_37011_end_0, end_mask = var_37011_end_mask_0, x = k_331_cast_fp16)[name = tensor<string, []>("op_37011_cast_fp16")];
            tensor<int32, [4]> var_37015_begin_0 = const()[name = tensor<string, []>("op_37015_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_37015_end_0 = const()[name = tensor<string, []>("op_37015_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_37015_end_mask_0 = const()[name = tensor<string, []>("op_37015_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37015_cast_fp16 = slice_by_index(begin = var_37015_begin_0, end = var_37015_end_0, end_mask = var_37015_end_mask_0, x = k_331_cast_fp16)[name = tensor<string, []>("op_37015_cast_fp16")];
            tensor<int32, [4]> var_37019_begin_0 = const()[name = tensor<string, []>("op_37019_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_37019_end_0 = const()[name = tensor<string, []>("op_37019_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_37019_end_mask_0 = const()[name = tensor<string, []>("op_37019_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37019_cast_fp16 = slice_by_index(begin = var_37019_begin_0, end = var_37019_end_0, end_mask = var_37019_end_mask_0, x = k_331_cast_fp16)[name = tensor<string, []>("op_37019_cast_fp16")];
            tensor<int32, [4]> var_37023_begin_0 = const()[name = tensor<string, []>("op_37023_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_37023_end_0 = const()[name = tensor<string, []>("op_37023_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_37023_end_mask_0 = const()[name = tensor<string, []>("op_37023_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37023_cast_fp16 = slice_by_index(begin = var_37023_begin_0, end = var_37023_end_0, end_mask = var_37023_end_mask_0, x = k_331_cast_fp16)[name = tensor<string, []>("op_37023_cast_fp16")];
            tensor<int32, [4]> var_37027_begin_0 = const()[name = tensor<string, []>("op_37027_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_37027_end_0 = const()[name = tensor<string, []>("op_37027_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_37027_end_mask_0 = const()[name = tensor<string, []>("op_37027_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37027_cast_fp16 = slice_by_index(begin = var_37027_begin_0, end = var_37027_end_0, end_mask = var_37027_end_mask_0, x = k_331_cast_fp16)[name = tensor<string, []>("op_37027_cast_fp16")];
            tensor<int32, [4]> var_37031_begin_0 = const()[name = tensor<string, []>("op_37031_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_37031_end_0 = const()[name = tensor<string, []>("op_37031_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_37031_end_mask_0 = const()[name = tensor<string, []>("op_37031_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37031_cast_fp16 = slice_by_index(begin = var_37031_begin_0, end = var_37031_end_0, end_mask = var_37031_end_mask_0, x = k_331_cast_fp16)[name = tensor<string, []>("op_37031_cast_fp16")];
            tensor<int32, [4]> var_37035_begin_0 = const()[name = tensor<string, []>("op_37035_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_37035_end_0 = const()[name = tensor<string, []>("op_37035_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_37035_end_mask_0 = const()[name = tensor<string, []>("op_37035_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37035_cast_fp16 = slice_by_index(begin = var_37035_begin_0, end = var_37035_end_0, end_mask = var_37035_end_mask_0, x = k_331_cast_fp16)[name = tensor<string, []>("op_37035_cast_fp16")];
            tensor<int32, [4]> var_37039_begin_0 = const()[name = tensor<string, []>("op_37039_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_37039_end_0 = const()[name = tensor<string, []>("op_37039_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_37039_end_mask_0 = const()[name = tensor<string, []>("op_37039_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37039_cast_fp16 = slice_by_index(begin = var_37039_begin_0, end = var_37039_end_0, end_mask = var_37039_end_mask_0, x = k_331_cast_fp16)[name = tensor<string, []>("op_37039_cast_fp16")];
            tensor<int32, [4]> var_37043_begin_0 = const()[name = tensor<string, []>("op_37043_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_37043_end_0 = const()[name = tensor<string, []>("op_37043_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_37043_end_mask_0 = const()[name = tensor<string, []>("op_37043_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37043_cast_fp16 = slice_by_index(begin = var_37043_begin_0, end = var_37043_end_0, end_mask = var_37043_end_mask_0, x = k_331_cast_fp16)[name = tensor<string, []>("op_37043_cast_fp16")];
            tensor<int32, [4]> var_37045_begin_0 = const()[name = tensor<string, []>("op_37045_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_37045_end_0 = const()[name = tensor<string, []>("op_37045_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_37045_end_mask_0 = const()[name = tensor<string, []>("op_37045_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37045_cast_fp16 = slice_by_index(begin = var_37045_begin_0, end = var_37045_end_0, end_mask = var_37045_end_mask_0, x = v_165_cast_fp16)[name = tensor<string, []>("op_37045_cast_fp16")];
            tensor<int32, [4]> var_37049_begin_0 = const()[name = tensor<string, []>("op_37049_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_37049_end_0 = const()[name = tensor<string, []>("op_37049_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_37049_end_mask_0 = const()[name = tensor<string, []>("op_37049_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37049_cast_fp16 = slice_by_index(begin = var_37049_begin_0, end = var_37049_end_0, end_mask = var_37049_end_mask_0, x = v_165_cast_fp16)[name = tensor<string, []>("op_37049_cast_fp16")];
            tensor<int32, [4]> var_37053_begin_0 = const()[name = tensor<string, []>("op_37053_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_37053_end_0 = const()[name = tensor<string, []>("op_37053_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_37053_end_mask_0 = const()[name = tensor<string, []>("op_37053_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37053_cast_fp16 = slice_by_index(begin = var_37053_begin_0, end = var_37053_end_0, end_mask = var_37053_end_mask_0, x = v_165_cast_fp16)[name = tensor<string, []>("op_37053_cast_fp16")];
            tensor<int32, [4]> var_37057_begin_0 = const()[name = tensor<string, []>("op_37057_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_37057_end_0 = const()[name = tensor<string, []>("op_37057_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_37057_end_mask_0 = const()[name = tensor<string, []>("op_37057_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37057_cast_fp16 = slice_by_index(begin = var_37057_begin_0, end = var_37057_end_0, end_mask = var_37057_end_mask_0, x = v_165_cast_fp16)[name = tensor<string, []>("op_37057_cast_fp16")];
            tensor<int32, [4]> var_37061_begin_0 = const()[name = tensor<string, []>("op_37061_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_37061_end_0 = const()[name = tensor<string, []>("op_37061_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_37061_end_mask_0 = const()[name = tensor<string, []>("op_37061_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37061_cast_fp16 = slice_by_index(begin = var_37061_begin_0, end = var_37061_end_0, end_mask = var_37061_end_mask_0, x = v_165_cast_fp16)[name = tensor<string, []>("op_37061_cast_fp16")];
            tensor<int32, [4]> var_37065_begin_0 = const()[name = tensor<string, []>("op_37065_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_37065_end_0 = const()[name = tensor<string, []>("op_37065_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_37065_end_mask_0 = const()[name = tensor<string, []>("op_37065_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37065_cast_fp16 = slice_by_index(begin = var_37065_begin_0, end = var_37065_end_0, end_mask = var_37065_end_mask_0, x = v_165_cast_fp16)[name = tensor<string, []>("op_37065_cast_fp16")];
            tensor<int32, [4]> var_37069_begin_0 = const()[name = tensor<string, []>("op_37069_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_37069_end_0 = const()[name = tensor<string, []>("op_37069_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_37069_end_mask_0 = const()[name = tensor<string, []>("op_37069_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37069_cast_fp16 = slice_by_index(begin = var_37069_begin_0, end = var_37069_end_0, end_mask = var_37069_end_mask_0, x = v_165_cast_fp16)[name = tensor<string, []>("op_37069_cast_fp16")];
            tensor<int32, [4]> var_37073_begin_0 = const()[name = tensor<string, []>("op_37073_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_37073_end_0 = const()[name = tensor<string, []>("op_37073_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_37073_end_mask_0 = const()[name = tensor<string, []>("op_37073_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37073_cast_fp16 = slice_by_index(begin = var_37073_begin_0, end = var_37073_end_0, end_mask = var_37073_end_mask_0, x = v_165_cast_fp16)[name = tensor<string, []>("op_37073_cast_fp16")];
            tensor<int32, [4]> var_37077_begin_0 = const()[name = tensor<string, []>("op_37077_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_37077_end_0 = const()[name = tensor<string, []>("op_37077_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_37077_end_mask_0 = const()[name = tensor<string, []>("op_37077_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37077_cast_fp16 = slice_by_index(begin = var_37077_begin_0, end = var_37077_end_0, end_mask = var_37077_end_mask_0, x = v_165_cast_fp16)[name = tensor<string, []>("op_37077_cast_fp16")];
            tensor<int32, [4]> var_37081_begin_0 = const()[name = tensor<string, []>("op_37081_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_37081_end_0 = const()[name = tensor<string, []>("op_37081_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_37081_end_mask_0 = const()[name = tensor<string, []>("op_37081_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37081_cast_fp16 = slice_by_index(begin = var_37081_begin_0, end = var_37081_end_0, end_mask = var_37081_end_mask_0, x = v_165_cast_fp16)[name = tensor<string, []>("op_37081_cast_fp16")];
            tensor<int32, [4]> var_37085_begin_0 = const()[name = tensor<string, []>("op_37085_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_37085_end_0 = const()[name = tensor<string, []>("op_37085_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_37085_end_mask_0 = const()[name = tensor<string, []>("op_37085_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37085_cast_fp16 = slice_by_index(begin = var_37085_begin_0, end = var_37085_end_0, end_mask = var_37085_end_mask_0, x = v_165_cast_fp16)[name = tensor<string, []>("op_37085_cast_fp16")];
            tensor<int32, [4]> var_37089_begin_0 = const()[name = tensor<string, []>("op_37089_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_37089_end_0 = const()[name = tensor<string, []>("op_37089_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_37089_end_mask_0 = const()[name = tensor<string, []>("op_37089_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37089_cast_fp16 = slice_by_index(begin = var_37089_begin_0, end = var_37089_end_0, end_mask = var_37089_end_mask_0, x = v_165_cast_fp16)[name = tensor<string, []>("op_37089_cast_fp16")];
            tensor<int32, [4]> var_37093_begin_0 = const()[name = tensor<string, []>("op_37093_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_37093_end_0 = const()[name = tensor<string, []>("op_37093_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_37093_end_mask_0 = const()[name = tensor<string, []>("op_37093_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37093_cast_fp16 = slice_by_index(begin = var_37093_begin_0, end = var_37093_end_0, end_mask = var_37093_end_mask_0, x = v_165_cast_fp16)[name = tensor<string, []>("op_37093_cast_fp16")];
            tensor<int32, [4]> var_37097_begin_0 = const()[name = tensor<string, []>("op_37097_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_37097_end_0 = const()[name = tensor<string, []>("op_37097_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_37097_end_mask_0 = const()[name = tensor<string, []>("op_37097_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37097_cast_fp16 = slice_by_index(begin = var_37097_begin_0, end = var_37097_end_0, end_mask = var_37097_end_mask_0, x = v_165_cast_fp16)[name = tensor<string, []>("op_37097_cast_fp16")];
            tensor<int32, [4]> var_37101_begin_0 = const()[name = tensor<string, []>("op_37101_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_37101_end_0 = const()[name = tensor<string, []>("op_37101_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_37101_end_mask_0 = const()[name = tensor<string, []>("op_37101_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37101_cast_fp16 = slice_by_index(begin = var_37101_begin_0, end = var_37101_end_0, end_mask = var_37101_end_mask_0, x = v_165_cast_fp16)[name = tensor<string, []>("op_37101_cast_fp16")];
            tensor<int32, [4]> var_37105_begin_0 = const()[name = tensor<string, []>("op_37105_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_37105_end_0 = const()[name = tensor<string, []>("op_37105_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_37105_end_mask_0 = const()[name = tensor<string, []>("op_37105_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37105_cast_fp16 = slice_by_index(begin = var_37105_begin_0, end = var_37105_end_0, end_mask = var_37105_end_mask_0, x = v_165_cast_fp16)[name = tensor<string, []>("op_37105_cast_fp16")];
            tensor<int32, [4]> var_37109_begin_0 = const()[name = tensor<string, []>("op_37109_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_37109_end_0 = const()[name = tensor<string, []>("op_37109_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_37109_end_mask_0 = const()[name = tensor<string, []>("op_37109_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37109_cast_fp16 = slice_by_index(begin = var_37109_begin_0, end = var_37109_end_0, end_mask = var_37109_end_mask_0, x = v_165_cast_fp16)[name = tensor<string, []>("op_37109_cast_fp16")];
            tensor<int32, [4]> var_37113_begin_0 = const()[name = tensor<string, []>("op_37113_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_37113_end_0 = const()[name = tensor<string, []>("op_37113_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_37113_end_mask_0 = const()[name = tensor<string, []>("op_37113_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37113_cast_fp16 = slice_by_index(begin = var_37113_begin_0, end = var_37113_end_0, end_mask = var_37113_end_mask_0, x = v_165_cast_fp16)[name = tensor<string, []>("op_37113_cast_fp16")];
            tensor<int32, [4]> var_37117_begin_0 = const()[name = tensor<string, []>("op_37117_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_37117_end_0 = const()[name = tensor<string, []>("op_37117_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_37117_end_mask_0 = const()[name = tensor<string, []>("op_37117_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37117_cast_fp16 = slice_by_index(begin = var_37117_begin_0, end = var_37117_end_0, end_mask = var_37117_end_mask_0, x = v_165_cast_fp16)[name = tensor<string, []>("op_37117_cast_fp16")];
            tensor<int32, [4]> var_37121_begin_0 = const()[name = tensor<string, []>("op_37121_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_37121_end_0 = const()[name = tensor<string, []>("op_37121_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_37121_end_mask_0 = const()[name = tensor<string, []>("op_37121_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37121_cast_fp16 = slice_by_index(begin = var_37121_begin_0, end = var_37121_end_0, end_mask = var_37121_end_mask_0, x = v_165_cast_fp16)[name = tensor<string, []>("op_37121_cast_fp16")];
            tensor<string, []> var_37125_equation_0 = const()[name = tensor<string, []>("op_37125_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37125_cast_fp16 = einsum(equation = var_37125_equation_0, values = (var_36967_cast_fp16, var_36884_cast_fp16))[name = tensor<string, []>("op_37125_cast_fp16")];
            tensor<fp16, []> var_37126_to_fp16 = const()[name = tensor<string, []>("op_37126_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3121_cast_fp16 = mul(x = var_37125_cast_fp16, y = var_37126_to_fp16)[name = tensor<string, []>("aw_3121_cast_fp16")];
            tensor<string, []> var_37129_equation_0 = const()[name = tensor<string, []>("op_37129_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37129_cast_fp16 = einsum(equation = var_37129_equation_0, values = (var_36971_cast_fp16, var_36888_cast_fp16))[name = tensor<string, []>("op_37129_cast_fp16")];
            tensor<fp16, []> var_37130_to_fp16 = const()[name = tensor<string, []>("op_37130_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3123_cast_fp16 = mul(x = var_37129_cast_fp16, y = var_37130_to_fp16)[name = tensor<string, []>("aw_3123_cast_fp16")];
            tensor<string, []> var_37133_equation_0 = const()[name = tensor<string, []>("op_37133_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37133_cast_fp16 = einsum(equation = var_37133_equation_0, values = (var_36975_cast_fp16, var_36892_cast_fp16))[name = tensor<string, []>("op_37133_cast_fp16")];
            tensor<fp16, []> var_37134_to_fp16 = const()[name = tensor<string, []>("op_37134_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3125_cast_fp16 = mul(x = var_37133_cast_fp16, y = var_37134_to_fp16)[name = tensor<string, []>("aw_3125_cast_fp16")];
            tensor<string, []> var_37137_equation_0 = const()[name = tensor<string, []>("op_37137_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37137_cast_fp16 = einsum(equation = var_37137_equation_0, values = (var_36979_cast_fp16, var_36896_cast_fp16))[name = tensor<string, []>("op_37137_cast_fp16")];
            tensor<fp16, []> var_37138_to_fp16 = const()[name = tensor<string, []>("op_37138_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3127_cast_fp16 = mul(x = var_37137_cast_fp16, y = var_37138_to_fp16)[name = tensor<string, []>("aw_3127_cast_fp16")];
            tensor<string, []> var_37141_equation_0 = const()[name = tensor<string, []>("op_37141_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37141_cast_fp16 = einsum(equation = var_37141_equation_0, values = (var_36983_cast_fp16, var_36900_cast_fp16))[name = tensor<string, []>("op_37141_cast_fp16")];
            tensor<fp16, []> var_37142_to_fp16 = const()[name = tensor<string, []>("op_37142_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3129_cast_fp16 = mul(x = var_37141_cast_fp16, y = var_37142_to_fp16)[name = tensor<string, []>("aw_3129_cast_fp16")];
            tensor<string, []> var_37145_equation_0 = const()[name = tensor<string, []>("op_37145_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37145_cast_fp16 = einsum(equation = var_37145_equation_0, values = (var_36987_cast_fp16, var_36904_cast_fp16))[name = tensor<string, []>("op_37145_cast_fp16")];
            tensor<fp16, []> var_37146_to_fp16 = const()[name = tensor<string, []>("op_37146_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3131_cast_fp16 = mul(x = var_37145_cast_fp16, y = var_37146_to_fp16)[name = tensor<string, []>("aw_3131_cast_fp16")];
            tensor<string, []> var_37149_equation_0 = const()[name = tensor<string, []>("op_37149_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37149_cast_fp16 = einsum(equation = var_37149_equation_0, values = (var_36991_cast_fp16, var_36908_cast_fp16))[name = tensor<string, []>("op_37149_cast_fp16")];
            tensor<fp16, []> var_37150_to_fp16 = const()[name = tensor<string, []>("op_37150_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3133_cast_fp16 = mul(x = var_37149_cast_fp16, y = var_37150_to_fp16)[name = tensor<string, []>("aw_3133_cast_fp16")];
            tensor<string, []> var_37153_equation_0 = const()[name = tensor<string, []>("op_37153_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37153_cast_fp16 = einsum(equation = var_37153_equation_0, values = (var_36995_cast_fp16, var_36912_cast_fp16))[name = tensor<string, []>("op_37153_cast_fp16")];
            tensor<fp16, []> var_37154_to_fp16 = const()[name = tensor<string, []>("op_37154_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3135_cast_fp16 = mul(x = var_37153_cast_fp16, y = var_37154_to_fp16)[name = tensor<string, []>("aw_3135_cast_fp16")];
            tensor<string, []> var_37157_equation_0 = const()[name = tensor<string, []>("op_37157_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37157_cast_fp16 = einsum(equation = var_37157_equation_0, values = (var_36999_cast_fp16, var_36916_cast_fp16))[name = tensor<string, []>("op_37157_cast_fp16")];
            tensor<fp16, []> var_37158_to_fp16 = const()[name = tensor<string, []>("op_37158_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3137_cast_fp16 = mul(x = var_37157_cast_fp16, y = var_37158_to_fp16)[name = tensor<string, []>("aw_3137_cast_fp16")];
            tensor<string, []> var_37161_equation_0 = const()[name = tensor<string, []>("op_37161_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37161_cast_fp16 = einsum(equation = var_37161_equation_0, values = (var_37003_cast_fp16, var_36920_cast_fp16))[name = tensor<string, []>("op_37161_cast_fp16")];
            tensor<fp16, []> var_37162_to_fp16 = const()[name = tensor<string, []>("op_37162_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3139_cast_fp16 = mul(x = var_37161_cast_fp16, y = var_37162_to_fp16)[name = tensor<string, []>("aw_3139_cast_fp16")];
            tensor<string, []> var_37165_equation_0 = const()[name = tensor<string, []>("op_37165_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37165_cast_fp16 = einsum(equation = var_37165_equation_0, values = (var_37007_cast_fp16, var_36924_cast_fp16))[name = tensor<string, []>("op_37165_cast_fp16")];
            tensor<fp16, []> var_37166_to_fp16 = const()[name = tensor<string, []>("op_37166_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3141_cast_fp16 = mul(x = var_37165_cast_fp16, y = var_37166_to_fp16)[name = tensor<string, []>("aw_3141_cast_fp16")];
            tensor<string, []> var_37169_equation_0 = const()[name = tensor<string, []>("op_37169_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37169_cast_fp16 = einsum(equation = var_37169_equation_0, values = (var_37011_cast_fp16, var_36928_cast_fp16))[name = tensor<string, []>("op_37169_cast_fp16")];
            tensor<fp16, []> var_37170_to_fp16 = const()[name = tensor<string, []>("op_37170_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3143_cast_fp16 = mul(x = var_37169_cast_fp16, y = var_37170_to_fp16)[name = tensor<string, []>("aw_3143_cast_fp16")];
            tensor<string, []> var_37173_equation_0 = const()[name = tensor<string, []>("op_37173_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37173_cast_fp16 = einsum(equation = var_37173_equation_0, values = (var_37015_cast_fp16, var_36932_cast_fp16))[name = tensor<string, []>("op_37173_cast_fp16")];
            tensor<fp16, []> var_37174_to_fp16 = const()[name = tensor<string, []>("op_37174_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3145_cast_fp16 = mul(x = var_37173_cast_fp16, y = var_37174_to_fp16)[name = tensor<string, []>("aw_3145_cast_fp16")];
            tensor<string, []> var_37177_equation_0 = const()[name = tensor<string, []>("op_37177_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37177_cast_fp16 = einsum(equation = var_37177_equation_0, values = (var_37019_cast_fp16, var_36936_cast_fp16))[name = tensor<string, []>("op_37177_cast_fp16")];
            tensor<fp16, []> var_37178_to_fp16 = const()[name = tensor<string, []>("op_37178_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3147_cast_fp16 = mul(x = var_37177_cast_fp16, y = var_37178_to_fp16)[name = tensor<string, []>("aw_3147_cast_fp16")];
            tensor<string, []> var_37181_equation_0 = const()[name = tensor<string, []>("op_37181_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37181_cast_fp16 = einsum(equation = var_37181_equation_0, values = (var_37023_cast_fp16, var_36940_cast_fp16))[name = tensor<string, []>("op_37181_cast_fp16")];
            tensor<fp16, []> var_37182_to_fp16 = const()[name = tensor<string, []>("op_37182_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3149_cast_fp16 = mul(x = var_37181_cast_fp16, y = var_37182_to_fp16)[name = tensor<string, []>("aw_3149_cast_fp16")];
            tensor<string, []> var_37185_equation_0 = const()[name = tensor<string, []>("op_37185_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37185_cast_fp16 = einsum(equation = var_37185_equation_0, values = (var_37027_cast_fp16, var_36944_cast_fp16))[name = tensor<string, []>("op_37185_cast_fp16")];
            tensor<fp16, []> var_37186_to_fp16 = const()[name = tensor<string, []>("op_37186_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3151_cast_fp16 = mul(x = var_37185_cast_fp16, y = var_37186_to_fp16)[name = tensor<string, []>("aw_3151_cast_fp16")];
            tensor<string, []> var_37189_equation_0 = const()[name = tensor<string, []>("op_37189_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37189_cast_fp16 = einsum(equation = var_37189_equation_0, values = (var_37031_cast_fp16, var_36948_cast_fp16))[name = tensor<string, []>("op_37189_cast_fp16")];
            tensor<fp16, []> var_37190_to_fp16 = const()[name = tensor<string, []>("op_37190_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3153_cast_fp16 = mul(x = var_37189_cast_fp16, y = var_37190_to_fp16)[name = tensor<string, []>("aw_3153_cast_fp16")];
            tensor<string, []> var_37193_equation_0 = const()[name = tensor<string, []>("op_37193_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37193_cast_fp16 = einsum(equation = var_37193_equation_0, values = (var_37035_cast_fp16, var_36952_cast_fp16))[name = tensor<string, []>("op_37193_cast_fp16")];
            tensor<fp16, []> var_37194_to_fp16 = const()[name = tensor<string, []>("op_37194_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3155_cast_fp16 = mul(x = var_37193_cast_fp16, y = var_37194_to_fp16)[name = tensor<string, []>("aw_3155_cast_fp16")];
            tensor<string, []> var_37197_equation_0 = const()[name = tensor<string, []>("op_37197_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37197_cast_fp16 = einsum(equation = var_37197_equation_0, values = (var_37039_cast_fp16, var_36956_cast_fp16))[name = tensor<string, []>("op_37197_cast_fp16")];
            tensor<fp16, []> var_37198_to_fp16 = const()[name = tensor<string, []>("op_37198_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3157_cast_fp16 = mul(x = var_37197_cast_fp16, y = var_37198_to_fp16)[name = tensor<string, []>("aw_3157_cast_fp16")];
            tensor<string, []> var_37201_equation_0 = const()[name = tensor<string, []>("op_37201_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37201_cast_fp16 = einsum(equation = var_37201_equation_0, values = (var_37043_cast_fp16, var_36960_cast_fp16))[name = tensor<string, []>("op_37201_cast_fp16")];
            tensor<fp16, []> var_37202_to_fp16 = const()[name = tensor<string, []>("op_37202_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3159_cast_fp16 = mul(x = var_37201_cast_fp16, y = var_37202_to_fp16)[name = tensor<string, []>("aw_3159_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37204_cast_fp16 = softmax(axis = var_30355, x = aw_3121_cast_fp16)[name = tensor<string, []>("op_37204_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37205_cast_fp16 = softmax(axis = var_30355, x = aw_3123_cast_fp16)[name = tensor<string, []>("op_37205_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37206_cast_fp16 = softmax(axis = var_30355, x = aw_3125_cast_fp16)[name = tensor<string, []>("op_37206_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37207_cast_fp16 = softmax(axis = var_30355, x = aw_3127_cast_fp16)[name = tensor<string, []>("op_37207_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37208_cast_fp16 = softmax(axis = var_30355, x = aw_3129_cast_fp16)[name = tensor<string, []>("op_37208_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37209_cast_fp16 = softmax(axis = var_30355, x = aw_3131_cast_fp16)[name = tensor<string, []>("op_37209_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37210_cast_fp16 = softmax(axis = var_30355, x = aw_3133_cast_fp16)[name = tensor<string, []>("op_37210_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37211_cast_fp16 = softmax(axis = var_30355, x = aw_3135_cast_fp16)[name = tensor<string, []>("op_37211_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37212_cast_fp16 = softmax(axis = var_30355, x = aw_3137_cast_fp16)[name = tensor<string, []>("op_37212_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37213_cast_fp16 = softmax(axis = var_30355, x = aw_3139_cast_fp16)[name = tensor<string, []>("op_37213_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37214_cast_fp16 = softmax(axis = var_30355, x = aw_3141_cast_fp16)[name = tensor<string, []>("op_37214_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37215_cast_fp16 = softmax(axis = var_30355, x = aw_3143_cast_fp16)[name = tensor<string, []>("op_37215_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37216_cast_fp16 = softmax(axis = var_30355, x = aw_3145_cast_fp16)[name = tensor<string, []>("op_37216_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37217_cast_fp16 = softmax(axis = var_30355, x = aw_3147_cast_fp16)[name = tensor<string, []>("op_37217_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37218_cast_fp16 = softmax(axis = var_30355, x = aw_3149_cast_fp16)[name = tensor<string, []>("op_37218_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37219_cast_fp16 = softmax(axis = var_30355, x = aw_3151_cast_fp16)[name = tensor<string, []>("op_37219_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37220_cast_fp16 = softmax(axis = var_30355, x = aw_3153_cast_fp16)[name = tensor<string, []>("op_37220_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37221_cast_fp16 = softmax(axis = var_30355, x = aw_3155_cast_fp16)[name = tensor<string, []>("op_37221_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37222_cast_fp16 = softmax(axis = var_30355, x = aw_3157_cast_fp16)[name = tensor<string, []>("op_37222_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37223_cast_fp16 = softmax(axis = var_30355, x = aw_3159_cast_fp16)[name = tensor<string, []>("op_37223_cast_fp16")];
            tensor<string, []> var_37225_equation_0 = const()[name = tensor<string, []>("op_37225_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37225_cast_fp16 = einsum(equation = var_37225_equation_0, values = (var_37045_cast_fp16, var_37204_cast_fp16))[name = tensor<string, []>("op_37225_cast_fp16")];
            tensor<string, []> var_37227_equation_0 = const()[name = tensor<string, []>("op_37227_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37227_cast_fp16 = einsum(equation = var_37227_equation_0, values = (var_37049_cast_fp16, var_37205_cast_fp16))[name = tensor<string, []>("op_37227_cast_fp16")];
            tensor<string, []> var_37229_equation_0 = const()[name = tensor<string, []>("op_37229_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37229_cast_fp16 = einsum(equation = var_37229_equation_0, values = (var_37053_cast_fp16, var_37206_cast_fp16))[name = tensor<string, []>("op_37229_cast_fp16")];
            tensor<string, []> var_37231_equation_0 = const()[name = tensor<string, []>("op_37231_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37231_cast_fp16 = einsum(equation = var_37231_equation_0, values = (var_37057_cast_fp16, var_37207_cast_fp16))[name = tensor<string, []>("op_37231_cast_fp16")];
            tensor<string, []> var_37233_equation_0 = const()[name = tensor<string, []>("op_37233_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37233_cast_fp16 = einsum(equation = var_37233_equation_0, values = (var_37061_cast_fp16, var_37208_cast_fp16))[name = tensor<string, []>("op_37233_cast_fp16")];
            tensor<string, []> var_37235_equation_0 = const()[name = tensor<string, []>("op_37235_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37235_cast_fp16 = einsum(equation = var_37235_equation_0, values = (var_37065_cast_fp16, var_37209_cast_fp16))[name = tensor<string, []>("op_37235_cast_fp16")];
            tensor<string, []> var_37237_equation_0 = const()[name = tensor<string, []>("op_37237_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37237_cast_fp16 = einsum(equation = var_37237_equation_0, values = (var_37069_cast_fp16, var_37210_cast_fp16))[name = tensor<string, []>("op_37237_cast_fp16")];
            tensor<string, []> var_37239_equation_0 = const()[name = tensor<string, []>("op_37239_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37239_cast_fp16 = einsum(equation = var_37239_equation_0, values = (var_37073_cast_fp16, var_37211_cast_fp16))[name = tensor<string, []>("op_37239_cast_fp16")];
            tensor<string, []> var_37241_equation_0 = const()[name = tensor<string, []>("op_37241_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37241_cast_fp16 = einsum(equation = var_37241_equation_0, values = (var_37077_cast_fp16, var_37212_cast_fp16))[name = tensor<string, []>("op_37241_cast_fp16")];
            tensor<string, []> var_37243_equation_0 = const()[name = tensor<string, []>("op_37243_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37243_cast_fp16 = einsum(equation = var_37243_equation_0, values = (var_37081_cast_fp16, var_37213_cast_fp16))[name = tensor<string, []>("op_37243_cast_fp16")];
            tensor<string, []> var_37245_equation_0 = const()[name = tensor<string, []>("op_37245_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37245_cast_fp16 = einsum(equation = var_37245_equation_0, values = (var_37085_cast_fp16, var_37214_cast_fp16))[name = tensor<string, []>("op_37245_cast_fp16")];
            tensor<string, []> var_37247_equation_0 = const()[name = tensor<string, []>("op_37247_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37247_cast_fp16 = einsum(equation = var_37247_equation_0, values = (var_37089_cast_fp16, var_37215_cast_fp16))[name = tensor<string, []>("op_37247_cast_fp16")];
            tensor<string, []> var_37249_equation_0 = const()[name = tensor<string, []>("op_37249_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37249_cast_fp16 = einsum(equation = var_37249_equation_0, values = (var_37093_cast_fp16, var_37216_cast_fp16))[name = tensor<string, []>("op_37249_cast_fp16")];
            tensor<string, []> var_37251_equation_0 = const()[name = tensor<string, []>("op_37251_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37251_cast_fp16 = einsum(equation = var_37251_equation_0, values = (var_37097_cast_fp16, var_37217_cast_fp16))[name = tensor<string, []>("op_37251_cast_fp16")];
            tensor<string, []> var_37253_equation_0 = const()[name = tensor<string, []>("op_37253_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37253_cast_fp16 = einsum(equation = var_37253_equation_0, values = (var_37101_cast_fp16, var_37218_cast_fp16))[name = tensor<string, []>("op_37253_cast_fp16")];
            tensor<string, []> var_37255_equation_0 = const()[name = tensor<string, []>("op_37255_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37255_cast_fp16 = einsum(equation = var_37255_equation_0, values = (var_37105_cast_fp16, var_37219_cast_fp16))[name = tensor<string, []>("op_37255_cast_fp16")];
            tensor<string, []> var_37257_equation_0 = const()[name = tensor<string, []>("op_37257_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37257_cast_fp16 = einsum(equation = var_37257_equation_0, values = (var_37109_cast_fp16, var_37220_cast_fp16))[name = tensor<string, []>("op_37257_cast_fp16")];
            tensor<string, []> var_37259_equation_0 = const()[name = tensor<string, []>("op_37259_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37259_cast_fp16 = einsum(equation = var_37259_equation_0, values = (var_37113_cast_fp16, var_37221_cast_fp16))[name = tensor<string, []>("op_37259_cast_fp16")];
            tensor<string, []> var_37261_equation_0 = const()[name = tensor<string, []>("op_37261_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37261_cast_fp16 = einsum(equation = var_37261_equation_0, values = (var_37117_cast_fp16, var_37222_cast_fp16))[name = tensor<string, []>("op_37261_cast_fp16")];
            tensor<string, []> var_37263_equation_0 = const()[name = tensor<string, []>("op_37263_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37263_cast_fp16 = einsum(equation = var_37263_equation_0, values = (var_37121_cast_fp16, var_37223_cast_fp16))[name = tensor<string, []>("op_37263_cast_fp16")];
            tensor<bool, []> input_495_interleave_0 = const()[name = tensor<string, []>("input_495_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_495_cast_fp16 = concat(axis = var_30355, interleave = input_495_interleave_0, values = (var_37225_cast_fp16, var_37227_cast_fp16, var_37229_cast_fp16, var_37231_cast_fp16, var_37233_cast_fp16, var_37235_cast_fp16, var_37237_cast_fp16, var_37239_cast_fp16, var_37241_cast_fp16, var_37243_cast_fp16, var_37245_cast_fp16, var_37247_cast_fp16, var_37249_cast_fp16, var_37251_cast_fp16, var_37253_cast_fp16, var_37255_cast_fp16, var_37257_cast_fp16, var_37259_cast_fp16, var_37261_cast_fp16, var_37263_cast_fp16))[name = tensor<string, []>("input_495_cast_fp16")];
            tensor<string, []> var_37273_pad_type_0 = const()[name = tensor<string, []>("op_37273_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_37273_strides_0 = const()[name = tensor<string, []>("op_37273_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_37273_pad_0 = const()[name = tensor<string, []>("op_37273_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_37273_dilations_0 = const()[name = tensor<string, []>("op_37273_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_37273_groups_0 = const()[name = tensor<string, []>("op_37273_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_7_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(190090560))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(191319424))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_7_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_7_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_7_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(191319616)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_37273_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_7_attn1_to_out_0_bias_to_fp16, dilations = var_37273_dilations_0, groups = var_37273_groups_0, pad = var_37273_pad_0, pad_type = var_37273_pad_type_0, strides = var_37273_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_7_attn1_to_out_0_weight_to_fp16_palettized, x = input_495_cast_fp16)[name = tensor<string, []>("op_37273_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_249_cast_fp16 = add(x = var_37273_cast_fp16, y = inputs_247_cast_fp16)[name = tensor<string, []>("inputs_249_cast_fp16")];
            tensor<int32, [1]> hidden_states_331_axes_0 = const()[name = tensor<string, []>("hidden_states_331_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_331_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_331_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(191322240)))];
            tensor<fp16, [1280]> hidden_states_331_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_331_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(191324864)))];
            tensor<fp16, []> var_37283_to_fp16 = const()[name = tensor<string, []>("op_37283_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_331_cast_fp16 = layer_norm(axes = hidden_states_331_axes_0, beta = hidden_states_331_beta_0_to_fp16, epsilon = var_37283_to_fp16, gamma = hidden_states_331_gamma_0_to_fp16, x = inputs_249_cast_fp16)[name = tensor<string, []>("hidden_states_331_cast_fp16")];
            tensor<string, []> q_167_pad_type_0 = const()[name = tensor<string, []>("q_167_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_167_strides_0 = const()[name = tensor<string, []>("q_167_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_167_pad_0 = const()[name = tensor<string, []>("q_167_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_167_dilations_0 = const()[name = tensor<string, []>("q_167_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_167_groups_0 = const()[name = tensor<string, []>("q_167_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_7_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(191327488))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(192556352))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_7_attn2_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_167_cast_fp16 = conv(dilations = q_167_dilations_0, groups = q_167_groups_0, pad = q_167_pad_0, pad_type = q_167_pad_type_0, strides = q_167_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_7_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_331_cast_fp16)[name = tensor<string, []>("q_167_cast_fp16")];
            tensor<string, []> k_333_pad_type_0 = const()[name = tensor<string, []>("k_333_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_333_strides_0 = const()[name = tensor<string, []>("k_333_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_333_pad_0 = const()[name = tensor<string, []>("k_333_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_333_dilations_0 = const()[name = tensor<string, []>("k_333_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_333_groups_0 = const()[name = tensor<string, []>("k_333_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_7_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(192556544))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(194522688))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_7_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> k_333_cast_fp16 = conv(dilations = k_333_dilations_0, groups = k_333_groups_0, pad = k_333_pad_0, pad_type = k_333_pad_type_0, strides = k_333_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_7_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("k_333_cast_fp16")];
            tensor<string, []> v_167_pad_type_0 = const()[name = tensor<string, []>("v_167_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_167_strides_0 = const()[name = tensor<string, []>("v_167_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_167_pad_0 = const()[name = tensor<string, []>("v_167_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_167_dilations_0 = const()[name = tensor<string, []>("v_167_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_167_groups_0 = const()[name = tensor<string, []>("v_167_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_7_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(194522880))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(196489024))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_7_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> v_167_cast_fp16 = conv(dilations = v_167_dilations_0, groups = v_167_groups_0, pad = v_167_pad_0, pad_type = v_167_pad_type_0, strides = v_167_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_7_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("v_167_cast_fp16")];
            tensor<int32, [4]> var_37316_begin_0 = const()[name = tensor<string, []>("op_37316_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_37316_end_0 = const()[name = tensor<string, []>("op_37316_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_37316_end_mask_0 = const()[name = tensor<string, []>("op_37316_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37316_cast_fp16 = slice_by_index(begin = var_37316_begin_0, end = var_37316_end_0, end_mask = var_37316_end_mask_0, x = q_167_cast_fp16)[name = tensor<string, []>("op_37316_cast_fp16")];
            tensor<int32, [4]> var_37320_begin_0 = const()[name = tensor<string, []>("op_37320_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_37320_end_0 = const()[name = tensor<string, []>("op_37320_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_37320_end_mask_0 = const()[name = tensor<string, []>("op_37320_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37320_cast_fp16 = slice_by_index(begin = var_37320_begin_0, end = var_37320_end_0, end_mask = var_37320_end_mask_0, x = q_167_cast_fp16)[name = tensor<string, []>("op_37320_cast_fp16")];
            tensor<int32, [4]> var_37324_begin_0 = const()[name = tensor<string, []>("op_37324_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_37324_end_0 = const()[name = tensor<string, []>("op_37324_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_37324_end_mask_0 = const()[name = tensor<string, []>("op_37324_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37324_cast_fp16 = slice_by_index(begin = var_37324_begin_0, end = var_37324_end_0, end_mask = var_37324_end_mask_0, x = q_167_cast_fp16)[name = tensor<string, []>("op_37324_cast_fp16")];
            tensor<int32, [4]> var_37328_begin_0 = const()[name = tensor<string, []>("op_37328_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_37328_end_0 = const()[name = tensor<string, []>("op_37328_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_37328_end_mask_0 = const()[name = tensor<string, []>("op_37328_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37328_cast_fp16 = slice_by_index(begin = var_37328_begin_0, end = var_37328_end_0, end_mask = var_37328_end_mask_0, x = q_167_cast_fp16)[name = tensor<string, []>("op_37328_cast_fp16")];
            tensor<int32, [4]> var_37332_begin_0 = const()[name = tensor<string, []>("op_37332_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_37332_end_0 = const()[name = tensor<string, []>("op_37332_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_37332_end_mask_0 = const()[name = tensor<string, []>("op_37332_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37332_cast_fp16 = slice_by_index(begin = var_37332_begin_0, end = var_37332_end_0, end_mask = var_37332_end_mask_0, x = q_167_cast_fp16)[name = tensor<string, []>("op_37332_cast_fp16")];
            tensor<int32, [4]> var_37336_begin_0 = const()[name = tensor<string, []>("op_37336_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_37336_end_0 = const()[name = tensor<string, []>("op_37336_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_37336_end_mask_0 = const()[name = tensor<string, []>("op_37336_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37336_cast_fp16 = slice_by_index(begin = var_37336_begin_0, end = var_37336_end_0, end_mask = var_37336_end_mask_0, x = q_167_cast_fp16)[name = tensor<string, []>("op_37336_cast_fp16")];
            tensor<int32, [4]> var_37340_begin_0 = const()[name = tensor<string, []>("op_37340_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_37340_end_0 = const()[name = tensor<string, []>("op_37340_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_37340_end_mask_0 = const()[name = tensor<string, []>("op_37340_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37340_cast_fp16 = slice_by_index(begin = var_37340_begin_0, end = var_37340_end_0, end_mask = var_37340_end_mask_0, x = q_167_cast_fp16)[name = tensor<string, []>("op_37340_cast_fp16")];
            tensor<int32, [4]> var_37344_begin_0 = const()[name = tensor<string, []>("op_37344_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_37344_end_0 = const()[name = tensor<string, []>("op_37344_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_37344_end_mask_0 = const()[name = tensor<string, []>("op_37344_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37344_cast_fp16 = slice_by_index(begin = var_37344_begin_0, end = var_37344_end_0, end_mask = var_37344_end_mask_0, x = q_167_cast_fp16)[name = tensor<string, []>("op_37344_cast_fp16")];
            tensor<int32, [4]> var_37348_begin_0 = const()[name = tensor<string, []>("op_37348_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_37348_end_0 = const()[name = tensor<string, []>("op_37348_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_37348_end_mask_0 = const()[name = tensor<string, []>("op_37348_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37348_cast_fp16 = slice_by_index(begin = var_37348_begin_0, end = var_37348_end_0, end_mask = var_37348_end_mask_0, x = q_167_cast_fp16)[name = tensor<string, []>("op_37348_cast_fp16")];
            tensor<int32, [4]> var_37352_begin_0 = const()[name = tensor<string, []>("op_37352_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_37352_end_0 = const()[name = tensor<string, []>("op_37352_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_37352_end_mask_0 = const()[name = tensor<string, []>("op_37352_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37352_cast_fp16 = slice_by_index(begin = var_37352_begin_0, end = var_37352_end_0, end_mask = var_37352_end_mask_0, x = q_167_cast_fp16)[name = tensor<string, []>("op_37352_cast_fp16")];
            tensor<int32, [4]> var_37356_begin_0 = const()[name = tensor<string, []>("op_37356_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_37356_end_0 = const()[name = tensor<string, []>("op_37356_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_37356_end_mask_0 = const()[name = tensor<string, []>("op_37356_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37356_cast_fp16 = slice_by_index(begin = var_37356_begin_0, end = var_37356_end_0, end_mask = var_37356_end_mask_0, x = q_167_cast_fp16)[name = tensor<string, []>("op_37356_cast_fp16")];
            tensor<int32, [4]> var_37360_begin_0 = const()[name = tensor<string, []>("op_37360_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_37360_end_0 = const()[name = tensor<string, []>("op_37360_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_37360_end_mask_0 = const()[name = tensor<string, []>("op_37360_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37360_cast_fp16 = slice_by_index(begin = var_37360_begin_0, end = var_37360_end_0, end_mask = var_37360_end_mask_0, x = q_167_cast_fp16)[name = tensor<string, []>("op_37360_cast_fp16")];
            tensor<int32, [4]> var_37364_begin_0 = const()[name = tensor<string, []>("op_37364_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_37364_end_0 = const()[name = tensor<string, []>("op_37364_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_37364_end_mask_0 = const()[name = tensor<string, []>("op_37364_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37364_cast_fp16 = slice_by_index(begin = var_37364_begin_0, end = var_37364_end_0, end_mask = var_37364_end_mask_0, x = q_167_cast_fp16)[name = tensor<string, []>("op_37364_cast_fp16")];
            tensor<int32, [4]> var_37368_begin_0 = const()[name = tensor<string, []>("op_37368_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_37368_end_0 = const()[name = tensor<string, []>("op_37368_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_37368_end_mask_0 = const()[name = tensor<string, []>("op_37368_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37368_cast_fp16 = slice_by_index(begin = var_37368_begin_0, end = var_37368_end_0, end_mask = var_37368_end_mask_0, x = q_167_cast_fp16)[name = tensor<string, []>("op_37368_cast_fp16")];
            tensor<int32, [4]> var_37372_begin_0 = const()[name = tensor<string, []>("op_37372_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_37372_end_0 = const()[name = tensor<string, []>("op_37372_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_37372_end_mask_0 = const()[name = tensor<string, []>("op_37372_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37372_cast_fp16 = slice_by_index(begin = var_37372_begin_0, end = var_37372_end_0, end_mask = var_37372_end_mask_0, x = q_167_cast_fp16)[name = tensor<string, []>("op_37372_cast_fp16")];
            tensor<int32, [4]> var_37376_begin_0 = const()[name = tensor<string, []>("op_37376_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_37376_end_0 = const()[name = tensor<string, []>("op_37376_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_37376_end_mask_0 = const()[name = tensor<string, []>("op_37376_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37376_cast_fp16 = slice_by_index(begin = var_37376_begin_0, end = var_37376_end_0, end_mask = var_37376_end_mask_0, x = q_167_cast_fp16)[name = tensor<string, []>("op_37376_cast_fp16")];
            tensor<int32, [4]> var_37380_begin_0 = const()[name = tensor<string, []>("op_37380_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_37380_end_0 = const()[name = tensor<string, []>("op_37380_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_37380_end_mask_0 = const()[name = tensor<string, []>("op_37380_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37380_cast_fp16 = slice_by_index(begin = var_37380_begin_0, end = var_37380_end_0, end_mask = var_37380_end_mask_0, x = q_167_cast_fp16)[name = tensor<string, []>("op_37380_cast_fp16")];
            tensor<int32, [4]> var_37384_begin_0 = const()[name = tensor<string, []>("op_37384_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_37384_end_0 = const()[name = tensor<string, []>("op_37384_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_37384_end_mask_0 = const()[name = tensor<string, []>("op_37384_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37384_cast_fp16 = slice_by_index(begin = var_37384_begin_0, end = var_37384_end_0, end_mask = var_37384_end_mask_0, x = q_167_cast_fp16)[name = tensor<string, []>("op_37384_cast_fp16")];
            tensor<int32, [4]> var_37388_begin_0 = const()[name = tensor<string, []>("op_37388_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_37388_end_0 = const()[name = tensor<string, []>("op_37388_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_37388_end_mask_0 = const()[name = tensor<string, []>("op_37388_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37388_cast_fp16 = slice_by_index(begin = var_37388_begin_0, end = var_37388_end_0, end_mask = var_37388_end_mask_0, x = q_167_cast_fp16)[name = tensor<string, []>("op_37388_cast_fp16")];
            tensor<int32, [4]> var_37392_begin_0 = const()[name = tensor<string, []>("op_37392_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_37392_end_0 = const()[name = tensor<string, []>("op_37392_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_37392_end_mask_0 = const()[name = tensor<string, []>("op_37392_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37392_cast_fp16 = slice_by_index(begin = var_37392_begin_0, end = var_37392_end_0, end_mask = var_37392_end_mask_0, x = q_167_cast_fp16)[name = tensor<string, []>("op_37392_cast_fp16")];
            tensor<int32, [4]> k_335_perm_0 = const()[name = tensor<string, []>("k_335_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_37399_begin_0 = const()[name = tensor<string, []>("op_37399_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_37399_end_0 = const()[name = tensor<string, []>("op_37399_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_37399_end_mask_0 = const()[name = tensor<string, []>("op_37399_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> k_335_cast_fp16 = transpose(perm = k_335_perm_0, x = k_333_cast_fp16)[name = tensor<string, []>("transpose_56")];
            tensor<fp16, [2, 77, 1, 64]> var_37399_cast_fp16 = slice_by_index(begin = var_37399_begin_0, end = var_37399_end_0, end_mask = var_37399_end_mask_0, x = k_335_cast_fp16)[name = tensor<string, []>("op_37399_cast_fp16")];
            tensor<int32, [4]> var_37403_begin_0 = const()[name = tensor<string, []>("op_37403_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_37403_end_0 = const()[name = tensor<string, []>("op_37403_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_37403_end_mask_0 = const()[name = tensor<string, []>("op_37403_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_37403_cast_fp16 = slice_by_index(begin = var_37403_begin_0, end = var_37403_end_0, end_mask = var_37403_end_mask_0, x = k_335_cast_fp16)[name = tensor<string, []>("op_37403_cast_fp16")];
            tensor<int32, [4]> var_37407_begin_0 = const()[name = tensor<string, []>("op_37407_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_37407_end_0 = const()[name = tensor<string, []>("op_37407_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_37407_end_mask_0 = const()[name = tensor<string, []>("op_37407_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_37407_cast_fp16 = slice_by_index(begin = var_37407_begin_0, end = var_37407_end_0, end_mask = var_37407_end_mask_0, x = k_335_cast_fp16)[name = tensor<string, []>("op_37407_cast_fp16")];
            tensor<int32, [4]> var_37411_begin_0 = const()[name = tensor<string, []>("op_37411_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_37411_end_0 = const()[name = tensor<string, []>("op_37411_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_37411_end_mask_0 = const()[name = tensor<string, []>("op_37411_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_37411_cast_fp16 = slice_by_index(begin = var_37411_begin_0, end = var_37411_end_0, end_mask = var_37411_end_mask_0, x = k_335_cast_fp16)[name = tensor<string, []>("op_37411_cast_fp16")];
            tensor<int32, [4]> var_37415_begin_0 = const()[name = tensor<string, []>("op_37415_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_37415_end_0 = const()[name = tensor<string, []>("op_37415_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_37415_end_mask_0 = const()[name = tensor<string, []>("op_37415_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_37415_cast_fp16 = slice_by_index(begin = var_37415_begin_0, end = var_37415_end_0, end_mask = var_37415_end_mask_0, x = k_335_cast_fp16)[name = tensor<string, []>("op_37415_cast_fp16")];
            tensor<int32, [4]> var_37419_begin_0 = const()[name = tensor<string, []>("op_37419_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_37419_end_0 = const()[name = tensor<string, []>("op_37419_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_37419_end_mask_0 = const()[name = tensor<string, []>("op_37419_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_37419_cast_fp16 = slice_by_index(begin = var_37419_begin_0, end = var_37419_end_0, end_mask = var_37419_end_mask_0, x = k_335_cast_fp16)[name = tensor<string, []>("op_37419_cast_fp16")];
            tensor<int32, [4]> var_37423_begin_0 = const()[name = tensor<string, []>("op_37423_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_37423_end_0 = const()[name = tensor<string, []>("op_37423_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_37423_end_mask_0 = const()[name = tensor<string, []>("op_37423_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_37423_cast_fp16 = slice_by_index(begin = var_37423_begin_0, end = var_37423_end_0, end_mask = var_37423_end_mask_0, x = k_335_cast_fp16)[name = tensor<string, []>("op_37423_cast_fp16")];
            tensor<int32, [4]> var_37427_begin_0 = const()[name = tensor<string, []>("op_37427_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_37427_end_0 = const()[name = tensor<string, []>("op_37427_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_37427_end_mask_0 = const()[name = tensor<string, []>("op_37427_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_37427_cast_fp16 = slice_by_index(begin = var_37427_begin_0, end = var_37427_end_0, end_mask = var_37427_end_mask_0, x = k_335_cast_fp16)[name = tensor<string, []>("op_37427_cast_fp16")];
            tensor<int32, [4]> var_37431_begin_0 = const()[name = tensor<string, []>("op_37431_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_37431_end_0 = const()[name = tensor<string, []>("op_37431_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_37431_end_mask_0 = const()[name = tensor<string, []>("op_37431_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_37431_cast_fp16 = slice_by_index(begin = var_37431_begin_0, end = var_37431_end_0, end_mask = var_37431_end_mask_0, x = k_335_cast_fp16)[name = tensor<string, []>("op_37431_cast_fp16")];
            tensor<int32, [4]> var_37435_begin_0 = const()[name = tensor<string, []>("op_37435_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_37435_end_0 = const()[name = tensor<string, []>("op_37435_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_37435_end_mask_0 = const()[name = tensor<string, []>("op_37435_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_37435_cast_fp16 = slice_by_index(begin = var_37435_begin_0, end = var_37435_end_0, end_mask = var_37435_end_mask_0, x = k_335_cast_fp16)[name = tensor<string, []>("op_37435_cast_fp16")];
            tensor<int32, [4]> var_37439_begin_0 = const()[name = tensor<string, []>("op_37439_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_37439_end_0 = const()[name = tensor<string, []>("op_37439_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_37439_end_mask_0 = const()[name = tensor<string, []>("op_37439_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_37439_cast_fp16 = slice_by_index(begin = var_37439_begin_0, end = var_37439_end_0, end_mask = var_37439_end_mask_0, x = k_335_cast_fp16)[name = tensor<string, []>("op_37439_cast_fp16")];
            tensor<int32, [4]> var_37443_begin_0 = const()[name = tensor<string, []>("op_37443_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_37443_end_0 = const()[name = tensor<string, []>("op_37443_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_37443_end_mask_0 = const()[name = tensor<string, []>("op_37443_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_37443_cast_fp16 = slice_by_index(begin = var_37443_begin_0, end = var_37443_end_0, end_mask = var_37443_end_mask_0, x = k_335_cast_fp16)[name = tensor<string, []>("op_37443_cast_fp16")];
            tensor<int32, [4]> var_37447_begin_0 = const()[name = tensor<string, []>("op_37447_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_37447_end_0 = const()[name = tensor<string, []>("op_37447_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_37447_end_mask_0 = const()[name = tensor<string, []>("op_37447_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_37447_cast_fp16 = slice_by_index(begin = var_37447_begin_0, end = var_37447_end_0, end_mask = var_37447_end_mask_0, x = k_335_cast_fp16)[name = tensor<string, []>("op_37447_cast_fp16")];
            tensor<int32, [4]> var_37451_begin_0 = const()[name = tensor<string, []>("op_37451_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_37451_end_0 = const()[name = tensor<string, []>("op_37451_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_37451_end_mask_0 = const()[name = tensor<string, []>("op_37451_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_37451_cast_fp16 = slice_by_index(begin = var_37451_begin_0, end = var_37451_end_0, end_mask = var_37451_end_mask_0, x = k_335_cast_fp16)[name = tensor<string, []>("op_37451_cast_fp16")];
            tensor<int32, [4]> var_37455_begin_0 = const()[name = tensor<string, []>("op_37455_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_37455_end_0 = const()[name = tensor<string, []>("op_37455_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_37455_end_mask_0 = const()[name = tensor<string, []>("op_37455_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_37455_cast_fp16 = slice_by_index(begin = var_37455_begin_0, end = var_37455_end_0, end_mask = var_37455_end_mask_0, x = k_335_cast_fp16)[name = tensor<string, []>("op_37455_cast_fp16")];
            tensor<int32, [4]> var_37459_begin_0 = const()[name = tensor<string, []>("op_37459_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_37459_end_0 = const()[name = tensor<string, []>("op_37459_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_37459_end_mask_0 = const()[name = tensor<string, []>("op_37459_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_37459_cast_fp16 = slice_by_index(begin = var_37459_begin_0, end = var_37459_end_0, end_mask = var_37459_end_mask_0, x = k_335_cast_fp16)[name = tensor<string, []>("op_37459_cast_fp16")];
            tensor<int32, [4]> var_37463_begin_0 = const()[name = tensor<string, []>("op_37463_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_37463_end_0 = const()[name = tensor<string, []>("op_37463_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_37463_end_mask_0 = const()[name = tensor<string, []>("op_37463_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_37463_cast_fp16 = slice_by_index(begin = var_37463_begin_0, end = var_37463_end_0, end_mask = var_37463_end_mask_0, x = k_335_cast_fp16)[name = tensor<string, []>("op_37463_cast_fp16")];
            tensor<int32, [4]> var_37467_begin_0 = const()[name = tensor<string, []>("op_37467_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_37467_end_0 = const()[name = tensor<string, []>("op_37467_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_37467_end_mask_0 = const()[name = tensor<string, []>("op_37467_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_37467_cast_fp16 = slice_by_index(begin = var_37467_begin_0, end = var_37467_end_0, end_mask = var_37467_end_mask_0, x = k_335_cast_fp16)[name = tensor<string, []>("op_37467_cast_fp16")];
            tensor<int32, [4]> var_37471_begin_0 = const()[name = tensor<string, []>("op_37471_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_37471_end_0 = const()[name = tensor<string, []>("op_37471_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_37471_end_mask_0 = const()[name = tensor<string, []>("op_37471_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_37471_cast_fp16 = slice_by_index(begin = var_37471_begin_0, end = var_37471_end_0, end_mask = var_37471_end_mask_0, x = k_335_cast_fp16)[name = tensor<string, []>("op_37471_cast_fp16")];
            tensor<int32, [4]> var_37475_begin_0 = const()[name = tensor<string, []>("op_37475_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_37475_end_0 = const()[name = tensor<string, []>("op_37475_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_37475_end_mask_0 = const()[name = tensor<string, []>("op_37475_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_37475_cast_fp16 = slice_by_index(begin = var_37475_begin_0, end = var_37475_end_0, end_mask = var_37475_end_mask_0, x = k_335_cast_fp16)[name = tensor<string, []>("op_37475_cast_fp16")];
            tensor<int32, [4]> var_37477_begin_0 = const()[name = tensor<string, []>("op_37477_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_37477_end_0 = const()[name = tensor<string, []>("op_37477_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_37477_end_mask_0 = const()[name = tensor<string, []>("op_37477_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_37477_cast_fp16 = slice_by_index(begin = var_37477_begin_0, end = var_37477_end_0, end_mask = var_37477_end_mask_0, x = v_167_cast_fp16)[name = tensor<string, []>("op_37477_cast_fp16")];
            tensor<int32, [4]> var_37481_begin_0 = const()[name = tensor<string, []>("op_37481_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_37481_end_0 = const()[name = tensor<string, []>("op_37481_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_37481_end_mask_0 = const()[name = tensor<string, []>("op_37481_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_37481_cast_fp16 = slice_by_index(begin = var_37481_begin_0, end = var_37481_end_0, end_mask = var_37481_end_mask_0, x = v_167_cast_fp16)[name = tensor<string, []>("op_37481_cast_fp16")];
            tensor<int32, [4]> var_37485_begin_0 = const()[name = tensor<string, []>("op_37485_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_37485_end_0 = const()[name = tensor<string, []>("op_37485_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_37485_end_mask_0 = const()[name = tensor<string, []>("op_37485_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_37485_cast_fp16 = slice_by_index(begin = var_37485_begin_0, end = var_37485_end_0, end_mask = var_37485_end_mask_0, x = v_167_cast_fp16)[name = tensor<string, []>("op_37485_cast_fp16")];
            tensor<int32, [4]> var_37489_begin_0 = const()[name = tensor<string, []>("op_37489_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_37489_end_0 = const()[name = tensor<string, []>("op_37489_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_37489_end_mask_0 = const()[name = tensor<string, []>("op_37489_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_37489_cast_fp16 = slice_by_index(begin = var_37489_begin_0, end = var_37489_end_0, end_mask = var_37489_end_mask_0, x = v_167_cast_fp16)[name = tensor<string, []>("op_37489_cast_fp16")];
            tensor<int32, [4]> var_37493_begin_0 = const()[name = tensor<string, []>("op_37493_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_37493_end_0 = const()[name = tensor<string, []>("op_37493_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_37493_end_mask_0 = const()[name = tensor<string, []>("op_37493_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_37493_cast_fp16 = slice_by_index(begin = var_37493_begin_0, end = var_37493_end_0, end_mask = var_37493_end_mask_0, x = v_167_cast_fp16)[name = tensor<string, []>("op_37493_cast_fp16")];
            tensor<int32, [4]> var_37497_begin_0 = const()[name = tensor<string, []>("op_37497_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_37497_end_0 = const()[name = tensor<string, []>("op_37497_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_37497_end_mask_0 = const()[name = tensor<string, []>("op_37497_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_37497_cast_fp16 = slice_by_index(begin = var_37497_begin_0, end = var_37497_end_0, end_mask = var_37497_end_mask_0, x = v_167_cast_fp16)[name = tensor<string, []>("op_37497_cast_fp16")];
            tensor<int32, [4]> var_37501_begin_0 = const()[name = tensor<string, []>("op_37501_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_37501_end_0 = const()[name = tensor<string, []>("op_37501_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_37501_end_mask_0 = const()[name = tensor<string, []>("op_37501_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_37501_cast_fp16 = slice_by_index(begin = var_37501_begin_0, end = var_37501_end_0, end_mask = var_37501_end_mask_0, x = v_167_cast_fp16)[name = tensor<string, []>("op_37501_cast_fp16")];
            tensor<int32, [4]> var_37505_begin_0 = const()[name = tensor<string, []>("op_37505_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_37505_end_0 = const()[name = tensor<string, []>("op_37505_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_37505_end_mask_0 = const()[name = tensor<string, []>("op_37505_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_37505_cast_fp16 = slice_by_index(begin = var_37505_begin_0, end = var_37505_end_0, end_mask = var_37505_end_mask_0, x = v_167_cast_fp16)[name = tensor<string, []>("op_37505_cast_fp16")];
            tensor<int32, [4]> var_37509_begin_0 = const()[name = tensor<string, []>("op_37509_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_37509_end_0 = const()[name = tensor<string, []>("op_37509_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_37509_end_mask_0 = const()[name = tensor<string, []>("op_37509_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_37509_cast_fp16 = slice_by_index(begin = var_37509_begin_0, end = var_37509_end_0, end_mask = var_37509_end_mask_0, x = v_167_cast_fp16)[name = tensor<string, []>("op_37509_cast_fp16")];
            tensor<int32, [4]> var_37513_begin_0 = const()[name = tensor<string, []>("op_37513_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_37513_end_0 = const()[name = tensor<string, []>("op_37513_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_37513_end_mask_0 = const()[name = tensor<string, []>("op_37513_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_37513_cast_fp16 = slice_by_index(begin = var_37513_begin_0, end = var_37513_end_0, end_mask = var_37513_end_mask_0, x = v_167_cast_fp16)[name = tensor<string, []>("op_37513_cast_fp16")];
            tensor<int32, [4]> var_37517_begin_0 = const()[name = tensor<string, []>("op_37517_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_37517_end_0 = const()[name = tensor<string, []>("op_37517_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_37517_end_mask_0 = const()[name = tensor<string, []>("op_37517_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_37517_cast_fp16 = slice_by_index(begin = var_37517_begin_0, end = var_37517_end_0, end_mask = var_37517_end_mask_0, x = v_167_cast_fp16)[name = tensor<string, []>("op_37517_cast_fp16")];
            tensor<int32, [4]> var_37521_begin_0 = const()[name = tensor<string, []>("op_37521_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_37521_end_0 = const()[name = tensor<string, []>("op_37521_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_37521_end_mask_0 = const()[name = tensor<string, []>("op_37521_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_37521_cast_fp16 = slice_by_index(begin = var_37521_begin_0, end = var_37521_end_0, end_mask = var_37521_end_mask_0, x = v_167_cast_fp16)[name = tensor<string, []>("op_37521_cast_fp16")];
            tensor<int32, [4]> var_37525_begin_0 = const()[name = tensor<string, []>("op_37525_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_37525_end_0 = const()[name = tensor<string, []>("op_37525_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_37525_end_mask_0 = const()[name = tensor<string, []>("op_37525_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_37525_cast_fp16 = slice_by_index(begin = var_37525_begin_0, end = var_37525_end_0, end_mask = var_37525_end_mask_0, x = v_167_cast_fp16)[name = tensor<string, []>("op_37525_cast_fp16")];
            tensor<int32, [4]> var_37529_begin_0 = const()[name = tensor<string, []>("op_37529_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_37529_end_0 = const()[name = tensor<string, []>("op_37529_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_37529_end_mask_0 = const()[name = tensor<string, []>("op_37529_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_37529_cast_fp16 = slice_by_index(begin = var_37529_begin_0, end = var_37529_end_0, end_mask = var_37529_end_mask_0, x = v_167_cast_fp16)[name = tensor<string, []>("op_37529_cast_fp16")];
            tensor<int32, [4]> var_37533_begin_0 = const()[name = tensor<string, []>("op_37533_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_37533_end_0 = const()[name = tensor<string, []>("op_37533_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_37533_end_mask_0 = const()[name = tensor<string, []>("op_37533_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_37533_cast_fp16 = slice_by_index(begin = var_37533_begin_0, end = var_37533_end_0, end_mask = var_37533_end_mask_0, x = v_167_cast_fp16)[name = tensor<string, []>("op_37533_cast_fp16")];
            tensor<int32, [4]> var_37537_begin_0 = const()[name = tensor<string, []>("op_37537_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_37537_end_0 = const()[name = tensor<string, []>("op_37537_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_37537_end_mask_0 = const()[name = tensor<string, []>("op_37537_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_37537_cast_fp16 = slice_by_index(begin = var_37537_begin_0, end = var_37537_end_0, end_mask = var_37537_end_mask_0, x = v_167_cast_fp16)[name = tensor<string, []>("op_37537_cast_fp16")];
            tensor<int32, [4]> var_37541_begin_0 = const()[name = tensor<string, []>("op_37541_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_37541_end_0 = const()[name = tensor<string, []>("op_37541_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_37541_end_mask_0 = const()[name = tensor<string, []>("op_37541_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_37541_cast_fp16 = slice_by_index(begin = var_37541_begin_0, end = var_37541_end_0, end_mask = var_37541_end_mask_0, x = v_167_cast_fp16)[name = tensor<string, []>("op_37541_cast_fp16")];
            tensor<int32, [4]> var_37545_begin_0 = const()[name = tensor<string, []>("op_37545_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_37545_end_0 = const()[name = tensor<string, []>("op_37545_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_37545_end_mask_0 = const()[name = tensor<string, []>("op_37545_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_37545_cast_fp16 = slice_by_index(begin = var_37545_begin_0, end = var_37545_end_0, end_mask = var_37545_end_mask_0, x = v_167_cast_fp16)[name = tensor<string, []>("op_37545_cast_fp16")];
            tensor<int32, [4]> var_37549_begin_0 = const()[name = tensor<string, []>("op_37549_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_37549_end_0 = const()[name = tensor<string, []>("op_37549_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_37549_end_mask_0 = const()[name = tensor<string, []>("op_37549_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_37549_cast_fp16 = slice_by_index(begin = var_37549_begin_0, end = var_37549_end_0, end_mask = var_37549_end_mask_0, x = v_167_cast_fp16)[name = tensor<string, []>("op_37549_cast_fp16")];
            tensor<int32, [4]> var_37553_begin_0 = const()[name = tensor<string, []>("op_37553_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_37553_end_0 = const()[name = tensor<string, []>("op_37553_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_37553_end_mask_0 = const()[name = tensor<string, []>("op_37553_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_37553_cast_fp16 = slice_by_index(begin = var_37553_begin_0, end = var_37553_end_0, end_mask = var_37553_end_mask_0, x = v_167_cast_fp16)[name = tensor<string, []>("op_37553_cast_fp16")];
            tensor<string, []> var_37557_equation_0 = const()[name = tensor<string, []>("op_37557_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_37557_cast_fp16 = einsum(equation = var_37557_equation_0, values = (var_37399_cast_fp16, var_37316_cast_fp16))[name = tensor<string, []>("op_37557_cast_fp16")];
            tensor<fp16, []> var_37558_to_fp16 = const()[name = tensor<string, []>("op_37558_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3161_cast_fp16 = mul(x = var_37557_cast_fp16, y = var_37558_to_fp16)[name = tensor<string, []>("aw_3161_cast_fp16")];
            tensor<string, []> var_37561_equation_0 = const()[name = tensor<string, []>("op_37561_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_37561_cast_fp16 = einsum(equation = var_37561_equation_0, values = (var_37403_cast_fp16, var_37320_cast_fp16))[name = tensor<string, []>("op_37561_cast_fp16")];
            tensor<fp16, []> var_37562_to_fp16 = const()[name = tensor<string, []>("op_37562_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3163_cast_fp16 = mul(x = var_37561_cast_fp16, y = var_37562_to_fp16)[name = tensor<string, []>("aw_3163_cast_fp16")];
            tensor<string, []> var_37565_equation_0 = const()[name = tensor<string, []>("op_37565_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_37565_cast_fp16 = einsum(equation = var_37565_equation_0, values = (var_37407_cast_fp16, var_37324_cast_fp16))[name = tensor<string, []>("op_37565_cast_fp16")];
            tensor<fp16, []> var_37566_to_fp16 = const()[name = tensor<string, []>("op_37566_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3165_cast_fp16 = mul(x = var_37565_cast_fp16, y = var_37566_to_fp16)[name = tensor<string, []>("aw_3165_cast_fp16")];
            tensor<string, []> var_37569_equation_0 = const()[name = tensor<string, []>("op_37569_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_37569_cast_fp16 = einsum(equation = var_37569_equation_0, values = (var_37411_cast_fp16, var_37328_cast_fp16))[name = tensor<string, []>("op_37569_cast_fp16")];
            tensor<fp16, []> var_37570_to_fp16 = const()[name = tensor<string, []>("op_37570_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3167_cast_fp16 = mul(x = var_37569_cast_fp16, y = var_37570_to_fp16)[name = tensor<string, []>("aw_3167_cast_fp16")];
            tensor<string, []> var_37573_equation_0 = const()[name = tensor<string, []>("op_37573_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_37573_cast_fp16 = einsum(equation = var_37573_equation_0, values = (var_37415_cast_fp16, var_37332_cast_fp16))[name = tensor<string, []>("op_37573_cast_fp16")];
            tensor<fp16, []> var_37574_to_fp16 = const()[name = tensor<string, []>("op_37574_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3169_cast_fp16 = mul(x = var_37573_cast_fp16, y = var_37574_to_fp16)[name = tensor<string, []>("aw_3169_cast_fp16")];
            tensor<string, []> var_37577_equation_0 = const()[name = tensor<string, []>("op_37577_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_37577_cast_fp16 = einsum(equation = var_37577_equation_0, values = (var_37419_cast_fp16, var_37336_cast_fp16))[name = tensor<string, []>("op_37577_cast_fp16")];
            tensor<fp16, []> var_37578_to_fp16 = const()[name = tensor<string, []>("op_37578_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3171_cast_fp16 = mul(x = var_37577_cast_fp16, y = var_37578_to_fp16)[name = tensor<string, []>("aw_3171_cast_fp16")];
            tensor<string, []> var_37581_equation_0 = const()[name = tensor<string, []>("op_37581_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_37581_cast_fp16 = einsum(equation = var_37581_equation_0, values = (var_37423_cast_fp16, var_37340_cast_fp16))[name = tensor<string, []>("op_37581_cast_fp16")];
            tensor<fp16, []> var_37582_to_fp16 = const()[name = tensor<string, []>("op_37582_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3173_cast_fp16 = mul(x = var_37581_cast_fp16, y = var_37582_to_fp16)[name = tensor<string, []>("aw_3173_cast_fp16")];
            tensor<string, []> var_37585_equation_0 = const()[name = tensor<string, []>("op_37585_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_37585_cast_fp16 = einsum(equation = var_37585_equation_0, values = (var_37427_cast_fp16, var_37344_cast_fp16))[name = tensor<string, []>("op_37585_cast_fp16")];
            tensor<fp16, []> var_37586_to_fp16 = const()[name = tensor<string, []>("op_37586_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3175_cast_fp16 = mul(x = var_37585_cast_fp16, y = var_37586_to_fp16)[name = tensor<string, []>("aw_3175_cast_fp16")];
            tensor<string, []> var_37589_equation_0 = const()[name = tensor<string, []>("op_37589_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_37589_cast_fp16 = einsum(equation = var_37589_equation_0, values = (var_37431_cast_fp16, var_37348_cast_fp16))[name = tensor<string, []>("op_37589_cast_fp16")];
            tensor<fp16, []> var_37590_to_fp16 = const()[name = tensor<string, []>("op_37590_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3177_cast_fp16 = mul(x = var_37589_cast_fp16, y = var_37590_to_fp16)[name = tensor<string, []>("aw_3177_cast_fp16")];
            tensor<string, []> var_37593_equation_0 = const()[name = tensor<string, []>("op_37593_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_37593_cast_fp16 = einsum(equation = var_37593_equation_0, values = (var_37435_cast_fp16, var_37352_cast_fp16))[name = tensor<string, []>("op_37593_cast_fp16")];
            tensor<fp16, []> var_37594_to_fp16 = const()[name = tensor<string, []>("op_37594_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3179_cast_fp16 = mul(x = var_37593_cast_fp16, y = var_37594_to_fp16)[name = tensor<string, []>("aw_3179_cast_fp16")];
            tensor<string, []> var_37597_equation_0 = const()[name = tensor<string, []>("op_37597_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_37597_cast_fp16 = einsum(equation = var_37597_equation_0, values = (var_37439_cast_fp16, var_37356_cast_fp16))[name = tensor<string, []>("op_37597_cast_fp16")];
            tensor<fp16, []> var_37598_to_fp16 = const()[name = tensor<string, []>("op_37598_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3181_cast_fp16 = mul(x = var_37597_cast_fp16, y = var_37598_to_fp16)[name = tensor<string, []>("aw_3181_cast_fp16")];
            tensor<string, []> var_37601_equation_0 = const()[name = tensor<string, []>("op_37601_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_37601_cast_fp16 = einsum(equation = var_37601_equation_0, values = (var_37443_cast_fp16, var_37360_cast_fp16))[name = tensor<string, []>("op_37601_cast_fp16")];
            tensor<fp16, []> var_37602_to_fp16 = const()[name = tensor<string, []>("op_37602_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3183_cast_fp16 = mul(x = var_37601_cast_fp16, y = var_37602_to_fp16)[name = tensor<string, []>("aw_3183_cast_fp16")];
            tensor<string, []> var_37605_equation_0 = const()[name = tensor<string, []>("op_37605_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_37605_cast_fp16 = einsum(equation = var_37605_equation_0, values = (var_37447_cast_fp16, var_37364_cast_fp16))[name = tensor<string, []>("op_37605_cast_fp16")];
            tensor<fp16, []> var_37606_to_fp16 = const()[name = tensor<string, []>("op_37606_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3185_cast_fp16 = mul(x = var_37605_cast_fp16, y = var_37606_to_fp16)[name = tensor<string, []>("aw_3185_cast_fp16")];
            tensor<string, []> var_37609_equation_0 = const()[name = tensor<string, []>("op_37609_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_37609_cast_fp16 = einsum(equation = var_37609_equation_0, values = (var_37451_cast_fp16, var_37368_cast_fp16))[name = tensor<string, []>("op_37609_cast_fp16")];
            tensor<fp16, []> var_37610_to_fp16 = const()[name = tensor<string, []>("op_37610_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3187_cast_fp16 = mul(x = var_37609_cast_fp16, y = var_37610_to_fp16)[name = tensor<string, []>("aw_3187_cast_fp16")];
            tensor<string, []> var_37613_equation_0 = const()[name = tensor<string, []>("op_37613_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_37613_cast_fp16 = einsum(equation = var_37613_equation_0, values = (var_37455_cast_fp16, var_37372_cast_fp16))[name = tensor<string, []>("op_37613_cast_fp16")];
            tensor<fp16, []> var_37614_to_fp16 = const()[name = tensor<string, []>("op_37614_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3189_cast_fp16 = mul(x = var_37613_cast_fp16, y = var_37614_to_fp16)[name = tensor<string, []>("aw_3189_cast_fp16")];
            tensor<string, []> var_37617_equation_0 = const()[name = tensor<string, []>("op_37617_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_37617_cast_fp16 = einsum(equation = var_37617_equation_0, values = (var_37459_cast_fp16, var_37376_cast_fp16))[name = tensor<string, []>("op_37617_cast_fp16")];
            tensor<fp16, []> var_37618_to_fp16 = const()[name = tensor<string, []>("op_37618_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3191_cast_fp16 = mul(x = var_37617_cast_fp16, y = var_37618_to_fp16)[name = tensor<string, []>("aw_3191_cast_fp16")];
            tensor<string, []> var_37621_equation_0 = const()[name = tensor<string, []>("op_37621_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_37621_cast_fp16 = einsum(equation = var_37621_equation_0, values = (var_37463_cast_fp16, var_37380_cast_fp16))[name = tensor<string, []>("op_37621_cast_fp16")];
            tensor<fp16, []> var_37622_to_fp16 = const()[name = tensor<string, []>("op_37622_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3193_cast_fp16 = mul(x = var_37621_cast_fp16, y = var_37622_to_fp16)[name = tensor<string, []>("aw_3193_cast_fp16")];
            tensor<string, []> var_37625_equation_0 = const()[name = tensor<string, []>("op_37625_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_37625_cast_fp16 = einsum(equation = var_37625_equation_0, values = (var_37467_cast_fp16, var_37384_cast_fp16))[name = tensor<string, []>("op_37625_cast_fp16")];
            tensor<fp16, []> var_37626_to_fp16 = const()[name = tensor<string, []>("op_37626_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3195_cast_fp16 = mul(x = var_37625_cast_fp16, y = var_37626_to_fp16)[name = tensor<string, []>("aw_3195_cast_fp16")];
            tensor<string, []> var_37629_equation_0 = const()[name = tensor<string, []>("op_37629_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_37629_cast_fp16 = einsum(equation = var_37629_equation_0, values = (var_37471_cast_fp16, var_37388_cast_fp16))[name = tensor<string, []>("op_37629_cast_fp16")];
            tensor<fp16, []> var_37630_to_fp16 = const()[name = tensor<string, []>("op_37630_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3197_cast_fp16 = mul(x = var_37629_cast_fp16, y = var_37630_to_fp16)[name = tensor<string, []>("aw_3197_cast_fp16")];
            tensor<string, []> var_37633_equation_0 = const()[name = tensor<string, []>("op_37633_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_37633_cast_fp16 = einsum(equation = var_37633_equation_0, values = (var_37475_cast_fp16, var_37392_cast_fp16))[name = tensor<string, []>("op_37633_cast_fp16")];
            tensor<fp16, []> var_37634_to_fp16 = const()[name = tensor<string, []>("op_37634_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3199_cast_fp16 = mul(x = var_37633_cast_fp16, y = var_37634_to_fp16)[name = tensor<string, []>("aw_3199_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_37636_cast_fp16 = softmax(axis = var_30355, x = aw_3161_cast_fp16)[name = tensor<string, []>("op_37636_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_37637_cast_fp16 = softmax(axis = var_30355, x = aw_3163_cast_fp16)[name = tensor<string, []>("op_37637_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_37638_cast_fp16 = softmax(axis = var_30355, x = aw_3165_cast_fp16)[name = tensor<string, []>("op_37638_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_37639_cast_fp16 = softmax(axis = var_30355, x = aw_3167_cast_fp16)[name = tensor<string, []>("op_37639_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_37640_cast_fp16 = softmax(axis = var_30355, x = aw_3169_cast_fp16)[name = tensor<string, []>("op_37640_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_37641_cast_fp16 = softmax(axis = var_30355, x = aw_3171_cast_fp16)[name = tensor<string, []>("op_37641_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_37642_cast_fp16 = softmax(axis = var_30355, x = aw_3173_cast_fp16)[name = tensor<string, []>("op_37642_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_37643_cast_fp16 = softmax(axis = var_30355, x = aw_3175_cast_fp16)[name = tensor<string, []>("op_37643_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_37644_cast_fp16 = softmax(axis = var_30355, x = aw_3177_cast_fp16)[name = tensor<string, []>("op_37644_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_37645_cast_fp16 = softmax(axis = var_30355, x = aw_3179_cast_fp16)[name = tensor<string, []>("op_37645_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_37646_cast_fp16 = softmax(axis = var_30355, x = aw_3181_cast_fp16)[name = tensor<string, []>("op_37646_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_37647_cast_fp16 = softmax(axis = var_30355, x = aw_3183_cast_fp16)[name = tensor<string, []>("op_37647_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_37648_cast_fp16 = softmax(axis = var_30355, x = aw_3185_cast_fp16)[name = tensor<string, []>("op_37648_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_37649_cast_fp16 = softmax(axis = var_30355, x = aw_3187_cast_fp16)[name = tensor<string, []>("op_37649_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_37650_cast_fp16 = softmax(axis = var_30355, x = aw_3189_cast_fp16)[name = tensor<string, []>("op_37650_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_37651_cast_fp16 = softmax(axis = var_30355, x = aw_3191_cast_fp16)[name = tensor<string, []>("op_37651_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_37652_cast_fp16 = softmax(axis = var_30355, x = aw_3193_cast_fp16)[name = tensor<string, []>("op_37652_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_37653_cast_fp16 = softmax(axis = var_30355, x = aw_3195_cast_fp16)[name = tensor<string, []>("op_37653_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_37654_cast_fp16 = softmax(axis = var_30355, x = aw_3197_cast_fp16)[name = tensor<string, []>("op_37654_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_37655_cast_fp16 = softmax(axis = var_30355, x = aw_3199_cast_fp16)[name = tensor<string, []>("op_37655_cast_fp16")];
            tensor<string, []> var_37657_equation_0 = const()[name = tensor<string, []>("op_37657_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37657_cast_fp16 = einsum(equation = var_37657_equation_0, values = (var_37477_cast_fp16, var_37636_cast_fp16))[name = tensor<string, []>("op_37657_cast_fp16")];
            tensor<string, []> var_37659_equation_0 = const()[name = tensor<string, []>("op_37659_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37659_cast_fp16 = einsum(equation = var_37659_equation_0, values = (var_37481_cast_fp16, var_37637_cast_fp16))[name = tensor<string, []>("op_37659_cast_fp16")];
            tensor<string, []> var_37661_equation_0 = const()[name = tensor<string, []>("op_37661_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37661_cast_fp16 = einsum(equation = var_37661_equation_0, values = (var_37485_cast_fp16, var_37638_cast_fp16))[name = tensor<string, []>("op_37661_cast_fp16")];
            tensor<string, []> var_37663_equation_0 = const()[name = tensor<string, []>("op_37663_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37663_cast_fp16 = einsum(equation = var_37663_equation_0, values = (var_37489_cast_fp16, var_37639_cast_fp16))[name = tensor<string, []>("op_37663_cast_fp16")];
            tensor<string, []> var_37665_equation_0 = const()[name = tensor<string, []>("op_37665_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37665_cast_fp16 = einsum(equation = var_37665_equation_0, values = (var_37493_cast_fp16, var_37640_cast_fp16))[name = tensor<string, []>("op_37665_cast_fp16")];
            tensor<string, []> var_37667_equation_0 = const()[name = tensor<string, []>("op_37667_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37667_cast_fp16 = einsum(equation = var_37667_equation_0, values = (var_37497_cast_fp16, var_37641_cast_fp16))[name = tensor<string, []>("op_37667_cast_fp16")];
            tensor<string, []> var_37669_equation_0 = const()[name = tensor<string, []>("op_37669_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37669_cast_fp16 = einsum(equation = var_37669_equation_0, values = (var_37501_cast_fp16, var_37642_cast_fp16))[name = tensor<string, []>("op_37669_cast_fp16")];
            tensor<string, []> var_37671_equation_0 = const()[name = tensor<string, []>("op_37671_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37671_cast_fp16 = einsum(equation = var_37671_equation_0, values = (var_37505_cast_fp16, var_37643_cast_fp16))[name = tensor<string, []>("op_37671_cast_fp16")];
            tensor<string, []> var_37673_equation_0 = const()[name = tensor<string, []>("op_37673_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37673_cast_fp16 = einsum(equation = var_37673_equation_0, values = (var_37509_cast_fp16, var_37644_cast_fp16))[name = tensor<string, []>("op_37673_cast_fp16")];
            tensor<string, []> var_37675_equation_0 = const()[name = tensor<string, []>("op_37675_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37675_cast_fp16 = einsum(equation = var_37675_equation_0, values = (var_37513_cast_fp16, var_37645_cast_fp16))[name = tensor<string, []>("op_37675_cast_fp16")];
            tensor<string, []> var_37677_equation_0 = const()[name = tensor<string, []>("op_37677_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37677_cast_fp16 = einsum(equation = var_37677_equation_0, values = (var_37517_cast_fp16, var_37646_cast_fp16))[name = tensor<string, []>("op_37677_cast_fp16")];
            tensor<string, []> var_37679_equation_0 = const()[name = tensor<string, []>("op_37679_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37679_cast_fp16 = einsum(equation = var_37679_equation_0, values = (var_37521_cast_fp16, var_37647_cast_fp16))[name = tensor<string, []>("op_37679_cast_fp16")];
            tensor<string, []> var_37681_equation_0 = const()[name = tensor<string, []>("op_37681_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37681_cast_fp16 = einsum(equation = var_37681_equation_0, values = (var_37525_cast_fp16, var_37648_cast_fp16))[name = tensor<string, []>("op_37681_cast_fp16")];
            tensor<string, []> var_37683_equation_0 = const()[name = tensor<string, []>("op_37683_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37683_cast_fp16 = einsum(equation = var_37683_equation_0, values = (var_37529_cast_fp16, var_37649_cast_fp16))[name = tensor<string, []>("op_37683_cast_fp16")];
            tensor<string, []> var_37685_equation_0 = const()[name = tensor<string, []>("op_37685_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37685_cast_fp16 = einsum(equation = var_37685_equation_0, values = (var_37533_cast_fp16, var_37650_cast_fp16))[name = tensor<string, []>("op_37685_cast_fp16")];
            tensor<string, []> var_37687_equation_0 = const()[name = tensor<string, []>("op_37687_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37687_cast_fp16 = einsum(equation = var_37687_equation_0, values = (var_37537_cast_fp16, var_37651_cast_fp16))[name = tensor<string, []>("op_37687_cast_fp16")];
            tensor<string, []> var_37689_equation_0 = const()[name = tensor<string, []>("op_37689_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37689_cast_fp16 = einsum(equation = var_37689_equation_0, values = (var_37541_cast_fp16, var_37652_cast_fp16))[name = tensor<string, []>("op_37689_cast_fp16")];
            tensor<string, []> var_37691_equation_0 = const()[name = tensor<string, []>("op_37691_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37691_cast_fp16 = einsum(equation = var_37691_equation_0, values = (var_37545_cast_fp16, var_37653_cast_fp16))[name = tensor<string, []>("op_37691_cast_fp16")];
            tensor<string, []> var_37693_equation_0 = const()[name = tensor<string, []>("op_37693_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37693_cast_fp16 = einsum(equation = var_37693_equation_0, values = (var_37549_cast_fp16, var_37654_cast_fp16))[name = tensor<string, []>("op_37693_cast_fp16")];
            tensor<string, []> var_37695_equation_0 = const()[name = tensor<string, []>("op_37695_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37695_cast_fp16 = einsum(equation = var_37695_equation_0, values = (var_37553_cast_fp16, var_37655_cast_fp16))[name = tensor<string, []>("op_37695_cast_fp16")];
            tensor<bool, []> input_497_interleave_0 = const()[name = tensor<string, []>("input_497_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_497_cast_fp16 = concat(axis = var_30355, interleave = input_497_interleave_0, values = (var_37657_cast_fp16, var_37659_cast_fp16, var_37661_cast_fp16, var_37663_cast_fp16, var_37665_cast_fp16, var_37667_cast_fp16, var_37669_cast_fp16, var_37671_cast_fp16, var_37673_cast_fp16, var_37675_cast_fp16, var_37677_cast_fp16, var_37679_cast_fp16, var_37681_cast_fp16, var_37683_cast_fp16, var_37685_cast_fp16, var_37687_cast_fp16, var_37689_cast_fp16, var_37691_cast_fp16, var_37693_cast_fp16, var_37695_cast_fp16))[name = tensor<string, []>("input_497_cast_fp16")];
            tensor<string, []> var_37705_pad_type_0 = const()[name = tensor<string, []>("op_37705_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_37705_strides_0 = const()[name = tensor<string, []>("op_37705_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_37705_pad_0 = const()[name = tensor<string, []>("op_37705_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_37705_dilations_0 = const()[name = tensor<string, []>("op_37705_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_37705_groups_0 = const()[name = tensor<string, []>("op_37705_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_7_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(196489216))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(197718080))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_7_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_7_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_7_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(197718272)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_37705_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_7_attn2_to_out_0_bias_to_fp16, dilations = var_37705_dilations_0, groups = var_37705_groups_0, pad = var_37705_pad_0, pad_type = var_37705_pad_type_0, strides = var_37705_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_7_attn2_to_out_0_weight_to_fp16_palettized, x = input_497_cast_fp16)[name = tensor<string, []>("op_37705_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_251_cast_fp16 = add(x = var_37705_cast_fp16, y = inputs_249_cast_fp16)[name = tensor<string, []>("inputs_251_cast_fp16")];
            tensor<int32, [1]> input_499_axes_0 = const()[name = tensor<string, []>("input_499_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> input_499_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_499_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(197720896)))];
            tensor<fp16, [1280]> input_499_beta_0_to_fp16 = const()[name = tensor<string, []>("input_499_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(197723520)))];
            tensor<fp16, []> var_37715_to_fp16 = const()[name = tensor<string, []>("op_37715_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> input_499_cast_fp16 = layer_norm(axes = input_499_axes_0, beta = input_499_beta_0_to_fp16, epsilon = var_37715_to_fp16, gamma = input_499_gamma_0_to_fp16, x = inputs_251_cast_fp16)[name = tensor<string, []>("input_499_cast_fp16")];
            tensor<string, []> var_37735_pad_type_0 = const()[name = tensor<string, []>("op_37735_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_37735_strides_0 = const()[name = tensor<string, []>("op_37735_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_37735_pad_0 = const()[name = tensor<string, []>("op_37735_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_37735_dilations_0 = const()[name = tensor<string, []>("op_37735_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_37735_groups_0 = const()[name = tensor<string, []>("op_37735_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_7_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [9830400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(197726144))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(207556608))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_7_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([10240, 1280, 1, 1])];
            tensor<fp16, [10240]> up_blocks_0_attentions_0_transformer_blocks_7_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_7_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(207556800)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_37735_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_7_ff_net_0_proj_bias_to_fp16, dilations = var_37735_dilations_0, groups = var_37735_groups_0, pad = var_37735_pad_0, pad_type = var_37735_pad_type_0, strides = var_37735_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_7_ff_net_0_proj_weight_to_fp16_palettized, x = input_499_cast_fp16)[name = tensor<string, []>("op_37735_cast_fp16")];
            tensor<int32, [2]> var_37736_split_sizes_0 = const()[name = tensor<string, []>("op_37736_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_37736_axis_0 = const()[name = tensor<string, []>("op_37736_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_37736_cast_fp16_0, tensor<fp16, [2, 5120, 1, 1024]> var_37736_cast_fp16_1 = split(axis = var_37736_axis_0, split_sizes = var_37736_split_sizes_0, x = var_37735_cast_fp16)[name = tensor<string, []>("op_37736_cast_fp16")];
            tensor<string, []> var_37738_mode_0 = const()[name = tensor<string, []>("op_37738_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_37738_cast_fp16 = gelu(mode = var_37738_mode_0, x = var_37736_cast_fp16_1)[name = tensor<string, []>("op_37738_cast_fp16")];
            tensor<fp16, [2, 5120, 1, 1024]> input_501_cast_fp16 = mul(x = var_37736_cast_fp16_0, y = var_37738_cast_fp16)[name = tensor<string, []>("input_501_cast_fp16")];
            tensor<string, []> var_37746_pad_type_0 = const()[name = tensor<string, []>("op_37746_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_37746_strides_0 = const()[name = tensor<string, []>("op_37746_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_37746_pad_0 = const()[name = tensor<string, []>("op_37746_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_37746_dilations_0 = const()[name = tensor<string, []>("op_37746_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_37746_groups_0 = const()[name = tensor<string, []>("op_37746_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_7_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [4915200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(207577344))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(212492608))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_7_ff_net_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 5120, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_7_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_7_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(212492800)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_37746_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_7_ff_net_2_bias_to_fp16, dilations = var_37746_dilations_0, groups = var_37746_groups_0, pad = var_37746_pad_0, pad_type = var_37746_pad_type_0, strides = var_37746_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_7_ff_net_2_weight_to_fp16_palettized, x = input_501_cast_fp16)[name = tensor<string, []>("op_37746_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_253_cast_fp16 = add(x = var_37746_cast_fp16, y = inputs_251_cast_fp16)[name = tensor<string, []>("inputs_253_cast_fp16")];
            tensor<int32, [1]> hidden_states_335_axes_0 = const()[name = tensor<string, []>("hidden_states_335_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_335_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_335_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(212495424)))];
            tensor<fp16, [1280]> hidden_states_335_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_335_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(212498048)))];
            tensor<fp16, []> var_37762_to_fp16 = const()[name = tensor<string, []>("op_37762_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_335_cast_fp16 = layer_norm(axes = hidden_states_335_axes_0, beta = hidden_states_335_beta_0_to_fp16, epsilon = var_37762_to_fp16, gamma = hidden_states_335_gamma_0_to_fp16, x = inputs_253_cast_fp16)[name = tensor<string, []>("hidden_states_335_cast_fp16")];
            tensor<string, []> q_169_pad_type_0 = const()[name = tensor<string, []>("q_169_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_169_strides_0 = const()[name = tensor<string, []>("q_169_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_169_pad_0 = const()[name = tensor<string, []>("q_169_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_169_dilations_0 = const()[name = tensor<string, []>("q_169_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_169_groups_0 = const()[name = tensor<string, []>("q_169_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_8_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(212500672))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(213729536))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_8_attn1_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_169_cast_fp16 = conv(dilations = q_169_dilations_0, groups = q_169_groups_0, pad = q_169_pad_0, pad_type = q_169_pad_type_0, strides = q_169_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_8_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_335_cast_fp16)[name = tensor<string, []>("q_169_cast_fp16")];
            tensor<string, []> k_337_pad_type_0 = const()[name = tensor<string, []>("k_337_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_337_strides_0 = const()[name = tensor<string, []>("k_337_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_337_pad_0 = const()[name = tensor<string, []>("k_337_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_337_dilations_0 = const()[name = tensor<string, []>("k_337_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_337_groups_0 = const()[name = tensor<string, []>("k_337_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_8_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(213729728))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(214958592))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_8_attn1_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> k_337_cast_fp16 = conv(dilations = k_337_dilations_0, groups = k_337_groups_0, pad = k_337_pad_0, pad_type = k_337_pad_type_0, strides = k_337_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_8_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_335_cast_fp16)[name = tensor<string, []>("k_337_cast_fp16")];
            tensor<string, []> v_169_pad_type_0 = const()[name = tensor<string, []>("v_169_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_169_strides_0 = const()[name = tensor<string, []>("v_169_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_169_pad_0 = const()[name = tensor<string, []>("v_169_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_169_dilations_0 = const()[name = tensor<string, []>("v_169_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_169_groups_0 = const()[name = tensor<string, []>("v_169_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_8_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(214958784))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(216187648))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_8_attn1_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> v_169_cast_fp16 = conv(dilations = v_169_dilations_0, groups = v_169_groups_0, pad = v_169_pad_0, pad_type = v_169_pad_type_0, strides = v_169_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_8_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_335_cast_fp16)[name = tensor<string, []>("v_169_cast_fp16")];
            tensor<int32, [4]> var_37795_begin_0 = const()[name = tensor<string, []>("op_37795_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_37795_end_0 = const()[name = tensor<string, []>("op_37795_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_37795_end_mask_0 = const()[name = tensor<string, []>("op_37795_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37795_cast_fp16 = slice_by_index(begin = var_37795_begin_0, end = var_37795_end_0, end_mask = var_37795_end_mask_0, x = q_169_cast_fp16)[name = tensor<string, []>("op_37795_cast_fp16")];
            tensor<int32, [4]> var_37799_begin_0 = const()[name = tensor<string, []>("op_37799_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_37799_end_0 = const()[name = tensor<string, []>("op_37799_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_37799_end_mask_0 = const()[name = tensor<string, []>("op_37799_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37799_cast_fp16 = slice_by_index(begin = var_37799_begin_0, end = var_37799_end_0, end_mask = var_37799_end_mask_0, x = q_169_cast_fp16)[name = tensor<string, []>("op_37799_cast_fp16")];
            tensor<int32, [4]> var_37803_begin_0 = const()[name = tensor<string, []>("op_37803_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_37803_end_0 = const()[name = tensor<string, []>("op_37803_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_37803_end_mask_0 = const()[name = tensor<string, []>("op_37803_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37803_cast_fp16 = slice_by_index(begin = var_37803_begin_0, end = var_37803_end_0, end_mask = var_37803_end_mask_0, x = q_169_cast_fp16)[name = tensor<string, []>("op_37803_cast_fp16")];
            tensor<int32, [4]> var_37807_begin_0 = const()[name = tensor<string, []>("op_37807_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_37807_end_0 = const()[name = tensor<string, []>("op_37807_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_37807_end_mask_0 = const()[name = tensor<string, []>("op_37807_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37807_cast_fp16 = slice_by_index(begin = var_37807_begin_0, end = var_37807_end_0, end_mask = var_37807_end_mask_0, x = q_169_cast_fp16)[name = tensor<string, []>("op_37807_cast_fp16")];
            tensor<int32, [4]> var_37811_begin_0 = const()[name = tensor<string, []>("op_37811_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_37811_end_0 = const()[name = tensor<string, []>("op_37811_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_37811_end_mask_0 = const()[name = tensor<string, []>("op_37811_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37811_cast_fp16 = slice_by_index(begin = var_37811_begin_0, end = var_37811_end_0, end_mask = var_37811_end_mask_0, x = q_169_cast_fp16)[name = tensor<string, []>("op_37811_cast_fp16")];
            tensor<int32, [4]> var_37815_begin_0 = const()[name = tensor<string, []>("op_37815_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_37815_end_0 = const()[name = tensor<string, []>("op_37815_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_37815_end_mask_0 = const()[name = tensor<string, []>("op_37815_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37815_cast_fp16 = slice_by_index(begin = var_37815_begin_0, end = var_37815_end_0, end_mask = var_37815_end_mask_0, x = q_169_cast_fp16)[name = tensor<string, []>("op_37815_cast_fp16")];
            tensor<int32, [4]> var_37819_begin_0 = const()[name = tensor<string, []>("op_37819_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_37819_end_0 = const()[name = tensor<string, []>("op_37819_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_37819_end_mask_0 = const()[name = tensor<string, []>("op_37819_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37819_cast_fp16 = slice_by_index(begin = var_37819_begin_0, end = var_37819_end_0, end_mask = var_37819_end_mask_0, x = q_169_cast_fp16)[name = tensor<string, []>("op_37819_cast_fp16")];
            tensor<int32, [4]> var_37823_begin_0 = const()[name = tensor<string, []>("op_37823_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_37823_end_0 = const()[name = tensor<string, []>("op_37823_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_37823_end_mask_0 = const()[name = tensor<string, []>("op_37823_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37823_cast_fp16 = slice_by_index(begin = var_37823_begin_0, end = var_37823_end_0, end_mask = var_37823_end_mask_0, x = q_169_cast_fp16)[name = tensor<string, []>("op_37823_cast_fp16")];
            tensor<int32, [4]> var_37827_begin_0 = const()[name = tensor<string, []>("op_37827_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_37827_end_0 = const()[name = tensor<string, []>("op_37827_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_37827_end_mask_0 = const()[name = tensor<string, []>("op_37827_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37827_cast_fp16 = slice_by_index(begin = var_37827_begin_0, end = var_37827_end_0, end_mask = var_37827_end_mask_0, x = q_169_cast_fp16)[name = tensor<string, []>("op_37827_cast_fp16")];
            tensor<int32, [4]> var_37831_begin_0 = const()[name = tensor<string, []>("op_37831_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_37831_end_0 = const()[name = tensor<string, []>("op_37831_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_37831_end_mask_0 = const()[name = tensor<string, []>("op_37831_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37831_cast_fp16 = slice_by_index(begin = var_37831_begin_0, end = var_37831_end_0, end_mask = var_37831_end_mask_0, x = q_169_cast_fp16)[name = tensor<string, []>("op_37831_cast_fp16")];
            tensor<int32, [4]> var_37835_begin_0 = const()[name = tensor<string, []>("op_37835_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_37835_end_0 = const()[name = tensor<string, []>("op_37835_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_37835_end_mask_0 = const()[name = tensor<string, []>("op_37835_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37835_cast_fp16 = slice_by_index(begin = var_37835_begin_0, end = var_37835_end_0, end_mask = var_37835_end_mask_0, x = q_169_cast_fp16)[name = tensor<string, []>("op_37835_cast_fp16")];
            tensor<int32, [4]> var_37839_begin_0 = const()[name = tensor<string, []>("op_37839_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_37839_end_0 = const()[name = tensor<string, []>("op_37839_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_37839_end_mask_0 = const()[name = tensor<string, []>("op_37839_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37839_cast_fp16 = slice_by_index(begin = var_37839_begin_0, end = var_37839_end_0, end_mask = var_37839_end_mask_0, x = q_169_cast_fp16)[name = tensor<string, []>("op_37839_cast_fp16")];
            tensor<int32, [4]> var_37843_begin_0 = const()[name = tensor<string, []>("op_37843_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_37843_end_0 = const()[name = tensor<string, []>("op_37843_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_37843_end_mask_0 = const()[name = tensor<string, []>("op_37843_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37843_cast_fp16 = slice_by_index(begin = var_37843_begin_0, end = var_37843_end_0, end_mask = var_37843_end_mask_0, x = q_169_cast_fp16)[name = tensor<string, []>("op_37843_cast_fp16")];
            tensor<int32, [4]> var_37847_begin_0 = const()[name = tensor<string, []>("op_37847_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_37847_end_0 = const()[name = tensor<string, []>("op_37847_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_37847_end_mask_0 = const()[name = tensor<string, []>("op_37847_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37847_cast_fp16 = slice_by_index(begin = var_37847_begin_0, end = var_37847_end_0, end_mask = var_37847_end_mask_0, x = q_169_cast_fp16)[name = tensor<string, []>("op_37847_cast_fp16")];
            tensor<int32, [4]> var_37851_begin_0 = const()[name = tensor<string, []>("op_37851_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_37851_end_0 = const()[name = tensor<string, []>("op_37851_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_37851_end_mask_0 = const()[name = tensor<string, []>("op_37851_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37851_cast_fp16 = slice_by_index(begin = var_37851_begin_0, end = var_37851_end_0, end_mask = var_37851_end_mask_0, x = q_169_cast_fp16)[name = tensor<string, []>("op_37851_cast_fp16")];
            tensor<int32, [4]> var_37855_begin_0 = const()[name = tensor<string, []>("op_37855_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_37855_end_0 = const()[name = tensor<string, []>("op_37855_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_37855_end_mask_0 = const()[name = tensor<string, []>("op_37855_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37855_cast_fp16 = slice_by_index(begin = var_37855_begin_0, end = var_37855_end_0, end_mask = var_37855_end_mask_0, x = q_169_cast_fp16)[name = tensor<string, []>("op_37855_cast_fp16")];
            tensor<int32, [4]> var_37859_begin_0 = const()[name = tensor<string, []>("op_37859_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_37859_end_0 = const()[name = tensor<string, []>("op_37859_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_37859_end_mask_0 = const()[name = tensor<string, []>("op_37859_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37859_cast_fp16 = slice_by_index(begin = var_37859_begin_0, end = var_37859_end_0, end_mask = var_37859_end_mask_0, x = q_169_cast_fp16)[name = tensor<string, []>("op_37859_cast_fp16")];
            tensor<int32, [4]> var_37863_begin_0 = const()[name = tensor<string, []>("op_37863_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_37863_end_0 = const()[name = tensor<string, []>("op_37863_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_37863_end_mask_0 = const()[name = tensor<string, []>("op_37863_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37863_cast_fp16 = slice_by_index(begin = var_37863_begin_0, end = var_37863_end_0, end_mask = var_37863_end_mask_0, x = q_169_cast_fp16)[name = tensor<string, []>("op_37863_cast_fp16")];
            tensor<int32, [4]> var_37867_begin_0 = const()[name = tensor<string, []>("op_37867_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_37867_end_0 = const()[name = tensor<string, []>("op_37867_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_37867_end_mask_0 = const()[name = tensor<string, []>("op_37867_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37867_cast_fp16 = slice_by_index(begin = var_37867_begin_0, end = var_37867_end_0, end_mask = var_37867_end_mask_0, x = q_169_cast_fp16)[name = tensor<string, []>("op_37867_cast_fp16")];
            tensor<int32, [4]> var_37871_begin_0 = const()[name = tensor<string, []>("op_37871_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_37871_end_0 = const()[name = tensor<string, []>("op_37871_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_37871_end_mask_0 = const()[name = tensor<string, []>("op_37871_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37871_cast_fp16 = slice_by_index(begin = var_37871_begin_0, end = var_37871_end_0, end_mask = var_37871_end_mask_0, x = q_169_cast_fp16)[name = tensor<string, []>("op_37871_cast_fp16")];
            tensor<int32, [4]> k_339_perm_0 = const()[name = tensor<string, []>("k_339_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_37878_begin_0 = const()[name = tensor<string, []>("op_37878_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_37878_end_0 = const()[name = tensor<string, []>("op_37878_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_37878_end_mask_0 = const()[name = tensor<string, []>("op_37878_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> k_339_cast_fp16 = transpose(perm = k_339_perm_0, x = k_337_cast_fp16)[name = tensor<string, []>("transpose_55")];
            tensor<fp16, [2, 1024, 1, 64]> var_37878_cast_fp16 = slice_by_index(begin = var_37878_begin_0, end = var_37878_end_0, end_mask = var_37878_end_mask_0, x = k_339_cast_fp16)[name = tensor<string, []>("op_37878_cast_fp16")];
            tensor<int32, [4]> var_37882_begin_0 = const()[name = tensor<string, []>("op_37882_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_37882_end_0 = const()[name = tensor<string, []>("op_37882_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_37882_end_mask_0 = const()[name = tensor<string, []>("op_37882_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37882_cast_fp16 = slice_by_index(begin = var_37882_begin_0, end = var_37882_end_0, end_mask = var_37882_end_mask_0, x = k_339_cast_fp16)[name = tensor<string, []>("op_37882_cast_fp16")];
            tensor<int32, [4]> var_37886_begin_0 = const()[name = tensor<string, []>("op_37886_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_37886_end_0 = const()[name = tensor<string, []>("op_37886_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_37886_end_mask_0 = const()[name = tensor<string, []>("op_37886_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37886_cast_fp16 = slice_by_index(begin = var_37886_begin_0, end = var_37886_end_0, end_mask = var_37886_end_mask_0, x = k_339_cast_fp16)[name = tensor<string, []>("op_37886_cast_fp16")];
            tensor<int32, [4]> var_37890_begin_0 = const()[name = tensor<string, []>("op_37890_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_37890_end_0 = const()[name = tensor<string, []>("op_37890_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_37890_end_mask_0 = const()[name = tensor<string, []>("op_37890_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37890_cast_fp16 = slice_by_index(begin = var_37890_begin_0, end = var_37890_end_0, end_mask = var_37890_end_mask_0, x = k_339_cast_fp16)[name = tensor<string, []>("op_37890_cast_fp16")];
            tensor<int32, [4]> var_37894_begin_0 = const()[name = tensor<string, []>("op_37894_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_37894_end_0 = const()[name = tensor<string, []>("op_37894_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_37894_end_mask_0 = const()[name = tensor<string, []>("op_37894_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37894_cast_fp16 = slice_by_index(begin = var_37894_begin_0, end = var_37894_end_0, end_mask = var_37894_end_mask_0, x = k_339_cast_fp16)[name = tensor<string, []>("op_37894_cast_fp16")];
            tensor<int32, [4]> var_37898_begin_0 = const()[name = tensor<string, []>("op_37898_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_37898_end_0 = const()[name = tensor<string, []>("op_37898_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_37898_end_mask_0 = const()[name = tensor<string, []>("op_37898_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37898_cast_fp16 = slice_by_index(begin = var_37898_begin_0, end = var_37898_end_0, end_mask = var_37898_end_mask_0, x = k_339_cast_fp16)[name = tensor<string, []>("op_37898_cast_fp16")];
            tensor<int32, [4]> var_37902_begin_0 = const()[name = tensor<string, []>("op_37902_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_37902_end_0 = const()[name = tensor<string, []>("op_37902_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_37902_end_mask_0 = const()[name = tensor<string, []>("op_37902_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37902_cast_fp16 = slice_by_index(begin = var_37902_begin_0, end = var_37902_end_0, end_mask = var_37902_end_mask_0, x = k_339_cast_fp16)[name = tensor<string, []>("op_37902_cast_fp16")];
            tensor<int32, [4]> var_37906_begin_0 = const()[name = tensor<string, []>("op_37906_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_37906_end_0 = const()[name = tensor<string, []>("op_37906_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_37906_end_mask_0 = const()[name = tensor<string, []>("op_37906_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37906_cast_fp16 = slice_by_index(begin = var_37906_begin_0, end = var_37906_end_0, end_mask = var_37906_end_mask_0, x = k_339_cast_fp16)[name = tensor<string, []>("op_37906_cast_fp16")];
            tensor<int32, [4]> var_37910_begin_0 = const()[name = tensor<string, []>("op_37910_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_37910_end_0 = const()[name = tensor<string, []>("op_37910_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_37910_end_mask_0 = const()[name = tensor<string, []>("op_37910_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37910_cast_fp16 = slice_by_index(begin = var_37910_begin_0, end = var_37910_end_0, end_mask = var_37910_end_mask_0, x = k_339_cast_fp16)[name = tensor<string, []>("op_37910_cast_fp16")];
            tensor<int32, [4]> var_37914_begin_0 = const()[name = tensor<string, []>("op_37914_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_37914_end_0 = const()[name = tensor<string, []>("op_37914_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_37914_end_mask_0 = const()[name = tensor<string, []>("op_37914_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37914_cast_fp16 = slice_by_index(begin = var_37914_begin_0, end = var_37914_end_0, end_mask = var_37914_end_mask_0, x = k_339_cast_fp16)[name = tensor<string, []>("op_37914_cast_fp16")];
            tensor<int32, [4]> var_37918_begin_0 = const()[name = tensor<string, []>("op_37918_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_37918_end_0 = const()[name = tensor<string, []>("op_37918_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_37918_end_mask_0 = const()[name = tensor<string, []>("op_37918_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37918_cast_fp16 = slice_by_index(begin = var_37918_begin_0, end = var_37918_end_0, end_mask = var_37918_end_mask_0, x = k_339_cast_fp16)[name = tensor<string, []>("op_37918_cast_fp16")];
            tensor<int32, [4]> var_37922_begin_0 = const()[name = tensor<string, []>("op_37922_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_37922_end_0 = const()[name = tensor<string, []>("op_37922_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_37922_end_mask_0 = const()[name = tensor<string, []>("op_37922_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37922_cast_fp16 = slice_by_index(begin = var_37922_begin_0, end = var_37922_end_0, end_mask = var_37922_end_mask_0, x = k_339_cast_fp16)[name = tensor<string, []>("op_37922_cast_fp16")];
            tensor<int32, [4]> var_37926_begin_0 = const()[name = tensor<string, []>("op_37926_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_37926_end_0 = const()[name = tensor<string, []>("op_37926_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_37926_end_mask_0 = const()[name = tensor<string, []>("op_37926_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37926_cast_fp16 = slice_by_index(begin = var_37926_begin_0, end = var_37926_end_0, end_mask = var_37926_end_mask_0, x = k_339_cast_fp16)[name = tensor<string, []>("op_37926_cast_fp16")];
            tensor<int32, [4]> var_37930_begin_0 = const()[name = tensor<string, []>("op_37930_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_37930_end_0 = const()[name = tensor<string, []>("op_37930_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_37930_end_mask_0 = const()[name = tensor<string, []>("op_37930_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37930_cast_fp16 = slice_by_index(begin = var_37930_begin_0, end = var_37930_end_0, end_mask = var_37930_end_mask_0, x = k_339_cast_fp16)[name = tensor<string, []>("op_37930_cast_fp16")];
            tensor<int32, [4]> var_37934_begin_0 = const()[name = tensor<string, []>("op_37934_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_37934_end_0 = const()[name = tensor<string, []>("op_37934_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_37934_end_mask_0 = const()[name = tensor<string, []>("op_37934_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37934_cast_fp16 = slice_by_index(begin = var_37934_begin_0, end = var_37934_end_0, end_mask = var_37934_end_mask_0, x = k_339_cast_fp16)[name = tensor<string, []>("op_37934_cast_fp16")];
            tensor<int32, [4]> var_37938_begin_0 = const()[name = tensor<string, []>("op_37938_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_37938_end_0 = const()[name = tensor<string, []>("op_37938_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_37938_end_mask_0 = const()[name = tensor<string, []>("op_37938_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37938_cast_fp16 = slice_by_index(begin = var_37938_begin_0, end = var_37938_end_0, end_mask = var_37938_end_mask_0, x = k_339_cast_fp16)[name = tensor<string, []>("op_37938_cast_fp16")];
            tensor<int32, [4]> var_37942_begin_0 = const()[name = tensor<string, []>("op_37942_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_37942_end_0 = const()[name = tensor<string, []>("op_37942_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_37942_end_mask_0 = const()[name = tensor<string, []>("op_37942_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37942_cast_fp16 = slice_by_index(begin = var_37942_begin_0, end = var_37942_end_0, end_mask = var_37942_end_mask_0, x = k_339_cast_fp16)[name = tensor<string, []>("op_37942_cast_fp16")];
            tensor<int32, [4]> var_37946_begin_0 = const()[name = tensor<string, []>("op_37946_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_37946_end_0 = const()[name = tensor<string, []>("op_37946_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_37946_end_mask_0 = const()[name = tensor<string, []>("op_37946_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37946_cast_fp16 = slice_by_index(begin = var_37946_begin_0, end = var_37946_end_0, end_mask = var_37946_end_mask_0, x = k_339_cast_fp16)[name = tensor<string, []>("op_37946_cast_fp16")];
            tensor<int32, [4]> var_37950_begin_0 = const()[name = tensor<string, []>("op_37950_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_37950_end_0 = const()[name = tensor<string, []>("op_37950_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_37950_end_mask_0 = const()[name = tensor<string, []>("op_37950_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37950_cast_fp16 = slice_by_index(begin = var_37950_begin_0, end = var_37950_end_0, end_mask = var_37950_end_mask_0, x = k_339_cast_fp16)[name = tensor<string, []>("op_37950_cast_fp16")];
            tensor<int32, [4]> var_37954_begin_0 = const()[name = tensor<string, []>("op_37954_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_37954_end_0 = const()[name = tensor<string, []>("op_37954_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_37954_end_mask_0 = const()[name = tensor<string, []>("op_37954_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37954_cast_fp16 = slice_by_index(begin = var_37954_begin_0, end = var_37954_end_0, end_mask = var_37954_end_mask_0, x = k_339_cast_fp16)[name = tensor<string, []>("op_37954_cast_fp16")];
            tensor<int32, [4]> var_37956_begin_0 = const()[name = tensor<string, []>("op_37956_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_37956_end_0 = const()[name = tensor<string, []>("op_37956_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_37956_end_mask_0 = const()[name = tensor<string, []>("op_37956_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37956_cast_fp16 = slice_by_index(begin = var_37956_begin_0, end = var_37956_end_0, end_mask = var_37956_end_mask_0, x = v_169_cast_fp16)[name = tensor<string, []>("op_37956_cast_fp16")];
            tensor<int32, [4]> var_37960_begin_0 = const()[name = tensor<string, []>("op_37960_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_37960_end_0 = const()[name = tensor<string, []>("op_37960_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_37960_end_mask_0 = const()[name = tensor<string, []>("op_37960_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37960_cast_fp16 = slice_by_index(begin = var_37960_begin_0, end = var_37960_end_0, end_mask = var_37960_end_mask_0, x = v_169_cast_fp16)[name = tensor<string, []>("op_37960_cast_fp16")];
            tensor<int32, [4]> var_37964_begin_0 = const()[name = tensor<string, []>("op_37964_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_37964_end_0 = const()[name = tensor<string, []>("op_37964_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_37964_end_mask_0 = const()[name = tensor<string, []>("op_37964_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37964_cast_fp16 = slice_by_index(begin = var_37964_begin_0, end = var_37964_end_0, end_mask = var_37964_end_mask_0, x = v_169_cast_fp16)[name = tensor<string, []>("op_37964_cast_fp16")];
            tensor<int32, [4]> var_37968_begin_0 = const()[name = tensor<string, []>("op_37968_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_37968_end_0 = const()[name = tensor<string, []>("op_37968_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_37968_end_mask_0 = const()[name = tensor<string, []>("op_37968_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37968_cast_fp16 = slice_by_index(begin = var_37968_begin_0, end = var_37968_end_0, end_mask = var_37968_end_mask_0, x = v_169_cast_fp16)[name = tensor<string, []>("op_37968_cast_fp16")];
            tensor<int32, [4]> var_37972_begin_0 = const()[name = tensor<string, []>("op_37972_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_37972_end_0 = const()[name = tensor<string, []>("op_37972_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_37972_end_mask_0 = const()[name = tensor<string, []>("op_37972_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37972_cast_fp16 = slice_by_index(begin = var_37972_begin_0, end = var_37972_end_0, end_mask = var_37972_end_mask_0, x = v_169_cast_fp16)[name = tensor<string, []>("op_37972_cast_fp16")];
            tensor<int32, [4]> var_37976_begin_0 = const()[name = tensor<string, []>("op_37976_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_37976_end_0 = const()[name = tensor<string, []>("op_37976_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_37976_end_mask_0 = const()[name = tensor<string, []>("op_37976_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37976_cast_fp16 = slice_by_index(begin = var_37976_begin_0, end = var_37976_end_0, end_mask = var_37976_end_mask_0, x = v_169_cast_fp16)[name = tensor<string, []>("op_37976_cast_fp16")];
            tensor<int32, [4]> var_37980_begin_0 = const()[name = tensor<string, []>("op_37980_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_37980_end_0 = const()[name = tensor<string, []>("op_37980_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_37980_end_mask_0 = const()[name = tensor<string, []>("op_37980_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37980_cast_fp16 = slice_by_index(begin = var_37980_begin_0, end = var_37980_end_0, end_mask = var_37980_end_mask_0, x = v_169_cast_fp16)[name = tensor<string, []>("op_37980_cast_fp16")];
            tensor<int32, [4]> var_37984_begin_0 = const()[name = tensor<string, []>("op_37984_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_37984_end_0 = const()[name = tensor<string, []>("op_37984_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_37984_end_mask_0 = const()[name = tensor<string, []>("op_37984_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37984_cast_fp16 = slice_by_index(begin = var_37984_begin_0, end = var_37984_end_0, end_mask = var_37984_end_mask_0, x = v_169_cast_fp16)[name = tensor<string, []>("op_37984_cast_fp16")];
            tensor<int32, [4]> var_37988_begin_0 = const()[name = tensor<string, []>("op_37988_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_37988_end_0 = const()[name = tensor<string, []>("op_37988_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_37988_end_mask_0 = const()[name = tensor<string, []>("op_37988_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37988_cast_fp16 = slice_by_index(begin = var_37988_begin_0, end = var_37988_end_0, end_mask = var_37988_end_mask_0, x = v_169_cast_fp16)[name = tensor<string, []>("op_37988_cast_fp16")];
            tensor<int32, [4]> var_37992_begin_0 = const()[name = tensor<string, []>("op_37992_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_37992_end_0 = const()[name = tensor<string, []>("op_37992_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_37992_end_mask_0 = const()[name = tensor<string, []>("op_37992_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37992_cast_fp16 = slice_by_index(begin = var_37992_begin_0, end = var_37992_end_0, end_mask = var_37992_end_mask_0, x = v_169_cast_fp16)[name = tensor<string, []>("op_37992_cast_fp16")];
            tensor<int32, [4]> var_37996_begin_0 = const()[name = tensor<string, []>("op_37996_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_37996_end_0 = const()[name = tensor<string, []>("op_37996_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_37996_end_mask_0 = const()[name = tensor<string, []>("op_37996_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37996_cast_fp16 = slice_by_index(begin = var_37996_begin_0, end = var_37996_end_0, end_mask = var_37996_end_mask_0, x = v_169_cast_fp16)[name = tensor<string, []>("op_37996_cast_fp16")];
            tensor<int32, [4]> var_38000_begin_0 = const()[name = tensor<string, []>("op_38000_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_38000_end_0 = const()[name = tensor<string, []>("op_38000_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_38000_end_mask_0 = const()[name = tensor<string, []>("op_38000_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38000_cast_fp16 = slice_by_index(begin = var_38000_begin_0, end = var_38000_end_0, end_mask = var_38000_end_mask_0, x = v_169_cast_fp16)[name = tensor<string, []>("op_38000_cast_fp16")];
            tensor<int32, [4]> var_38004_begin_0 = const()[name = tensor<string, []>("op_38004_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_38004_end_0 = const()[name = tensor<string, []>("op_38004_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_38004_end_mask_0 = const()[name = tensor<string, []>("op_38004_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38004_cast_fp16 = slice_by_index(begin = var_38004_begin_0, end = var_38004_end_0, end_mask = var_38004_end_mask_0, x = v_169_cast_fp16)[name = tensor<string, []>("op_38004_cast_fp16")];
            tensor<int32, [4]> var_38008_begin_0 = const()[name = tensor<string, []>("op_38008_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_38008_end_0 = const()[name = tensor<string, []>("op_38008_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_38008_end_mask_0 = const()[name = tensor<string, []>("op_38008_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38008_cast_fp16 = slice_by_index(begin = var_38008_begin_0, end = var_38008_end_0, end_mask = var_38008_end_mask_0, x = v_169_cast_fp16)[name = tensor<string, []>("op_38008_cast_fp16")];
            tensor<int32, [4]> var_38012_begin_0 = const()[name = tensor<string, []>("op_38012_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_38012_end_0 = const()[name = tensor<string, []>("op_38012_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_38012_end_mask_0 = const()[name = tensor<string, []>("op_38012_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38012_cast_fp16 = slice_by_index(begin = var_38012_begin_0, end = var_38012_end_0, end_mask = var_38012_end_mask_0, x = v_169_cast_fp16)[name = tensor<string, []>("op_38012_cast_fp16")];
            tensor<int32, [4]> var_38016_begin_0 = const()[name = tensor<string, []>("op_38016_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_38016_end_0 = const()[name = tensor<string, []>("op_38016_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_38016_end_mask_0 = const()[name = tensor<string, []>("op_38016_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38016_cast_fp16 = slice_by_index(begin = var_38016_begin_0, end = var_38016_end_0, end_mask = var_38016_end_mask_0, x = v_169_cast_fp16)[name = tensor<string, []>("op_38016_cast_fp16")];
            tensor<int32, [4]> var_38020_begin_0 = const()[name = tensor<string, []>("op_38020_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_38020_end_0 = const()[name = tensor<string, []>("op_38020_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_38020_end_mask_0 = const()[name = tensor<string, []>("op_38020_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38020_cast_fp16 = slice_by_index(begin = var_38020_begin_0, end = var_38020_end_0, end_mask = var_38020_end_mask_0, x = v_169_cast_fp16)[name = tensor<string, []>("op_38020_cast_fp16")];
            tensor<int32, [4]> var_38024_begin_0 = const()[name = tensor<string, []>("op_38024_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_38024_end_0 = const()[name = tensor<string, []>("op_38024_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_38024_end_mask_0 = const()[name = tensor<string, []>("op_38024_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38024_cast_fp16 = slice_by_index(begin = var_38024_begin_0, end = var_38024_end_0, end_mask = var_38024_end_mask_0, x = v_169_cast_fp16)[name = tensor<string, []>("op_38024_cast_fp16")];
            tensor<int32, [4]> var_38028_begin_0 = const()[name = tensor<string, []>("op_38028_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_38028_end_0 = const()[name = tensor<string, []>("op_38028_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_38028_end_mask_0 = const()[name = tensor<string, []>("op_38028_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38028_cast_fp16 = slice_by_index(begin = var_38028_begin_0, end = var_38028_end_0, end_mask = var_38028_end_mask_0, x = v_169_cast_fp16)[name = tensor<string, []>("op_38028_cast_fp16")];
            tensor<int32, [4]> var_38032_begin_0 = const()[name = tensor<string, []>("op_38032_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_38032_end_0 = const()[name = tensor<string, []>("op_38032_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_38032_end_mask_0 = const()[name = tensor<string, []>("op_38032_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38032_cast_fp16 = slice_by_index(begin = var_38032_begin_0, end = var_38032_end_0, end_mask = var_38032_end_mask_0, x = v_169_cast_fp16)[name = tensor<string, []>("op_38032_cast_fp16")];
            tensor<string, []> var_38036_equation_0 = const()[name = tensor<string, []>("op_38036_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38036_cast_fp16 = einsum(equation = var_38036_equation_0, values = (var_37878_cast_fp16, var_37795_cast_fp16))[name = tensor<string, []>("op_38036_cast_fp16")];
            tensor<fp16, []> var_38037_to_fp16 = const()[name = tensor<string, []>("op_38037_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3201_cast_fp16 = mul(x = var_38036_cast_fp16, y = var_38037_to_fp16)[name = tensor<string, []>("aw_3201_cast_fp16")];
            tensor<string, []> var_38040_equation_0 = const()[name = tensor<string, []>("op_38040_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38040_cast_fp16 = einsum(equation = var_38040_equation_0, values = (var_37882_cast_fp16, var_37799_cast_fp16))[name = tensor<string, []>("op_38040_cast_fp16")];
            tensor<fp16, []> var_38041_to_fp16 = const()[name = tensor<string, []>("op_38041_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3203_cast_fp16 = mul(x = var_38040_cast_fp16, y = var_38041_to_fp16)[name = tensor<string, []>("aw_3203_cast_fp16")];
            tensor<string, []> var_38044_equation_0 = const()[name = tensor<string, []>("op_38044_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38044_cast_fp16 = einsum(equation = var_38044_equation_0, values = (var_37886_cast_fp16, var_37803_cast_fp16))[name = tensor<string, []>("op_38044_cast_fp16")];
            tensor<fp16, []> var_38045_to_fp16 = const()[name = tensor<string, []>("op_38045_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3205_cast_fp16 = mul(x = var_38044_cast_fp16, y = var_38045_to_fp16)[name = tensor<string, []>("aw_3205_cast_fp16")];
            tensor<string, []> var_38048_equation_0 = const()[name = tensor<string, []>("op_38048_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38048_cast_fp16 = einsum(equation = var_38048_equation_0, values = (var_37890_cast_fp16, var_37807_cast_fp16))[name = tensor<string, []>("op_38048_cast_fp16")];
            tensor<fp16, []> var_38049_to_fp16 = const()[name = tensor<string, []>("op_38049_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3207_cast_fp16 = mul(x = var_38048_cast_fp16, y = var_38049_to_fp16)[name = tensor<string, []>("aw_3207_cast_fp16")];
            tensor<string, []> var_38052_equation_0 = const()[name = tensor<string, []>("op_38052_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38052_cast_fp16 = einsum(equation = var_38052_equation_0, values = (var_37894_cast_fp16, var_37811_cast_fp16))[name = tensor<string, []>("op_38052_cast_fp16")];
            tensor<fp16, []> var_38053_to_fp16 = const()[name = tensor<string, []>("op_38053_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3209_cast_fp16 = mul(x = var_38052_cast_fp16, y = var_38053_to_fp16)[name = tensor<string, []>("aw_3209_cast_fp16")];
            tensor<string, []> var_38056_equation_0 = const()[name = tensor<string, []>("op_38056_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38056_cast_fp16 = einsum(equation = var_38056_equation_0, values = (var_37898_cast_fp16, var_37815_cast_fp16))[name = tensor<string, []>("op_38056_cast_fp16")];
            tensor<fp16, []> var_38057_to_fp16 = const()[name = tensor<string, []>("op_38057_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3211_cast_fp16 = mul(x = var_38056_cast_fp16, y = var_38057_to_fp16)[name = tensor<string, []>("aw_3211_cast_fp16")];
            tensor<string, []> var_38060_equation_0 = const()[name = tensor<string, []>("op_38060_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38060_cast_fp16 = einsum(equation = var_38060_equation_0, values = (var_37902_cast_fp16, var_37819_cast_fp16))[name = tensor<string, []>("op_38060_cast_fp16")];
            tensor<fp16, []> var_38061_to_fp16 = const()[name = tensor<string, []>("op_38061_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3213_cast_fp16 = mul(x = var_38060_cast_fp16, y = var_38061_to_fp16)[name = tensor<string, []>("aw_3213_cast_fp16")];
            tensor<string, []> var_38064_equation_0 = const()[name = tensor<string, []>("op_38064_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38064_cast_fp16 = einsum(equation = var_38064_equation_0, values = (var_37906_cast_fp16, var_37823_cast_fp16))[name = tensor<string, []>("op_38064_cast_fp16")];
            tensor<fp16, []> var_38065_to_fp16 = const()[name = tensor<string, []>("op_38065_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3215_cast_fp16 = mul(x = var_38064_cast_fp16, y = var_38065_to_fp16)[name = tensor<string, []>("aw_3215_cast_fp16")];
            tensor<string, []> var_38068_equation_0 = const()[name = tensor<string, []>("op_38068_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38068_cast_fp16 = einsum(equation = var_38068_equation_0, values = (var_37910_cast_fp16, var_37827_cast_fp16))[name = tensor<string, []>("op_38068_cast_fp16")];
            tensor<fp16, []> var_38069_to_fp16 = const()[name = tensor<string, []>("op_38069_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3217_cast_fp16 = mul(x = var_38068_cast_fp16, y = var_38069_to_fp16)[name = tensor<string, []>("aw_3217_cast_fp16")];
            tensor<string, []> var_38072_equation_0 = const()[name = tensor<string, []>("op_38072_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38072_cast_fp16 = einsum(equation = var_38072_equation_0, values = (var_37914_cast_fp16, var_37831_cast_fp16))[name = tensor<string, []>("op_38072_cast_fp16")];
            tensor<fp16, []> var_38073_to_fp16 = const()[name = tensor<string, []>("op_38073_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3219_cast_fp16 = mul(x = var_38072_cast_fp16, y = var_38073_to_fp16)[name = tensor<string, []>("aw_3219_cast_fp16")];
            tensor<string, []> var_38076_equation_0 = const()[name = tensor<string, []>("op_38076_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38076_cast_fp16 = einsum(equation = var_38076_equation_0, values = (var_37918_cast_fp16, var_37835_cast_fp16))[name = tensor<string, []>("op_38076_cast_fp16")];
            tensor<fp16, []> var_38077_to_fp16 = const()[name = tensor<string, []>("op_38077_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3221_cast_fp16 = mul(x = var_38076_cast_fp16, y = var_38077_to_fp16)[name = tensor<string, []>("aw_3221_cast_fp16")];
            tensor<string, []> var_38080_equation_0 = const()[name = tensor<string, []>("op_38080_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38080_cast_fp16 = einsum(equation = var_38080_equation_0, values = (var_37922_cast_fp16, var_37839_cast_fp16))[name = tensor<string, []>("op_38080_cast_fp16")];
            tensor<fp16, []> var_38081_to_fp16 = const()[name = tensor<string, []>("op_38081_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3223_cast_fp16 = mul(x = var_38080_cast_fp16, y = var_38081_to_fp16)[name = tensor<string, []>("aw_3223_cast_fp16")];
            tensor<string, []> var_38084_equation_0 = const()[name = tensor<string, []>("op_38084_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38084_cast_fp16 = einsum(equation = var_38084_equation_0, values = (var_37926_cast_fp16, var_37843_cast_fp16))[name = tensor<string, []>("op_38084_cast_fp16")];
            tensor<fp16, []> var_38085_to_fp16 = const()[name = tensor<string, []>("op_38085_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3225_cast_fp16 = mul(x = var_38084_cast_fp16, y = var_38085_to_fp16)[name = tensor<string, []>("aw_3225_cast_fp16")];
            tensor<string, []> var_38088_equation_0 = const()[name = tensor<string, []>("op_38088_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38088_cast_fp16 = einsum(equation = var_38088_equation_0, values = (var_37930_cast_fp16, var_37847_cast_fp16))[name = tensor<string, []>("op_38088_cast_fp16")];
            tensor<fp16, []> var_38089_to_fp16 = const()[name = tensor<string, []>("op_38089_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3227_cast_fp16 = mul(x = var_38088_cast_fp16, y = var_38089_to_fp16)[name = tensor<string, []>("aw_3227_cast_fp16")];
            tensor<string, []> var_38092_equation_0 = const()[name = tensor<string, []>("op_38092_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38092_cast_fp16 = einsum(equation = var_38092_equation_0, values = (var_37934_cast_fp16, var_37851_cast_fp16))[name = tensor<string, []>("op_38092_cast_fp16")];
            tensor<fp16, []> var_38093_to_fp16 = const()[name = tensor<string, []>("op_38093_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3229_cast_fp16 = mul(x = var_38092_cast_fp16, y = var_38093_to_fp16)[name = tensor<string, []>("aw_3229_cast_fp16")];
            tensor<string, []> var_38096_equation_0 = const()[name = tensor<string, []>("op_38096_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38096_cast_fp16 = einsum(equation = var_38096_equation_0, values = (var_37938_cast_fp16, var_37855_cast_fp16))[name = tensor<string, []>("op_38096_cast_fp16")];
            tensor<fp16, []> var_38097_to_fp16 = const()[name = tensor<string, []>("op_38097_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3231_cast_fp16 = mul(x = var_38096_cast_fp16, y = var_38097_to_fp16)[name = tensor<string, []>("aw_3231_cast_fp16")];
            tensor<string, []> var_38100_equation_0 = const()[name = tensor<string, []>("op_38100_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38100_cast_fp16 = einsum(equation = var_38100_equation_0, values = (var_37942_cast_fp16, var_37859_cast_fp16))[name = tensor<string, []>("op_38100_cast_fp16")];
            tensor<fp16, []> var_38101_to_fp16 = const()[name = tensor<string, []>("op_38101_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3233_cast_fp16 = mul(x = var_38100_cast_fp16, y = var_38101_to_fp16)[name = tensor<string, []>("aw_3233_cast_fp16")];
            tensor<string, []> var_38104_equation_0 = const()[name = tensor<string, []>("op_38104_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38104_cast_fp16 = einsum(equation = var_38104_equation_0, values = (var_37946_cast_fp16, var_37863_cast_fp16))[name = tensor<string, []>("op_38104_cast_fp16")];
            tensor<fp16, []> var_38105_to_fp16 = const()[name = tensor<string, []>("op_38105_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3235_cast_fp16 = mul(x = var_38104_cast_fp16, y = var_38105_to_fp16)[name = tensor<string, []>("aw_3235_cast_fp16")];
            tensor<string, []> var_38108_equation_0 = const()[name = tensor<string, []>("op_38108_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38108_cast_fp16 = einsum(equation = var_38108_equation_0, values = (var_37950_cast_fp16, var_37867_cast_fp16))[name = tensor<string, []>("op_38108_cast_fp16")];
            tensor<fp16, []> var_38109_to_fp16 = const()[name = tensor<string, []>("op_38109_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3237_cast_fp16 = mul(x = var_38108_cast_fp16, y = var_38109_to_fp16)[name = tensor<string, []>("aw_3237_cast_fp16")];
            tensor<string, []> var_38112_equation_0 = const()[name = tensor<string, []>("op_38112_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38112_cast_fp16 = einsum(equation = var_38112_equation_0, values = (var_37954_cast_fp16, var_37871_cast_fp16))[name = tensor<string, []>("op_38112_cast_fp16")];
            tensor<fp16, []> var_38113_to_fp16 = const()[name = tensor<string, []>("op_38113_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3239_cast_fp16 = mul(x = var_38112_cast_fp16, y = var_38113_to_fp16)[name = tensor<string, []>("aw_3239_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38115_cast_fp16 = softmax(axis = var_30355, x = aw_3201_cast_fp16)[name = tensor<string, []>("op_38115_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38116_cast_fp16 = softmax(axis = var_30355, x = aw_3203_cast_fp16)[name = tensor<string, []>("op_38116_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38117_cast_fp16 = softmax(axis = var_30355, x = aw_3205_cast_fp16)[name = tensor<string, []>("op_38117_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38118_cast_fp16 = softmax(axis = var_30355, x = aw_3207_cast_fp16)[name = tensor<string, []>("op_38118_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38119_cast_fp16 = softmax(axis = var_30355, x = aw_3209_cast_fp16)[name = tensor<string, []>("op_38119_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38120_cast_fp16 = softmax(axis = var_30355, x = aw_3211_cast_fp16)[name = tensor<string, []>("op_38120_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38121_cast_fp16 = softmax(axis = var_30355, x = aw_3213_cast_fp16)[name = tensor<string, []>("op_38121_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38122_cast_fp16 = softmax(axis = var_30355, x = aw_3215_cast_fp16)[name = tensor<string, []>("op_38122_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38123_cast_fp16 = softmax(axis = var_30355, x = aw_3217_cast_fp16)[name = tensor<string, []>("op_38123_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38124_cast_fp16 = softmax(axis = var_30355, x = aw_3219_cast_fp16)[name = tensor<string, []>("op_38124_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38125_cast_fp16 = softmax(axis = var_30355, x = aw_3221_cast_fp16)[name = tensor<string, []>("op_38125_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38126_cast_fp16 = softmax(axis = var_30355, x = aw_3223_cast_fp16)[name = tensor<string, []>("op_38126_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38127_cast_fp16 = softmax(axis = var_30355, x = aw_3225_cast_fp16)[name = tensor<string, []>("op_38127_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38128_cast_fp16 = softmax(axis = var_30355, x = aw_3227_cast_fp16)[name = tensor<string, []>("op_38128_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38129_cast_fp16 = softmax(axis = var_30355, x = aw_3229_cast_fp16)[name = tensor<string, []>("op_38129_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38130_cast_fp16 = softmax(axis = var_30355, x = aw_3231_cast_fp16)[name = tensor<string, []>("op_38130_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38131_cast_fp16 = softmax(axis = var_30355, x = aw_3233_cast_fp16)[name = tensor<string, []>("op_38131_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38132_cast_fp16 = softmax(axis = var_30355, x = aw_3235_cast_fp16)[name = tensor<string, []>("op_38132_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38133_cast_fp16 = softmax(axis = var_30355, x = aw_3237_cast_fp16)[name = tensor<string, []>("op_38133_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38134_cast_fp16 = softmax(axis = var_30355, x = aw_3239_cast_fp16)[name = tensor<string, []>("op_38134_cast_fp16")];
            tensor<string, []> var_38136_equation_0 = const()[name = tensor<string, []>("op_38136_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38136_cast_fp16 = einsum(equation = var_38136_equation_0, values = (var_37956_cast_fp16, var_38115_cast_fp16))[name = tensor<string, []>("op_38136_cast_fp16")];
            tensor<string, []> var_38138_equation_0 = const()[name = tensor<string, []>("op_38138_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38138_cast_fp16 = einsum(equation = var_38138_equation_0, values = (var_37960_cast_fp16, var_38116_cast_fp16))[name = tensor<string, []>("op_38138_cast_fp16")];
            tensor<string, []> var_38140_equation_0 = const()[name = tensor<string, []>("op_38140_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38140_cast_fp16 = einsum(equation = var_38140_equation_0, values = (var_37964_cast_fp16, var_38117_cast_fp16))[name = tensor<string, []>("op_38140_cast_fp16")];
            tensor<string, []> var_38142_equation_0 = const()[name = tensor<string, []>("op_38142_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38142_cast_fp16 = einsum(equation = var_38142_equation_0, values = (var_37968_cast_fp16, var_38118_cast_fp16))[name = tensor<string, []>("op_38142_cast_fp16")];
            tensor<string, []> var_38144_equation_0 = const()[name = tensor<string, []>("op_38144_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38144_cast_fp16 = einsum(equation = var_38144_equation_0, values = (var_37972_cast_fp16, var_38119_cast_fp16))[name = tensor<string, []>("op_38144_cast_fp16")];
            tensor<string, []> var_38146_equation_0 = const()[name = tensor<string, []>("op_38146_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38146_cast_fp16 = einsum(equation = var_38146_equation_0, values = (var_37976_cast_fp16, var_38120_cast_fp16))[name = tensor<string, []>("op_38146_cast_fp16")];
            tensor<string, []> var_38148_equation_0 = const()[name = tensor<string, []>("op_38148_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38148_cast_fp16 = einsum(equation = var_38148_equation_0, values = (var_37980_cast_fp16, var_38121_cast_fp16))[name = tensor<string, []>("op_38148_cast_fp16")];
            tensor<string, []> var_38150_equation_0 = const()[name = tensor<string, []>("op_38150_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38150_cast_fp16 = einsum(equation = var_38150_equation_0, values = (var_37984_cast_fp16, var_38122_cast_fp16))[name = tensor<string, []>("op_38150_cast_fp16")];
            tensor<string, []> var_38152_equation_0 = const()[name = tensor<string, []>("op_38152_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38152_cast_fp16 = einsum(equation = var_38152_equation_0, values = (var_37988_cast_fp16, var_38123_cast_fp16))[name = tensor<string, []>("op_38152_cast_fp16")];
            tensor<string, []> var_38154_equation_0 = const()[name = tensor<string, []>("op_38154_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38154_cast_fp16 = einsum(equation = var_38154_equation_0, values = (var_37992_cast_fp16, var_38124_cast_fp16))[name = tensor<string, []>("op_38154_cast_fp16")];
            tensor<string, []> var_38156_equation_0 = const()[name = tensor<string, []>("op_38156_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38156_cast_fp16 = einsum(equation = var_38156_equation_0, values = (var_37996_cast_fp16, var_38125_cast_fp16))[name = tensor<string, []>("op_38156_cast_fp16")];
            tensor<string, []> var_38158_equation_0 = const()[name = tensor<string, []>("op_38158_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38158_cast_fp16 = einsum(equation = var_38158_equation_0, values = (var_38000_cast_fp16, var_38126_cast_fp16))[name = tensor<string, []>("op_38158_cast_fp16")];
            tensor<string, []> var_38160_equation_0 = const()[name = tensor<string, []>("op_38160_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38160_cast_fp16 = einsum(equation = var_38160_equation_0, values = (var_38004_cast_fp16, var_38127_cast_fp16))[name = tensor<string, []>("op_38160_cast_fp16")];
            tensor<string, []> var_38162_equation_0 = const()[name = tensor<string, []>("op_38162_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38162_cast_fp16 = einsum(equation = var_38162_equation_0, values = (var_38008_cast_fp16, var_38128_cast_fp16))[name = tensor<string, []>("op_38162_cast_fp16")];
            tensor<string, []> var_38164_equation_0 = const()[name = tensor<string, []>("op_38164_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38164_cast_fp16 = einsum(equation = var_38164_equation_0, values = (var_38012_cast_fp16, var_38129_cast_fp16))[name = tensor<string, []>("op_38164_cast_fp16")];
            tensor<string, []> var_38166_equation_0 = const()[name = tensor<string, []>("op_38166_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38166_cast_fp16 = einsum(equation = var_38166_equation_0, values = (var_38016_cast_fp16, var_38130_cast_fp16))[name = tensor<string, []>("op_38166_cast_fp16")];
            tensor<string, []> var_38168_equation_0 = const()[name = tensor<string, []>("op_38168_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38168_cast_fp16 = einsum(equation = var_38168_equation_0, values = (var_38020_cast_fp16, var_38131_cast_fp16))[name = tensor<string, []>("op_38168_cast_fp16")];
            tensor<string, []> var_38170_equation_0 = const()[name = tensor<string, []>("op_38170_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38170_cast_fp16 = einsum(equation = var_38170_equation_0, values = (var_38024_cast_fp16, var_38132_cast_fp16))[name = tensor<string, []>("op_38170_cast_fp16")];
            tensor<string, []> var_38172_equation_0 = const()[name = tensor<string, []>("op_38172_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38172_cast_fp16 = einsum(equation = var_38172_equation_0, values = (var_38028_cast_fp16, var_38133_cast_fp16))[name = tensor<string, []>("op_38172_cast_fp16")];
            tensor<string, []> var_38174_equation_0 = const()[name = tensor<string, []>("op_38174_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38174_cast_fp16 = einsum(equation = var_38174_equation_0, values = (var_38032_cast_fp16, var_38134_cast_fp16))[name = tensor<string, []>("op_38174_cast_fp16")];
            tensor<bool, []> input_503_interleave_0 = const()[name = tensor<string, []>("input_503_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_503_cast_fp16 = concat(axis = var_30355, interleave = input_503_interleave_0, values = (var_38136_cast_fp16, var_38138_cast_fp16, var_38140_cast_fp16, var_38142_cast_fp16, var_38144_cast_fp16, var_38146_cast_fp16, var_38148_cast_fp16, var_38150_cast_fp16, var_38152_cast_fp16, var_38154_cast_fp16, var_38156_cast_fp16, var_38158_cast_fp16, var_38160_cast_fp16, var_38162_cast_fp16, var_38164_cast_fp16, var_38166_cast_fp16, var_38168_cast_fp16, var_38170_cast_fp16, var_38172_cast_fp16, var_38174_cast_fp16))[name = tensor<string, []>("input_503_cast_fp16")];
            tensor<string, []> var_38184_pad_type_0 = const()[name = tensor<string, []>("op_38184_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_38184_strides_0 = const()[name = tensor<string, []>("op_38184_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_38184_pad_0 = const()[name = tensor<string, []>("op_38184_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_38184_dilations_0 = const()[name = tensor<string, []>("op_38184_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_38184_groups_0 = const()[name = tensor<string, []>("op_38184_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_8_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(216187840))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(217416704))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_8_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_8_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_8_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(217416896)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_38184_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_8_attn1_to_out_0_bias_to_fp16, dilations = var_38184_dilations_0, groups = var_38184_groups_0, pad = var_38184_pad_0, pad_type = var_38184_pad_type_0, strides = var_38184_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_8_attn1_to_out_0_weight_to_fp16_palettized, x = input_503_cast_fp16)[name = tensor<string, []>("op_38184_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_255_cast_fp16 = add(x = var_38184_cast_fp16, y = inputs_253_cast_fp16)[name = tensor<string, []>("inputs_255_cast_fp16")];
            tensor<int32, [1]> hidden_states_337_axes_0 = const()[name = tensor<string, []>("hidden_states_337_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_337_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_337_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(217419520)))];
            tensor<fp16, [1280]> hidden_states_337_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_337_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(217422144)))];
            tensor<fp16, []> var_38194_to_fp16 = const()[name = tensor<string, []>("op_38194_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_337_cast_fp16 = layer_norm(axes = hidden_states_337_axes_0, beta = hidden_states_337_beta_0_to_fp16, epsilon = var_38194_to_fp16, gamma = hidden_states_337_gamma_0_to_fp16, x = inputs_255_cast_fp16)[name = tensor<string, []>("hidden_states_337_cast_fp16")];
            tensor<string, []> q_171_pad_type_0 = const()[name = tensor<string, []>("q_171_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_171_strides_0 = const()[name = tensor<string, []>("q_171_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_171_pad_0 = const()[name = tensor<string, []>("q_171_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_171_dilations_0 = const()[name = tensor<string, []>("q_171_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_171_groups_0 = const()[name = tensor<string, []>("q_171_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_8_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(217424768))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(218653632))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_8_attn2_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_171_cast_fp16 = conv(dilations = q_171_dilations_0, groups = q_171_groups_0, pad = q_171_pad_0, pad_type = q_171_pad_type_0, strides = q_171_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_8_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_337_cast_fp16)[name = tensor<string, []>("q_171_cast_fp16")];
            tensor<string, []> k_341_pad_type_0 = const()[name = tensor<string, []>("k_341_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_341_strides_0 = const()[name = tensor<string, []>("k_341_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_341_pad_0 = const()[name = tensor<string, []>("k_341_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_341_dilations_0 = const()[name = tensor<string, []>("k_341_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_341_groups_0 = const()[name = tensor<string, []>("k_341_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_8_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(218653824))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(220619968))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_8_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> k_341_cast_fp16 = conv(dilations = k_341_dilations_0, groups = k_341_groups_0, pad = k_341_pad_0, pad_type = k_341_pad_type_0, strides = k_341_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_8_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("k_341_cast_fp16")];
            tensor<string, []> v_171_pad_type_0 = const()[name = tensor<string, []>("v_171_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_171_strides_0 = const()[name = tensor<string, []>("v_171_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_171_pad_0 = const()[name = tensor<string, []>("v_171_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_171_dilations_0 = const()[name = tensor<string, []>("v_171_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_171_groups_0 = const()[name = tensor<string, []>("v_171_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_8_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(220620160))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(222586304))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_8_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> v_171_cast_fp16 = conv(dilations = v_171_dilations_0, groups = v_171_groups_0, pad = v_171_pad_0, pad_type = v_171_pad_type_0, strides = v_171_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_8_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("v_171_cast_fp16")];
            tensor<int32, [4]> var_38227_begin_0 = const()[name = tensor<string, []>("op_38227_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_38227_end_0 = const()[name = tensor<string, []>("op_38227_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_38227_end_mask_0 = const()[name = tensor<string, []>("op_38227_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38227_cast_fp16 = slice_by_index(begin = var_38227_begin_0, end = var_38227_end_0, end_mask = var_38227_end_mask_0, x = q_171_cast_fp16)[name = tensor<string, []>("op_38227_cast_fp16")];
            tensor<int32, [4]> var_38231_begin_0 = const()[name = tensor<string, []>("op_38231_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_38231_end_0 = const()[name = tensor<string, []>("op_38231_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_38231_end_mask_0 = const()[name = tensor<string, []>("op_38231_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38231_cast_fp16 = slice_by_index(begin = var_38231_begin_0, end = var_38231_end_0, end_mask = var_38231_end_mask_0, x = q_171_cast_fp16)[name = tensor<string, []>("op_38231_cast_fp16")];
            tensor<int32, [4]> var_38235_begin_0 = const()[name = tensor<string, []>("op_38235_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_38235_end_0 = const()[name = tensor<string, []>("op_38235_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_38235_end_mask_0 = const()[name = tensor<string, []>("op_38235_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38235_cast_fp16 = slice_by_index(begin = var_38235_begin_0, end = var_38235_end_0, end_mask = var_38235_end_mask_0, x = q_171_cast_fp16)[name = tensor<string, []>("op_38235_cast_fp16")];
            tensor<int32, [4]> var_38239_begin_0 = const()[name = tensor<string, []>("op_38239_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_38239_end_0 = const()[name = tensor<string, []>("op_38239_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_38239_end_mask_0 = const()[name = tensor<string, []>("op_38239_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38239_cast_fp16 = slice_by_index(begin = var_38239_begin_0, end = var_38239_end_0, end_mask = var_38239_end_mask_0, x = q_171_cast_fp16)[name = tensor<string, []>("op_38239_cast_fp16")];
            tensor<int32, [4]> var_38243_begin_0 = const()[name = tensor<string, []>("op_38243_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_38243_end_0 = const()[name = tensor<string, []>("op_38243_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_38243_end_mask_0 = const()[name = tensor<string, []>("op_38243_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38243_cast_fp16 = slice_by_index(begin = var_38243_begin_0, end = var_38243_end_0, end_mask = var_38243_end_mask_0, x = q_171_cast_fp16)[name = tensor<string, []>("op_38243_cast_fp16")];
            tensor<int32, [4]> var_38247_begin_0 = const()[name = tensor<string, []>("op_38247_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_38247_end_0 = const()[name = tensor<string, []>("op_38247_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_38247_end_mask_0 = const()[name = tensor<string, []>("op_38247_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38247_cast_fp16 = slice_by_index(begin = var_38247_begin_0, end = var_38247_end_0, end_mask = var_38247_end_mask_0, x = q_171_cast_fp16)[name = tensor<string, []>("op_38247_cast_fp16")];
            tensor<int32, [4]> var_38251_begin_0 = const()[name = tensor<string, []>("op_38251_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_38251_end_0 = const()[name = tensor<string, []>("op_38251_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_38251_end_mask_0 = const()[name = tensor<string, []>("op_38251_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38251_cast_fp16 = slice_by_index(begin = var_38251_begin_0, end = var_38251_end_0, end_mask = var_38251_end_mask_0, x = q_171_cast_fp16)[name = tensor<string, []>("op_38251_cast_fp16")];
            tensor<int32, [4]> var_38255_begin_0 = const()[name = tensor<string, []>("op_38255_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_38255_end_0 = const()[name = tensor<string, []>("op_38255_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_38255_end_mask_0 = const()[name = tensor<string, []>("op_38255_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38255_cast_fp16 = slice_by_index(begin = var_38255_begin_0, end = var_38255_end_0, end_mask = var_38255_end_mask_0, x = q_171_cast_fp16)[name = tensor<string, []>("op_38255_cast_fp16")];
            tensor<int32, [4]> var_38259_begin_0 = const()[name = tensor<string, []>("op_38259_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_38259_end_0 = const()[name = tensor<string, []>("op_38259_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_38259_end_mask_0 = const()[name = tensor<string, []>("op_38259_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38259_cast_fp16 = slice_by_index(begin = var_38259_begin_0, end = var_38259_end_0, end_mask = var_38259_end_mask_0, x = q_171_cast_fp16)[name = tensor<string, []>("op_38259_cast_fp16")];
            tensor<int32, [4]> var_38263_begin_0 = const()[name = tensor<string, []>("op_38263_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_38263_end_0 = const()[name = tensor<string, []>("op_38263_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_38263_end_mask_0 = const()[name = tensor<string, []>("op_38263_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38263_cast_fp16 = slice_by_index(begin = var_38263_begin_0, end = var_38263_end_0, end_mask = var_38263_end_mask_0, x = q_171_cast_fp16)[name = tensor<string, []>("op_38263_cast_fp16")];
            tensor<int32, [4]> var_38267_begin_0 = const()[name = tensor<string, []>("op_38267_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_38267_end_0 = const()[name = tensor<string, []>("op_38267_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_38267_end_mask_0 = const()[name = tensor<string, []>("op_38267_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38267_cast_fp16 = slice_by_index(begin = var_38267_begin_0, end = var_38267_end_0, end_mask = var_38267_end_mask_0, x = q_171_cast_fp16)[name = tensor<string, []>("op_38267_cast_fp16")];
            tensor<int32, [4]> var_38271_begin_0 = const()[name = tensor<string, []>("op_38271_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_38271_end_0 = const()[name = tensor<string, []>("op_38271_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_38271_end_mask_0 = const()[name = tensor<string, []>("op_38271_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38271_cast_fp16 = slice_by_index(begin = var_38271_begin_0, end = var_38271_end_0, end_mask = var_38271_end_mask_0, x = q_171_cast_fp16)[name = tensor<string, []>("op_38271_cast_fp16")];
            tensor<int32, [4]> var_38275_begin_0 = const()[name = tensor<string, []>("op_38275_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_38275_end_0 = const()[name = tensor<string, []>("op_38275_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_38275_end_mask_0 = const()[name = tensor<string, []>("op_38275_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38275_cast_fp16 = slice_by_index(begin = var_38275_begin_0, end = var_38275_end_0, end_mask = var_38275_end_mask_0, x = q_171_cast_fp16)[name = tensor<string, []>("op_38275_cast_fp16")];
            tensor<int32, [4]> var_38279_begin_0 = const()[name = tensor<string, []>("op_38279_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_38279_end_0 = const()[name = tensor<string, []>("op_38279_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_38279_end_mask_0 = const()[name = tensor<string, []>("op_38279_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38279_cast_fp16 = slice_by_index(begin = var_38279_begin_0, end = var_38279_end_0, end_mask = var_38279_end_mask_0, x = q_171_cast_fp16)[name = tensor<string, []>("op_38279_cast_fp16")];
            tensor<int32, [4]> var_38283_begin_0 = const()[name = tensor<string, []>("op_38283_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_38283_end_0 = const()[name = tensor<string, []>("op_38283_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_38283_end_mask_0 = const()[name = tensor<string, []>("op_38283_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38283_cast_fp16 = slice_by_index(begin = var_38283_begin_0, end = var_38283_end_0, end_mask = var_38283_end_mask_0, x = q_171_cast_fp16)[name = tensor<string, []>("op_38283_cast_fp16")];
            tensor<int32, [4]> var_38287_begin_0 = const()[name = tensor<string, []>("op_38287_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_38287_end_0 = const()[name = tensor<string, []>("op_38287_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_38287_end_mask_0 = const()[name = tensor<string, []>("op_38287_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38287_cast_fp16 = slice_by_index(begin = var_38287_begin_0, end = var_38287_end_0, end_mask = var_38287_end_mask_0, x = q_171_cast_fp16)[name = tensor<string, []>("op_38287_cast_fp16")];
            tensor<int32, [4]> var_38291_begin_0 = const()[name = tensor<string, []>("op_38291_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_38291_end_0 = const()[name = tensor<string, []>("op_38291_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_38291_end_mask_0 = const()[name = tensor<string, []>("op_38291_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38291_cast_fp16 = slice_by_index(begin = var_38291_begin_0, end = var_38291_end_0, end_mask = var_38291_end_mask_0, x = q_171_cast_fp16)[name = tensor<string, []>("op_38291_cast_fp16")];
            tensor<int32, [4]> var_38295_begin_0 = const()[name = tensor<string, []>("op_38295_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_38295_end_0 = const()[name = tensor<string, []>("op_38295_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_38295_end_mask_0 = const()[name = tensor<string, []>("op_38295_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38295_cast_fp16 = slice_by_index(begin = var_38295_begin_0, end = var_38295_end_0, end_mask = var_38295_end_mask_0, x = q_171_cast_fp16)[name = tensor<string, []>("op_38295_cast_fp16")];
            tensor<int32, [4]> var_38299_begin_0 = const()[name = tensor<string, []>("op_38299_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_38299_end_0 = const()[name = tensor<string, []>("op_38299_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_38299_end_mask_0 = const()[name = tensor<string, []>("op_38299_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38299_cast_fp16 = slice_by_index(begin = var_38299_begin_0, end = var_38299_end_0, end_mask = var_38299_end_mask_0, x = q_171_cast_fp16)[name = tensor<string, []>("op_38299_cast_fp16")];
            tensor<int32, [4]> var_38303_begin_0 = const()[name = tensor<string, []>("op_38303_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_38303_end_0 = const()[name = tensor<string, []>("op_38303_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_38303_end_mask_0 = const()[name = tensor<string, []>("op_38303_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38303_cast_fp16 = slice_by_index(begin = var_38303_begin_0, end = var_38303_end_0, end_mask = var_38303_end_mask_0, x = q_171_cast_fp16)[name = tensor<string, []>("op_38303_cast_fp16")];
            tensor<int32, [4]> k_343_perm_0 = const()[name = tensor<string, []>("k_343_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_38310_begin_0 = const()[name = tensor<string, []>("op_38310_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_38310_end_0 = const()[name = tensor<string, []>("op_38310_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_38310_end_mask_0 = const()[name = tensor<string, []>("op_38310_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> k_343_cast_fp16 = transpose(perm = k_343_perm_0, x = k_341_cast_fp16)[name = tensor<string, []>("transpose_54")];
            tensor<fp16, [2, 77, 1, 64]> var_38310_cast_fp16 = slice_by_index(begin = var_38310_begin_0, end = var_38310_end_0, end_mask = var_38310_end_mask_0, x = k_343_cast_fp16)[name = tensor<string, []>("op_38310_cast_fp16")];
            tensor<int32, [4]> var_38314_begin_0 = const()[name = tensor<string, []>("op_38314_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_38314_end_0 = const()[name = tensor<string, []>("op_38314_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_38314_end_mask_0 = const()[name = tensor<string, []>("op_38314_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_38314_cast_fp16 = slice_by_index(begin = var_38314_begin_0, end = var_38314_end_0, end_mask = var_38314_end_mask_0, x = k_343_cast_fp16)[name = tensor<string, []>("op_38314_cast_fp16")];
            tensor<int32, [4]> var_38318_begin_0 = const()[name = tensor<string, []>("op_38318_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_38318_end_0 = const()[name = tensor<string, []>("op_38318_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_38318_end_mask_0 = const()[name = tensor<string, []>("op_38318_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_38318_cast_fp16 = slice_by_index(begin = var_38318_begin_0, end = var_38318_end_0, end_mask = var_38318_end_mask_0, x = k_343_cast_fp16)[name = tensor<string, []>("op_38318_cast_fp16")];
            tensor<int32, [4]> var_38322_begin_0 = const()[name = tensor<string, []>("op_38322_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_38322_end_0 = const()[name = tensor<string, []>("op_38322_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_38322_end_mask_0 = const()[name = tensor<string, []>("op_38322_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_38322_cast_fp16 = slice_by_index(begin = var_38322_begin_0, end = var_38322_end_0, end_mask = var_38322_end_mask_0, x = k_343_cast_fp16)[name = tensor<string, []>("op_38322_cast_fp16")];
            tensor<int32, [4]> var_38326_begin_0 = const()[name = tensor<string, []>("op_38326_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_38326_end_0 = const()[name = tensor<string, []>("op_38326_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_38326_end_mask_0 = const()[name = tensor<string, []>("op_38326_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_38326_cast_fp16 = slice_by_index(begin = var_38326_begin_0, end = var_38326_end_0, end_mask = var_38326_end_mask_0, x = k_343_cast_fp16)[name = tensor<string, []>("op_38326_cast_fp16")];
            tensor<int32, [4]> var_38330_begin_0 = const()[name = tensor<string, []>("op_38330_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_38330_end_0 = const()[name = tensor<string, []>("op_38330_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_38330_end_mask_0 = const()[name = tensor<string, []>("op_38330_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_38330_cast_fp16 = slice_by_index(begin = var_38330_begin_0, end = var_38330_end_0, end_mask = var_38330_end_mask_0, x = k_343_cast_fp16)[name = tensor<string, []>("op_38330_cast_fp16")];
            tensor<int32, [4]> var_38334_begin_0 = const()[name = tensor<string, []>("op_38334_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_38334_end_0 = const()[name = tensor<string, []>("op_38334_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_38334_end_mask_0 = const()[name = tensor<string, []>("op_38334_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_38334_cast_fp16 = slice_by_index(begin = var_38334_begin_0, end = var_38334_end_0, end_mask = var_38334_end_mask_0, x = k_343_cast_fp16)[name = tensor<string, []>("op_38334_cast_fp16")];
            tensor<int32, [4]> var_38338_begin_0 = const()[name = tensor<string, []>("op_38338_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_38338_end_0 = const()[name = tensor<string, []>("op_38338_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_38338_end_mask_0 = const()[name = tensor<string, []>("op_38338_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_38338_cast_fp16 = slice_by_index(begin = var_38338_begin_0, end = var_38338_end_0, end_mask = var_38338_end_mask_0, x = k_343_cast_fp16)[name = tensor<string, []>("op_38338_cast_fp16")];
            tensor<int32, [4]> var_38342_begin_0 = const()[name = tensor<string, []>("op_38342_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_38342_end_0 = const()[name = tensor<string, []>("op_38342_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_38342_end_mask_0 = const()[name = tensor<string, []>("op_38342_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_38342_cast_fp16 = slice_by_index(begin = var_38342_begin_0, end = var_38342_end_0, end_mask = var_38342_end_mask_0, x = k_343_cast_fp16)[name = tensor<string, []>("op_38342_cast_fp16")];
            tensor<int32, [4]> var_38346_begin_0 = const()[name = tensor<string, []>("op_38346_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_38346_end_0 = const()[name = tensor<string, []>("op_38346_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_38346_end_mask_0 = const()[name = tensor<string, []>("op_38346_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_38346_cast_fp16 = slice_by_index(begin = var_38346_begin_0, end = var_38346_end_0, end_mask = var_38346_end_mask_0, x = k_343_cast_fp16)[name = tensor<string, []>("op_38346_cast_fp16")];
            tensor<int32, [4]> var_38350_begin_0 = const()[name = tensor<string, []>("op_38350_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_38350_end_0 = const()[name = tensor<string, []>("op_38350_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_38350_end_mask_0 = const()[name = tensor<string, []>("op_38350_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_38350_cast_fp16 = slice_by_index(begin = var_38350_begin_0, end = var_38350_end_0, end_mask = var_38350_end_mask_0, x = k_343_cast_fp16)[name = tensor<string, []>("op_38350_cast_fp16")];
            tensor<int32, [4]> var_38354_begin_0 = const()[name = tensor<string, []>("op_38354_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_38354_end_0 = const()[name = tensor<string, []>("op_38354_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_38354_end_mask_0 = const()[name = tensor<string, []>("op_38354_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_38354_cast_fp16 = slice_by_index(begin = var_38354_begin_0, end = var_38354_end_0, end_mask = var_38354_end_mask_0, x = k_343_cast_fp16)[name = tensor<string, []>("op_38354_cast_fp16")];
            tensor<int32, [4]> var_38358_begin_0 = const()[name = tensor<string, []>("op_38358_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_38358_end_0 = const()[name = tensor<string, []>("op_38358_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_38358_end_mask_0 = const()[name = tensor<string, []>("op_38358_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_38358_cast_fp16 = slice_by_index(begin = var_38358_begin_0, end = var_38358_end_0, end_mask = var_38358_end_mask_0, x = k_343_cast_fp16)[name = tensor<string, []>("op_38358_cast_fp16")];
            tensor<int32, [4]> var_38362_begin_0 = const()[name = tensor<string, []>("op_38362_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_38362_end_0 = const()[name = tensor<string, []>("op_38362_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_38362_end_mask_0 = const()[name = tensor<string, []>("op_38362_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_38362_cast_fp16 = slice_by_index(begin = var_38362_begin_0, end = var_38362_end_0, end_mask = var_38362_end_mask_0, x = k_343_cast_fp16)[name = tensor<string, []>("op_38362_cast_fp16")];
            tensor<int32, [4]> var_38366_begin_0 = const()[name = tensor<string, []>("op_38366_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_38366_end_0 = const()[name = tensor<string, []>("op_38366_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_38366_end_mask_0 = const()[name = tensor<string, []>("op_38366_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_38366_cast_fp16 = slice_by_index(begin = var_38366_begin_0, end = var_38366_end_0, end_mask = var_38366_end_mask_0, x = k_343_cast_fp16)[name = tensor<string, []>("op_38366_cast_fp16")];
            tensor<int32, [4]> var_38370_begin_0 = const()[name = tensor<string, []>("op_38370_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_38370_end_0 = const()[name = tensor<string, []>("op_38370_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_38370_end_mask_0 = const()[name = tensor<string, []>("op_38370_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_38370_cast_fp16 = slice_by_index(begin = var_38370_begin_0, end = var_38370_end_0, end_mask = var_38370_end_mask_0, x = k_343_cast_fp16)[name = tensor<string, []>("op_38370_cast_fp16")];
            tensor<int32, [4]> var_38374_begin_0 = const()[name = tensor<string, []>("op_38374_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_38374_end_0 = const()[name = tensor<string, []>("op_38374_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_38374_end_mask_0 = const()[name = tensor<string, []>("op_38374_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_38374_cast_fp16 = slice_by_index(begin = var_38374_begin_0, end = var_38374_end_0, end_mask = var_38374_end_mask_0, x = k_343_cast_fp16)[name = tensor<string, []>("op_38374_cast_fp16")];
            tensor<int32, [4]> var_38378_begin_0 = const()[name = tensor<string, []>("op_38378_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_38378_end_0 = const()[name = tensor<string, []>("op_38378_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_38378_end_mask_0 = const()[name = tensor<string, []>("op_38378_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_38378_cast_fp16 = slice_by_index(begin = var_38378_begin_0, end = var_38378_end_0, end_mask = var_38378_end_mask_0, x = k_343_cast_fp16)[name = tensor<string, []>("op_38378_cast_fp16")];
            tensor<int32, [4]> var_38382_begin_0 = const()[name = tensor<string, []>("op_38382_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_38382_end_0 = const()[name = tensor<string, []>("op_38382_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_38382_end_mask_0 = const()[name = tensor<string, []>("op_38382_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_38382_cast_fp16 = slice_by_index(begin = var_38382_begin_0, end = var_38382_end_0, end_mask = var_38382_end_mask_0, x = k_343_cast_fp16)[name = tensor<string, []>("op_38382_cast_fp16")];
            tensor<int32, [4]> var_38386_begin_0 = const()[name = tensor<string, []>("op_38386_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_38386_end_0 = const()[name = tensor<string, []>("op_38386_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_38386_end_mask_0 = const()[name = tensor<string, []>("op_38386_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_38386_cast_fp16 = slice_by_index(begin = var_38386_begin_0, end = var_38386_end_0, end_mask = var_38386_end_mask_0, x = k_343_cast_fp16)[name = tensor<string, []>("op_38386_cast_fp16")];
            tensor<int32, [4]> var_38388_begin_0 = const()[name = tensor<string, []>("op_38388_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_38388_end_0 = const()[name = tensor<string, []>("op_38388_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_38388_end_mask_0 = const()[name = tensor<string, []>("op_38388_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_38388_cast_fp16 = slice_by_index(begin = var_38388_begin_0, end = var_38388_end_0, end_mask = var_38388_end_mask_0, x = v_171_cast_fp16)[name = tensor<string, []>("op_38388_cast_fp16")];
            tensor<int32, [4]> var_38392_begin_0 = const()[name = tensor<string, []>("op_38392_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_38392_end_0 = const()[name = tensor<string, []>("op_38392_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_38392_end_mask_0 = const()[name = tensor<string, []>("op_38392_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_38392_cast_fp16 = slice_by_index(begin = var_38392_begin_0, end = var_38392_end_0, end_mask = var_38392_end_mask_0, x = v_171_cast_fp16)[name = tensor<string, []>("op_38392_cast_fp16")];
            tensor<int32, [4]> var_38396_begin_0 = const()[name = tensor<string, []>("op_38396_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_38396_end_0 = const()[name = tensor<string, []>("op_38396_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_38396_end_mask_0 = const()[name = tensor<string, []>("op_38396_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_38396_cast_fp16 = slice_by_index(begin = var_38396_begin_0, end = var_38396_end_0, end_mask = var_38396_end_mask_0, x = v_171_cast_fp16)[name = tensor<string, []>("op_38396_cast_fp16")];
            tensor<int32, [4]> var_38400_begin_0 = const()[name = tensor<string, []>("op_38400_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_38400_end_0 = const()[name = tensor<string, []>("op_38400_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_38400_end_mask_0 = const()[name = tensor<string, []>("op_38400_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_38400_cast_fp16 = slice_by_index(begin = var_38400_begin_0, end = var_38400_end_0, end_mask = var_38400_end_mask_0, x = v_171_cast_fp16)[name = tensor<string, []>("op_38400_cast_fp16")];
            tensor<int32, [4]> var_38404_begin_0 = const()[name = tensor<string, []>("op_38404_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_38404_end_0 = const()[name = tensor<string, []>("op_38404_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_38404_end_mask_0 = const()[name = tensor<string, []>("op_38404_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_38404_cast_fp16 = slice_by_index(begin = var_38404_begin_0, end = var_38404_end_0, end_mask = var_38404_end_mask_0, x = v_171_cast_fp16)[name = tensor<string, []>("op_38404_cast_fp16")];
            tensor<int32, [4]> var_38408_begin_0 = const()[name = tensor<string, []>("op_38408_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_38408_end_0 = const()[name = tensor<string, []>("op_38408_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_38408_end_mask_0 = const()[name = tensor<string, []>("op_38408_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_38408_cast_fp16 = slice_by_index(begin = var_38408_begin_0, end = var_38408_end_0, end_mask = var_38408_end_mask_0, x = v_171_cast_fp16)[name = tensor<string, []>("op_38408_cast_fp16")];
            tensor<int32, [4]> var_38412_begin_0 = const()[name = tensor<string, []>("op_38412_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_38412_end_0 = const()[name = tensor<string, []>("op_38412_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_38412_end_mask_0 = const()[name = tensor<string, []>("op_38412_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_38412_cast_fp16 = slice_by_index(begin = var_38412_begin_0, end = var_38412_end_0, end_mask = var_38412_end_mask_0, x = v_171_cast_fp16)[name = tensor<string, []>("op_38412_cast_fp16")];
            tensor<int32, [4]> var_38416_begin_0 = const()[name = tensor<string, []>("op_38416_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_38416_end_0 = const()[name = tensor<string, []>("op_38416_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_38416_end_mask_0 = const()[name = tensor<string, []>("op_38416_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_38416_cast_fp16 = slice_by_index(begin = var_38416_begin_0, end = var_38416_end_0, end_mask = var_38416_end_mask_0, x = v_171_cast_fp16)[name = tensor<string, []>("op_38416_cast_fp16")];
            tensor<int32, [4]> var_38420_begin_0 = const()[name = tensor<string, []>("op_38420_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_38420_end_0 = const()[name = tensor<string, []>("op_38420_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_38420_end_mask_0 = const()[name = tensor<string, []>("op_38420_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_38420_cast_fp16 = slice_by_index(begin = var_38420_begin_0, end = var_38420_end_0, end_mask = var_38420_end_mask_0, x = v_171_cast_fp16)[name = tensor<string, []>("op_38420_cast_fp16")];
            tensor<int32, [4]> var_38424_begin_0 = const()[name = tensor<string, []>("op_38424_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_38424_end_0 = const()[name = tensor<string, []>("op_38424_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_38424_end_mask_0 = const()[name = tensor<string, []>("op_38424_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_38424_cast_fp16 = slice_by_index(begin = var_38424_begin_0, end = var_38424_end_0, end_mask = var_38424_end_mask_0, x = v_171_cast_fp16)[name = tensor<string, []>("op_38424_cast_fp16")];
            tensor<int32, [4]> var_38428_begin_0 = const()[name = tensor<string, []>("op_38428_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_38428_end_0 = const()[name = tensor<string, []>("op_38428_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_38428_end_mask_0 = const()[name = tensor<string, []>("op_38428_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_38428_cast_fp16 = slice_by_index(begin = var_38428_begin_0, end = var_38428_end_0, end_mask = var_38428_end_mask_0, x = v_171_cast_fp16)[name = tensor<string, []>("op_38428_cast_fp16")];
            tensor<int32, [4]> var_38432_begin_0 = const()[name = tensor<string, []>("op_38432_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_38432_end_0 = const()[name = tensor<string, []>("op_38432_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_38432_end_mask_0 = const()[name = tensor<string, []>("op_38432_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_38432_cast_fp16 = slice_by_index(begin = var_38432_begin_0, end = var_38432_end_0, end_mask = var_38432_end_mask_0, x = v_171_cast_fp16)[name = tensor<string, []>("op_38432_cast_fp16")];
            tensor<int32, [4]> var_38436_begin_0 = const()[name = tensor<string, []>("op_38436_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_38436_end_0 = const()[name = tensor<string, []>("op_38436_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_38436_end_mask_0 = const()[name = tensor<string, []>("op_38436_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_38436_cast_fp16 = slice_by_index(begin = var_38436_begin_0, end = var_38436_end_0, end_mask = var_38436_end_mask_0, x = v_171_cast_fp16)[name = tensor<string, []>("op_38436_cast_fp16")];
            tensor<int32, [4]> var_38440_begin_0 = const()[name = tensor<string, []>("op_38440_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_38440_end_0 = const()[name = tensor<string, []>("op_38440_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_38440_end_mask_0 = const()[name = tensor<string, []>("op_38440_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_38440_cast_fp16 = slice_by_index(begin = var_38440_begin_0, end = var_38440_end_0, end_mask = var_38440_end_mask_0, x = v_171_cast_fp16)[name = tensor<string, []>("op_38440_cast_fp16")];
            tensor<int32, [4]> var_38444_begin_0 = const()[name = tensor<string, []>("op_38444_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_38444_end_0 = const()[name = tensor<string, []>("op_38444_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_38444_end_mask_0 = const()[name = tensor<string, []>("op_38444_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_38444_cast_fp16 = slice_by_index(begin = var_38444_begin_0, end = var_38444_end_0, end_mask = var_38444_end_mask_0, x = v_171_cast_fp16)[name = tensor<string, []>("op_38444_cast_fp16")];
            tensor<int32, [4]> var_38448_begin_0 = const()[name = tensor<string, []>("op_38448_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_38448_end_0 = const()[name = tensor<string, []>("op_38448_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_38448_end_mask_0 = const()[name = tensor<string, []>("op_38448_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_38448_cast_fp16 = slice_by_index(begin = var_38448_begin_0, end = var_38448_end_0, end_mask = var_38448_end_mask_0, x = v_171_cast_fp16)[name = tensor<string, []>("op_38448_cast_fp16")];
            tensor<int32, [4]> var_38452_begin_0 = const()[name = tensor<string, []>("op_38452_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_38452_end_0 = const()[name = tensor<string, []>("op_38452_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_38452_end_mask_0 = const()[name = tensor<string, []>("op_38452_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_38452_cast_fp16 = slice_by_index(begin = var_38452_begin_0, end = var_38452_end_0, end_mask = var_38452_end_mask_0, x = v_171_cast_fp16)[name = tensor<string, []>("op_38452_cast_fp16")];
            tensor<int32, [4]> var_38456_begin_0 = const()[name = tensor<string, []>("op_38456_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_38456_end_0 = const()[name = tensor<string, []>("op_38456_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_38456_end_mask_0 = const()[name = tensor<string, []>("op_38456_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_38456_cast_fp16 = slice_by_index(begin = var_38456_begin_0, end = var_38456_end_0, end_mask = var_38456_end_mask_0, x = v_171_cast_fp16)[name = tensor<string, []>("op_38456_cast_fp16")];
            tensor<int32, [4]> var_38460_begin_0 = const()[name = tensor<string, []>("op_38460_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_38460_end_0 = const()[name = tensor<string, []>("op_38460_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_38460_end_mask_0 = const()[name = tensor<string, []>("op_38460_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_38460_cast_fp16 = slice_by_index(begin = var_38460_begin_0, end = var_38460_end_0, end_mask = var_38460_end_mask_0, x = v_171_cast_fp16)[name = tensor<string, []>("op_38460_cast_fp16")];
            tensor<int32, [4]> var_38464_begin_0 = const()[name = tensor<string, []>("op_38464_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_38464_end_0 = const()[name = tensor<string, []>("op_38464_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_38464_end_mask_0 = const()[name = tensor<string, []>("op_38464_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_38464_cast_fp16 = slice_by_index(begin = var_38464_begin_0, end = var_38464_end_0, end_mask = var_38464_end_mask_0, x = v_171_cast_fp16)[name = tensor<string, []>("op_38464_cast_fp16")];
            tensor<string, []> var_38468_equation_0 = const()[name = tensor<string, []>("op_38468_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_38468_cast_fp16 = einsum(equation = var_38468_equation_0, values = (var_38310_cast_fp16, var_38227_cast_fp16))[name = tensor<string, []>("op_38468_cast_fp16")];
            tensor<fp16, []> var_38469_to_fp16 = const()[name = tensor<string, []>("op_38469_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3241_cast_fp16 = mul(x = var_38468_cast_fp16, y = var_38469_to_fp16)[name = tensor<string, []>("aw_3241_cast_fp16")];
            tensor<string, []> var_38472_equation_0 = const()[name = tensor<string, []>("op_38472_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_38472_cast_fp16 = einsum(equation = var_38472_equation_0, values = (var_38314_cast_fp16, var_38231_cast_fp16))[name = tensor<string, []>("op_38472_cast_fp16")];
            tensor<fp16, []> var_38473_to_fp16 = const()[name = tensor<string, []>("op_38473_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3243_cast_fp16 = mul(x = var_38472_cast_fp16, y = var_38473_to_fp16)[name = tensor<string, []>("aw_3243_cast_fp16")];
            tensor<string, []> var_38476_equation_0 = const()[name = tensor<string, []>("op_38476_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_38476_cast_fp16 = einsum(equation = var_38476_equation_0, values = (var_38318_cast_fp16, var_38235_cast_fp16))[name = tensor<string, []>("op_38476_cast_fp16")];
            tensor<fp16, []> var_38477_to_fp16 = const()[name = tensor<string, []>("op_38477_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3245_cast_fp16 = mul(x = var_38476_cast_fp16, y = var_38477_to_fp16)[name = tensor<string, []>("aw_3245_cast_fp16")];
            tensor<string, []> var_38480_equation_0 = const()[name = tensor<string, []>("op_38480_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_38480_cast_fp16 = einsum(equation = var_38480_equation_0, values = (var_38322_cast_fp16, var_38239_cast_fp16))[name = tensor<string, []>("op_38480_cast_fp16")];
            tensor<fp16, []> var_38481_to_fp16 = const()[name = tensor<string, []>("op_38481_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3247_cast_fp16 = mul(x = var_38480_cast_fp16, y = var_38481_to_fp16)[name = tensor<string, []>("aw_3247_cast_fp16")];
            tensor<string, []> var_38484_equation_0 = const()[name = tensor<string, []>("op_38484_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_38484_cast_fp16 = einsum(equation = var_38484_equation_0, values = (var_38326_cast_fp16, var_38243_cast_fp16))[name = tensor<string, []>("op_38484_cast_fp16")];
            tensor<fp16, []> var_38485_to_fp16 = const()[name = tensor<string, []>("op_38485_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3249_cast_fp16 = mul(x = var_38484_cast_fp16, y = var_38485_to_fp16)[name = tensor<string, []>("aw_3249_cast_fp16")];
            tensor<string, []> var_38488_equation_0 = const()[name = tensor<string, []>("op_38488_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_38488_cast_fp16 = einsum(equation = var_38488_equation_0, values = (var_38330_cast_fp16, var_38247_cast_fp16))[name = tensor<string, []>("op_38488_cast_fp16")];
            tensor<fp16, []> var_38489_to_fp16 = const()[name = tensor<string, []>("op_38489_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3251_cast_fp16 = mul(x = var_38488_cast_fp16, y = var_38489_to_fp16)[name = tensor<string, []>("aw_3251_cast_fp16")];
            tensor<string, []> var_38492_equation_0 = const()[name = tensor<string, []>("op_38492_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_38492_cast_fp16 = einsum(equation = var_38492_equation_0, values = (var_38334_cast_fp16, var_38251_cast_fp16))[name = tensor<string, []>("op_38492_cast_fp16")];
            tensor<fp16, []> var_38493_to_fp16 = const()[name = tensor<string, []>("op_38493_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3253_cast_fp16 = mul(x = var_38492_cast_fp16, y = var_38493_to_fp16)[name = tensor<string, []>("aw_3253_cast_fp16")];
            tensor<string, []> var_38496_equation_0 = const()[name = tensor<string, []>("op_38496_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_38496_cast_fp16 = einsum(equation = var_38496_equation_0, values = (var_38338_cast_fp16, var_38255_cast_fp16))[name = tensor<string, []>("op_38496_cast_fp16")];
            tensor<fp16, []> var_38497_to_fp16 = const()[name = tensor<string, []>("op_38497_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3255_cast_fp16 = mul(x = var_38496_cast_fp16, y = var_38497_to_fp16)[name = tensor<string, []>("aw_3255_cast_fp16")];
            tensor<string, []> var_38500_equation_0 = const()[name = tensor<string, []>("op_38500_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_38500_cast_fp16 = einsum(equation = var_38500_equation_0, values = (var_38342_cast_fp16, var_38259_cast_fp16))[name = tensor<string, []>("op_38500_cast_fp16")];
            tensor<fp16, []> var_38501_to_fp16 = const()[name = tensor<string, []>("op_38501_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3257_cast_fp16 = mul(x = var_38500_cast_fp16, y = var_38501_to_fp16)[name = tensor<string, []>("aw_3257_cast_fp16")];
            tensor<string, []> var_38504_equation_0 = const()[name = tensor<string, []>("op_38504_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_38504_cast_fp16 = einsum(equation = var_38504_equation_0, values = (var_38346_cast_fp16, var_38263_cast_fp16))[name = tensor<string, []>("op_38504_cast_fp16")];
            tensor<fp16, []> var_38505_to_fp16 = const()[name = tensor<string, []>("op_38505_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3259_cast_fp16 = mul(x = var_38504_cast_fp16, y = var_38505_to_fp16)[name = tensor<string, []>("aw_3259_cast_fp16")];
            tensor<string, []> var_38508_equation_0 = const()[name = tensor<string, []>("op_38508_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_38508_cast_fp16 = einsum(equation = var_38508_equation_0, values = (var_38350_cast_fp16, var_38267_cast_fp16))[name = tensor<string, []>("op_38508_cast_fp16")];
            tensor<fp16, []> var_38509_to_fp16 = const()[name = tensor<string, []>("op_38509_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3261_cast_fp16 = mul(x = var_38508_cast_fp16, y = var_38509_to_fp16)[name = tensor<string, []>("aw_3261_cast_fp16")];
            tensor<string, []> var_38512_equation_0 = const()[name = tensor<string, []>("op_38512_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_38512_cast_fp16 = einsum(equation = var_38512_equation_0, values = (var_38354_cast_fp16, var_38271_cast_fp16))[name = tensor<string, []>("op_38512_cast_fp16")];
            tensor<fp16, []> var_38513_to_fp16 = const()[name = tensor<string, []>("op_38513_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3263_cast_fp16 = mul(x = var_38512_cast_fp16, y = var_38513_to_fp16)[name = tensor<string, []>("aw_3263_cast_fp16")];
            tensor<string, []> var_38516_equation_0 = const()[name = tensor<string, []>("op_38516_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_38516_cast_fp16 = einsum(equation = var_38516_equation_0, values = (var_38358_cast_fp16, var_38275_cast_fp16))[name = tensor<string, []>("op_38516_cast_fp16")];
            tensor<fp16, []> var_38517_to_fp16 = const()[name = tensor<string, []>("op_38517_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3265_cast_fp16 = mul(x = var_38516_cast_fp16, y = var_38517_to_fp16)[name = tensor<string, []>("aw_3265_cast_fp16")];
            tensor<string, []> var_38520_equation_0 = const()[name = tensor<string, []>("op_38520_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_38520_cast_fp16 = einsum(equation = var_38520_equation_0, values = (var_38362_cast_fp16, var_38279_cast_fp16))[name = tensor<string, []>("op_38520_cast_fp16")];
            tensor<fp16, []> var_38521_to_fp16 = const()[name = tensor<string, []>("op_38521_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3267_cast_fp16 = mul(x = var_38520_cast_fp16, y = var_38521_to_fp16)[name = tensor<string, []>("aw_3267_cast_fp16")];
            tensor<string, []> var_38524_equation_0 = const()[name = tensor<string, []>("op_38524_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_38524_cast_fp16 = einsum(equation = var_38524_equation_0, values = (var_38366_cast_fp16, var_38283_cast_fp16))[name = tensor<string, []>("op_38524_cast_fp16")];
            tensor<fp16, []> var_38525_to_fp16 = const()[name = tensor<string, []>("op_38525_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3269_cast_fp16 = mul(x = var_38524_cast_fp16, y = var_38525_to_fp16)[name = tensor<string, []>("aw_3269_cast_fp16")];
            tensor<string, []> var_38528_equation_0 = const()[name = tensor<string, []>("op_38528_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_38528_cast_fp16 = einsum(equation = var_38528_equation_0, values = (var_38370_cast_fp16, var_38287_cast_fp16))[name = tensor<string, []>("op_38528_cast_fp16")];
            tensor<fp16, []> var_38529_to_fp16 = const()[name = tensor<string, []>("op_38529_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3271_cast_fp16 = mul(x = var_38528_cast_fp16, y = var_38529_to_fp16)[name = tensor<string, []>("aw_3271_cast_fp16")];
            tensor<string, []> var_38532_equation_0 = const()[name = tensor<string, []>("op_38532_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_38532_cast_fp16 = einsum(equation = var_38532_equation_0, values = (var_38374_cast_fp16, var_38291_cast_fp16))[name = tensor<string, []>("op_38532_cast_fp16")];
            tensor<fp16, []> var_38533_to_fp16 = const()[name = tensor<string, []>("op_38533_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3273_cast_fp16 = mul(x = var_38532_cast_fp16, y = var_38533_to_fp16)[name = tensor<string, []>("aw_3273_cast_fp16")];
            tensor<string, []> var_38536_equation_0 = const()[name = tensor<string, []>("op_38536_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_38536_cast_fp16 = einsum(equation = var_38536_equation_0, values = (var_38378_cast_fp16, var_38295_cast_fp16))[name = tensor<string, []>("op_38536_cast_fp16")];
            tensor<fp16, []> var_38537_to_fp16 = const()[name = tensor<string, []>("op_38537_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3275_cast_fp16 = mul(x = var_38536_cast_fp16, y = var_38537_to_fp16)[name = tensor<string, []>("aw_3275_cast_fp16")];
            tensor<string, []> var_38540_equation_0 = const()[name = tensor<string, []>("op_38540_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_38540_cast_fp16 = einsum(equation = var_38540_equation_0, values = (var_38382_cast_fp16, var_38299_cast_fp16))[name = tensor<string, []>("op_38540_cast_fp16")];
            tensor<fp16, []> var_38541_to_fp16 = const()[name = tensor<string, []>("op_38541_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3277_cast_fp16 = mul(x = var_38540_cast_fp16, y = var_38541_to_fp16)[name = tensor<string, []>("aw_3277_cast_fp16")];
            tensor<string, []> var_38544_equation_0 = const()[name = tensor<string, []>("op_38544_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_38544_cast_fp16 = einsum(equation = var_38544_equation_0, values = (var_38386_cast_fp16, var_38303_cast_fp16))[name = tensor<string, []>("op_38544_cast_fp16")];
            tensor<fp16, []> var_38545_to_fp16 = const()[name = tensor<string, []>("op_38545_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3279_cast_fp16 = mul(x = var_38544_cast_fp16, y = var_38545_to_fp16)[name = tensor<string, []>("aw_3279_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_38547_cast_fp16 = softmax(axis = var_30355, x = aw_3241_cast_fp16)[name = tensor<string, []>("op_38547_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_38548_cast_fp16 = softmax(axis = var_30355, x = aw_3243_cast_fp16)[name = tensor<string, []>("op_38548_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_38549_cast_fp16 = softmax(axis = var_30355, x = aw_3245_cast_fp16)[name = tensor<string, []>("op_38549_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_38550_cast_fp16 = softmax(axis = var_30355, x = aw_3247_cast_fp16)[name = tensor<string, []>("op_38550_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_38551_cast_fp16 = softmax(axis = var_30355, x = aw_3249_cast_fp16)[name = tensor<string, []>("op_38551_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_38552_cast_fp16 = softmax(axis = var_30355, x = aw_3251_cast_fp16)[name = tensor<string, []>("op_38552_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_38553_cast_fp16 = softmax(axis = var_30355, x = aw_3253_cast_fp16)[name = tensor<string, []>("op_38553_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_38554_cast_fp16 = softmax(axis = var_30355, x = aw_3255_cast_fp16)[name = tensor<string, []>("op_38554_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_38555_cast_fp16 = softmax(axis = var_30355, x = aw_3257_cast_fp16)[name = tensor<string, []>("op_38555_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_38556_cast_fp16 = softmax(axis = var_30355, x = aw_3259_cast_fp16)[name = tensor<string, []>("op_38556_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_38557_cast_fp16 = softmax(axis = var_30355, x = aw_3261_cast_fp16)[name = tensor<string, []>("op_38557_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_38558_cast_fp16 = softmax(axis = var_30355, x = aw_3263_cast_fp16)[name = tensor<string, []>("op_38558_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_38559_cast_fp16 = softmax(axis = var_30355, x = aw_3265_cast_fp16)[name = tensor<string, []>("op_38559_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_38560_cast_fp16 = softmax(axis = var_30355, x = aw_3267_cast_fp16)[name = tensor<string, []>("op_38560_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_38561_cast_fp16 = softmax(axis = var_30355, x = aw_3269_cast_fp16)[name = tensor<string, []>("op_38561_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_38562_cast_fp16 = softmax(axis = var_30355, x = aw_3271_cast_fp16)[name = tensor<string, []>("op_38562_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_38563_cast_fp16 = softmax(axis = var_30355, x = aw_3273_cast_fp16)[name = tensor<string, []>("op_38563_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_38564_cast_fp16 = softmax(axis = var_30355, x = aw_3275_cast_fp16)[name = tensor<string, []>("op_38564_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_38565_cast_fp16 = softmax(axis = var_30355, x = aw_3277_cast_fp16)[name = tensor<string, []>("op_38565_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_38566_cast_fp16 = softmax(axis = var_30355, x = aw_3279_cast_fp16)[name = tensor<string, []>("op_38566_cast_fp16")];
            tensor<string, []> var_38568_equation_0 = const()[name = tensor<string, []>("op_38568_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38568_cast_fp16 = einsum(equation = var_38568_equation_0, values = (var_38388_cast_fp16, var_38547_cast_fp16))[name = tensor<string, []>("op_38568_cast_fp16")];
            tensor<string, []> var_38570_equation_0 = const()[name = tensor<string, []>("op_38570_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38570_cast_fp16 = einsum(equation = var_38570_equation_0, values = (var_38392_cast_fp16, var_38548_cast_fp16))[name = tensor<string, []>("op_38570_cast_fp16")];
            tensor<string, []> var_38572_equation_0 = const()[name = tensor<string, []>("op_38572_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38572_cast_fp16 = einsum(equation = var_38572_equation_0, values = (var_38396_cast_fp16, var_38549_cast_fp16))[name = tensor<string, []>("op_38572_cast_fp16")];
            tensor<string, []> var_38574_equation_0 = const()[name = tensor<string, []>("op_38574_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38574_cast_fp16 = einsum(equation = var_38574_equation_0, values = (var_38400_cast_fp16, var_38550_cast_fp16))[name = tensor<string, []>("op_38574_cast_fp16")];
            tensor<string, []> var_38576_equation_0 = const()[name = tensor<string, []>("op_38576_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38576_cast_fp16 = einsum(equation = var_38576_equation_0, values = (var_38404_cast_fp16, var_38551_cast_fp16))[name = tensor<string, []>("op_38576_cast_fp16")];
            tensor<string, []> var_38578_equation_0 = const()[name = tensor<string, []>("op_38578_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38578_cast_fp16 = einsum(equation = var_38578_equation_0, values = (var_38408_cast_fp16, var_38552_cast_fp16))[name = tensor<string, []>("op_38578_cast_fp16")];
            tensor<string, []> var_38580_equation_0 = const()[name = tensor<string, []>("op_38580_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38580_cast_fp16 = einsum(equation = var_38580_equation_0, values = (var_38412_cast_fp16, var_38553_cast_fp16))[name = tensor<string, []>("op_38580_cast_fp16")];
            tensor<string, []> var_38582_equation_0 = const()[name = tensor<string, []>("op_38582_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38582_cast_fp16 = einsum(equation = var_38582_equation_0, values = (var_38416_cast_fp16, var_38554_cast_fp16))[name = tensor<string, []>("op_38582_cast_fp16")];
            tensor<string, []> var_38584_equation_0 = const()[name = tensor<string, []>("op_38584_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38584_cast_fp16 = einsum(equation = var_38584_equation_0, values = (var_38420_cast_fp16, var_38555_cast_fp16))[name = tensor<string, []>("op_38584_cast_fp16")];
            tensor<string, []> var_38586_equation_0 = const()[name = tensor<string, []>("op_38586_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38586_cast_fp16 = einsum(equation = var_38586_equation_0, values = (var_38424_cast_fp16, var_38556_cast_fp16))[name = tensor<string, []>("op_38586_cast_fp16")];
            tensor<string, []> var_38588_equation_0 = const()[name = tensor<string, []>("op_38588_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38588_cast_fp16 = einsum(equation = var_38588_equation_0, values = (var_38428_cast_fp16, var_38557_cast_fp16))[name = tensor<string, []>("op_38588_cast_fp16")];
            tensor<string, []> var_38590_equation_0 = const()[name = tensor<string, []>("op_38590_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38590_cast_fp16 = einsum(equation = var_38590_equation_0, values = (var_38432_cast_fp16, var_38558_cast_fp16))[name = tensor<string, []>("op_38590_cast_fp16")];
            tensor<string, []> var_38592_equation_0 = const()[name = tensor<string, []>("op_38592_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38592_cast_fp16 = einsum(equation = var_38592_equation_0, values = (var_38436_cast_fp16, var_38559_cast_fp16))[name = tensor<string, []>("op_38592_cast_fp16")];
            tensor<string, []> var_38594_equation_0 = const()[name = tensor<string, []>("op_38594_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38594_cast_fp16 = einsum(equation = var_38594_equation_0, values = (var_38440_cast_fp16, var_38560_cast_fp16))[name = tensor<string, []>("op_38594_cast_fp16")];
            tensor<string, []> var_38596_equation_0 = const()[name = tensor<string, []>("op_38596_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38596_cast_fp16 = einsum(equation = var_38596_equation_0, values = (var_38444_cast_fp16, var_38561_cast_fp16))[name = tensor<string, []>("op_38596_cast_fp16")];
            tensor<string, []> var_38598_equation_0 = const()[name = tensor<string, []>("op_38598_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38598_cast_fp16 = einsum(equation = var_38598_equation_0, values = (var_38448_cast_fp16, var_38562_cast_fp16))[name = tensor<string, []>("op_38598_cast_fp16")];
            tensor<string, []> var_38600_equation_0 = const()[name = tensor<string, []>("op_38600_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38600_cast_fp16 = einsum(equation = var_38600_equation_0, values = (var_38452_cast_fp16, var_38563_cast_fp16))[name = tensor<string, []>("op_38600_cast_fp16")];
            tensor<string, []> var_38602_equation_0 = const()[name = tensor<string, []>("op_38602_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38602_cast_fp16 = einsum(equation = var_38602_equation_0, values = (var_38456_cast_fp16, var_38564_cast_fp16))[name = tensor<string, []>("op_38602_cast_fp16")];
            tensor<string, []> var_38604_equation_0 = const()[name = tensor<string, []>("op_38604_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38604_cast_fp16 = einsum(equation = var_38604_equation_0, values = (var_38460_cast_fp16, var_38565_cast_fp16))[name = tensor<string, []>("op_38604_cast_fp16")];
            tensor<string, []> var_38606_equation_0 = const()[name = tensor<string, []>("op_38606_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38606_cast_fp16 = einsum(equation = var_38606_equation_0, values = (var_38464_cast_fp16, var_38566_cast_fp16))[name = tensor<string, []>("op_38606_cast_fp16")];
            tensor<bool, []> input_505_interleave_0 = const()[name = tensor<string, []>("input_505_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_505_cast_fp16 = concat(axis = var_30355, interleave = input_505_interleave_0, values = (var_38568_cast_fp16, var_38570_cast_fp16, var_38572_cast_fp16, var_38574_cast_fp16, var_38576_cast_fp16, var_38578_cast_fp16, var_38580_cast_fp16, var_38582_cast_fp16, var_38584_cast_fp16, var_38586_cast_fp16, var_38588_cast_fp16, var_38590_cast_fp16, var_38592_cast_fp16, var_38594_cast_fp16, var_38596_cast_fp16, var_38598_cast_fp16, var_38600_cast_fp16, var_38602_cast_fp16, var_38604_cast_fp16, var_38606_cast_fp16))[name = tensor<string, []>("input_505_cast_fp16")];
            tensor<string, []> var_38616_pad_type_0 = const()[name = tensor<string, []>("op_38616_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_38616_strides_0 = const()[name = tensor<string, []>("op_38616_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_38616_pad_0 = const()[name = tensor<string, []>("op_38616_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_38616_dilations_0 = const()[name = tensor<string, []>("op_38616_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_38616_groups_0 = const()[name = tensor<string, []>("op_38616_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_8_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(222586496))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(223815360))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_8_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_8_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_8_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(223815552)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_38616_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_8_attn2_to_out_0_bias_to_fp16, dilations = var_38616_dilations_0, groups = var_38616_groups_0, pad = var_38616_pad_0, pad_type = var_38616_pad_type_0, strides = var_38616_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_8_attn2_to_out_0_weight_to_fp16_palettized, x = input_505_cast_fp16)[name = tensor<string, []>("op_38616_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_257_cast_fp16 = add(x = var_38616_cast_fp16, y = inputs_255_cast_fp16)[name = tensor<string, []>("inputs_257_cast_fp16")];
            tensor<int32, [1]> input_507_axes_0 = const()[name = tensor<string, []>("input_507_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> input_507_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_507_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(223818176)))];
            tensor<fp16, [1280]> input_507_beta_0_to_fp16 = const()[name = tensor<string, []>("input_507_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(223820800)))];
            tensor<fp16, []> var_38626_to_fp16 = const()[name = tensor<string, []>("op_38626_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> input_507_cast_fp16 = layer_norm(axes = input_507_axes_0, beta = input_507_beta_0_to_fp16, epsilon = var_38626_to_fp16, gamma = input_507_gamma_0_to_fp16, x = inputs_257_cast_fp16)[name = tensor<string, []>("input_507_cast_fp16")];
            tensor<string, []> var_38646_pad_type_0 = const()[name = tensor<string, []>("op_38646_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_38646_strides_0 = const()[name = tensor<string, []>("op_38646_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_38646_pad_0 = const()[name = tensor<string, []>("op_38646_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_38646_dilations_0 = const()[name = tensor<string, []>("op_38646_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_38646_groups_0 = const()[name = tensor<string, []>("op_38646_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_8_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [9830400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(223823424))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(233653888))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_8_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([10240, 1280, 1, 1])];
            tensor<fp16, [10240]> up_blocks_0_attentions_0_transformer_blocks_8_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_8_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(233654080)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_38646_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_8_ff_net_0_proj_bias_to_fp16, dilations = var_38646_dilations_0, groups = var_38646_groups_0, pad = var_38646_pad_0, pad_type = var_38646_pad_type_0, strides = var_38646_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_8_ff_net_0_proj_weight_to_fp16_palettized, x = input_507_cast_fp16)[name = tensor<string, []>("op_38646_cast_fp16")];
            tensor<int32, [2]> var_38647_split_sizes_0 = const()[name = tensor<string, []>("op_38647_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_38647_axis_0 = const()[name = tensor<string, []>("op_38647_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_38647_cast_fp16_0, tensor<fp16, [2, 5120, 1, 1024]> var_38647_cast_fp16_1 = split(axis = var_38647_axis_0, split_sizes = var_38647_split_sizes_0, x = var_38646_cast_fp16)[name = tensor<string, []>("op_38647_cast_fp16")];
            tensor<string, []> var_38649_mode_0 = const()[name = tensor<string, []>("op_38649_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_38649_cast_fp16 = gelu(mode = var_38649_mode_0, x = var_38647_cast_fp16_1)[name = tensor<string, []>("op_38649_cast_fp16")];
            tensor<fp16, [2, 5120, 1, 1024]> input_509_cast_fp16 = mul(x = var_38647_cast_fp16_0, y = var_38649_cast_fp16)[name = tensor<string, []>("input_509_cast_fp16")];
            tensor<string, []> var_38657_pad_type_0 = const()[name = tensor<string, []>("op_38657_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_38657_strides_0 = const()[name = tensor<string, []>("op_38657_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_38657_pad_0 = const()[name = tensor<string, []>("op_38657_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_38657_dilations_0 = const()[name = tensor<string, []>("op_38657_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_38657_groups_0 = const()[name = tensor<string, []>("op_38657_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_8_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [4915200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(233674624))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(238589888))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_8_ff_net_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 5120, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_8_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_8_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(238590080)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_38657_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_8_ff_net_2_bias_to_fp16, dilations = var_38657_dilations_0, groups = var_38657_groups_0, pad = var_38657_pad_0, pad_type = var_38657_pad_type_0, strides = var_38657_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_8_ff_net_2_weight_to_fp16_palettized, x = input_509_cast_fp16)[name = tensor<string, []>("op_38657_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_259_cast_fp16 = add(x = var_38657_cast_fp16, y = inputs_257_cast_fp16)[name = tensor<string, []>("inputs_259_cast_fp16")];
            tensor<int32, [1]> hidden_states_341_axes_0 = const()[name = tensor<string, []>("hidden_states_341_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_341_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_341_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(238592704)))];
            tensor<fp16, [1280]> hidden_states_341_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_341_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(238595328)))];
            tensor<fp16, []> var_38673_to_fp16 = const()[name = tensor<string, []>("op_38673_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_341_cast_fp16 = layer_norm(axes = hidden_states_341_axes_0, beta = hidden_states_341_beta_0_to_fp16, epsilon = var_38673_to_fp16, gamma = hidden_states_341_gamma_0_to_fp16, x = inputs_259_cast_fp16)[name = tensor<string, []>("hidden_states_341_cast_fp16")];
            tensor<string, []> q_173_pad_type_0 = const()[name = tensor<string, []>("q_173_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_173_strides_0 = const()[name = tensor<string, []>("q_173_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_173_pad_0 = const()[name = tensor<string, []>("q_173_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_173_dilations_0 = const()[name = tensor<string, []>("q_173_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_173_groups_0 = const()[name = tensor<string, []>("q_173_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_9_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(238597952))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(239826816))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_9_attn1_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_173_cast_fp16 = conv(dilations = q_173_dilations_0, groups = q_173_groups_0, pad = q_173_pad_0, pad_type = q_173_pad_type_0, strides = q_173_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_9_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_341_cast_fp16)[name = tensor<string, []>("q_173_cast_fp16")];
            tensor<string, []> k_345_pad_type_0 = const()[name = tensor<string, []>("k_345_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_345_strides_0 = const()[name = tensor<string, []>("k_345_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_345_pad_0 = const()[name = tensor<string, []>("k_345_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_345_dilations_0 = const()[name = tensor<string, []>("k_345_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_345_groups_0 = const()[name = tensor<string, []>("k_345_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_9_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(239827008))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(241055872))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_9_attn1_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> k_345_cast_fp16 = conv(dilations = k_345_dilations_0, groups = k_345_groups_0, pad = k_345_pad_0, pad_type = k_345_pad_type_0, strides = k_345_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_9_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_341_cast_fp16)[name = tensor<string, []>("k_345_cast_fp16")];
            tensor<string, []> v_173_pad_type_0 = const()[name = tensor<string, []>("v_173_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_173_strides_0 = const()[name = tensor<string, []>("v_173_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_173_pad_0 = const()[name = tensor<string, []>("v_173_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_173_dilations_0 = const()[name = tensor<string, []>("v_173_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_173_groups_0 = const()[name = tensor<string, []>("v_173_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_9_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(241056064))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(242284928))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_9_attn1_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> v_173_cast_fp16 = conv(dilations = v_173_dilations_0, groups = v_173_groups_0, pad = v_173_pad_0, pad_type = v_173_pad_type_0, strides = v_173_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_9_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_341_cast_fp16)[name = tensor<string, []>("v_173_cast_fp16")];
            tensor<int32, [4]> var_38706_begin_0 = const()[name = tensor<string, []>("op_38706_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_38706_end_0 = const()[name = tensor<string, []>("op_38706_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_38706_end_mask_0 = const()[name = tensor<string, []>("op_38706_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38706_cast_fp16 = slice_by_index(begin = var_38706_begin_0, end = var_38706_end_0, end_mask = var_38706_end_mask_0, x = q_173_cast_fp16)[name = tensor<string, []>("op_38706_cast_fp16")];
            tensor<int32, [4]> var_38710_begin_0 = const()[name = tensor<string, []>("op_38710_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_38710_end_0 = const()[name = tensor<string, []>("op_38710_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_38710_end_mask_0 = const()[name = tensor<string, []>("op_38710_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38710_cast_fp16 = slice_by_index(begin = var_38710_begin_0, end = var_38710_end_0, end_mask = var_38710_end_mask_0, x = q_173_cast_fp16)[name = tensor<string, []>("op_38710_cast_fp16")];
            tensor<int32, [4]> var_38714_begin_0 = const()[name = tensor<string, []>("op_38714_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_38714_end_0 = const()[name = tensor<string, []>("op_38714_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_38714_end_mask_0 = const()[name = tensor<string, []>("op_38714_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38714_cast_fp16 = slice_by_index(begin = var_38714_begin_0, end = var_38714_end_0, end_mask = var_38714_end_mask_0, x = q_173_cast_fp16)[name = tensor<string, []>("op_38714_cast_fp16")];
            tensor<int32, [4]> var_38718_begin_0 = const()[name = tensor<string, []>("op_38718_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_38718_end_0 = const()[name = tensor<string, []>("op_38718_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_38718_end_mask_0 = const()[name = tensor<string, []>("op_38718_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38718_cast_fp16 = slice_by_index(begin = var_38718_begin_0, end = var_38718_end_0, end_mask = var_38718_end_mask_0, x = q_173_cast_fp16)[name = tensor<string, []>("op_38718_cast_fp16")];
            tensor<int32, [4]> var_38722_begin_0 = const()[name = tensor<string, []>("op_38722_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_38722_end_0 = const()[name = tensor<string, []>("op_38722_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_38722_end_mask_0 = const()[name = tensor<string, []>("op_38722_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38722_cast_fp16 = slice_by_index(begin = var_38722_begin_0, end = var_38722_end_0, end_mask = var_38722_end_mask_0, x = q_173_cast_fp16)[name = tensor<string, []>("op_38722_cast_fp16")];
            tensor<int32, [4]> var_38726_begin_0 = const()[name = tensor<string, []>("op_38726_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_38726_end_0 = const()[name = tensor<string, []>("op_38726_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_38726_end_mask_0 = const()[name = tensor<string, []>("op_38726_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38726_cast_fp16 = slice_by_index(begin = var_38726_begin_0, end = var_38726_end_0, end_mask = var_38726_end_mask_0, x = q_173_cast_fp16)[name = tensor<string, []>("op_38726_cast_fp16")];
            tensor<int32, [4]> var_38730_begin_0 = const()[name = tensor<string, []>("op_38730_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_38730_end_0 = const()[name = tensor<string, []>("op_38730_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_38730_end_mask_0 = const()[name = tensor<string, []>("op_38730_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38730_cast_fp16 = slice_by_index(begin = var_38730_begin_0, end = var_38730_end_0, end_mask = var_38730_end_mask_0, x = q_173_cast_fp16)[name = tensor<string, []>("op_38730_cast_fp16")];
            tensor<int32, [4]> var_38734_begin_0 = const()[name = tensor<string, []>("op_38734_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_38734_end_0 = const()[name = tensor<string, []>("op_38734_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_38734_end_mask_0 = const()[name = tensor<string, []>("op_38734_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38734_cast_fp16 = slice_by_index(begin = var_38734_begin_0, end = var_38734_end_0, end_mask = var_38734_end_mask_0, x = q_173_cast_fp16)[name = tensor<string, []>("op_38734_cast_fp16")];
            tensor<int32, [4]> var_38738_begin_0 = const()[name = tensor<string, []>("op_38738_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_38738_end_0 = const()[name = tensor<string, []>("op_38738_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_38738_end_mask_0 = const()[name = tensor<string, []>("op_38738_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38738_cast_fp16 = slice_by_index(begin = var_38738_begin_0, end = var_38738_end_0, end_mask = var_38738_end_mask_0, x = q_173_cast_fp16)[name = tensor<string, []>("op_38738_cast_fp16")];
            tensor<int32, [4]> var_38742_begin_0 = const()[name = tensor<string, []>("op_38742_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_38742_end_0 = const()[name = tensor<string, []>("op_38742_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_38742_end_mask_0 = const()[name = tensor<string, []>("op_38742_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38742_cast_fp16 = slice_by_index(begin = var_38742_begin_0, end = var_38742_end_0, end_mask = var_38742_end_mask_0, x = q_173_cast_fp16)[name = tensor<string, []>("op_38742_cast_fp16")];
            tensor<int32, [4]> var_38746_begin_0 = const()[name = tensor<string, []>("op_38746_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_38746_end_0 = const()[name = tensor<string, []>("op_38746_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_38746_end_mask_0 = const()[name = tensor<string, []>("op_38746_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38746_cast_fp16 = slice_by_index(begin = var_38746_begin_0, end = var_38746_end_0, end_mask = var_38746_end_mask_0, x = q_173_cast_fp16)[name = tensor<string, []>("op_38746_cast_fp16")];
            tensor<int32, [4]> var_38750_begin_0 = const()[name = tensor<string, []>("op_38750_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_38750_end_0 = const()[name = tensor<string, []>("op_38750_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_38750_end_mask_0 = const()[name = tensor<string, []>("op_38750_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38750_cast_fp16 = slice_by_index(begin = var_38750_begin_0, end = var_38750_end_0, end_mask = var_38750_end_mask_0, x = q_173_cast_fp16)[name = tensor<string, []>("op_38750_cast_fp16")];
            tensor<int32, [4]> var_38754_begin_0 = const()[name = tensor<string, []>("op_38754_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_38754_end_0 = const()[name = tensor<string, []>("op_38754_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_38754_end_mask_0 = const()[name = tensor<string, []>("op_38754_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38754_cast_fp16 = slice_by_index(begin = var_38754_begin_0, end = var_38754_end_0, end_mask = var_38754_end_mask_0, x = q_173_cast_fp16)[name = tensor<string, []>("op_38754_cast_fp16")];
            tensor<int32, [4]> var_38758_begin_0 = const()[name = tensor<string, []>("op_38758_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_38758_end_0 = const()[name = tensor<string, []>("op_38758_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_38758_end_mask_0 = const()[name = tensor<string, []>("op_38758_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38758_cast_fp16 = slice_by_index(begin = var_38758_begin_0, end = var_38758_end_0, end_mask = var_38758_end_mask_0, x = q_173_cast_fp16)[name = tensor<string, []>("op_38758_cast_fp16")];
            tensor<int32, [4]> var_38762_begin_0 = const()[name = tensor<string, []>("op_38762_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_38762_end_0 = const()[name = tensor<string, []>("op_38762_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_38762_end_mask_0 = const()[name = tensor<string, []>("op_38762_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38762_cast_fp16 = slice_by_index(begin = var_38762_begin_0, end = var_38762_end_0, end_mask = var_38762_end_mask_0, x = q_173_cast_fp16)[name = tensor<string, []>("op_38762_cast_fp16")];
            tensor<int32, [4]> var_38766_begin_0 = const()[name = tensor<string, []>("op_38766_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_38766_end_0 = const()[name = tensor<string, []>("op_38766_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_38766_end_mask_0 = const()[name = tensor<string, []>("op_38766_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38766_cast_fp16 = slice_by_index(begin = var_38766_begin_0, end = var_38766_end_0, end_mask = var_38766_end_mask_0, x = q_173_cast_fp16)[name = tensor<string, []>("op_38766_cast_fp16")];
            tensor<int32, [4]> var_38770_begin_0 = const()[name = tensor<string, []>("op_38770_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_38770_end_0 = const()[name = tensor<string, []>("op_38770_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_38770_end_mask_0 = const()[name = tensor<string, []>("op_38770_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38770_cast_fp16 = slice_by_index(begin = var_38770_begin_0, end = var_38770_end_0, end_mask = var_38770_end_mask_0, x = q_173_cast_fp16)[name = tensor<string, []>("op_38770_cast_fp16")];
            tensor<int32, [4]> var_38774_begin_0 = const()[name = tensor<string, []>("op_38774_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_38774_end_0 = const()[name = tensor<string, []>("op_38774_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_38774_end_mask_0 = const()[name = tensor<string, []>("op_38774_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38774_cast_fp16 = slice_by_index(begin = var_38774_begin_0, end = var_38774_end_0, end_mask = var_38774_end_mask_0, x = q_173_cast_fp16)[name = tensor<string, []>("op_38774_cast_fp16")];
            tensor<int32, [4]> var_38778_begin_0 = const()[name = tensor<string, []>("op_38778_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_38778_end_0 = const()[name = tensor<string, []>("op_38778_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_38778_end_mask_0 = const()[name = tensor<string, []>("op_38778_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38778_cast_fp16 = slice_by_index(begin = var_38778_begin_0, end = var_38778_end_0, end_mask = var_38778_end_mask_0, x = q_173_cast_fp16)[name = tensor<string, []>("op_38778_cast_fp16")];
            tensor<int32, [4]> var_38782_begin_0 = const()[name = tensor<string, []>("op_38782_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_38782_end_0 = const()[name = tensor<string, []>("op_38782_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_38782_end_mask_0 = const()[name = tensor<string, []>("op_38782_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38782_cast_fp16 = slice_by_index(begin = var_38782_begin_0, end = var_38782_end_0, end_mask = var_38782_end_mask_0, x = q_173_cast_fp16)[name = tensor<string, []>("op_38782_cast_fp16")];
            tensor<int32, [4]> k_347_perm_0 = const()[name = tensor<string, []>("k_347_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_38789_begin_0 = const()[name = tensor<string, []>("op_38789_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_38789_end_0 = const()[name = tensor<string, []>("op_38789_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_38789_end_mask_0 = const()[name = tensor<string, []>("op_38789_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> k_347_cast_fp16 = transpose(perm = k_347_perm_0, x = k_345_cast_fp16)[name = tensor<string, []>("transpose_53")];
            tensor<fp16, [2, 1024, 1, 64]> var_38789_cast_fp16 = slice_by_index(begin = var_38789_begin_0, end = var_38789_end_0, end_mask = var_38789_end_mask_0, x = k_347_cast_fp16)[name = tensor<string, []>("op_38789_cast_fp16")];
            tensor<int32, [4]> var_38793_begin_0 = const()[name = tensor<string, []>("op_38793_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_38793_end_0 = const()[name = tensor<string, []>("op_38793_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_38793_end_mask_0 = const()[name = tensor<string, []>("op_38793_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_38793_cast_fp16 = slice_by_index(begin = var_38793_begin_0, end = var_38793_end_0, end_mask = var_38793_end_mask_0, x = k_347_cast_fp16)[name = tensor<string, []>("op_38793_cast_fp16")];
            tensor<int32, [4]> var_38797_begin_0 = const()[name = tensor<string, []>("op_38797_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_38797_end_0 = const()[name = tensor<string, []>("op_38797_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_38797_end_mask_0 = const()[name = tensor<string, []>("op_38797_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_38797_cast_fp16 = slice_by_index(begin = var_38797_begin_0, end = var_38797_end_0, end_mask = var_38797_end_mask_0, x = k_347_cast_fp16)[name = tensor<string, []>("op_38797_cast_fp16")];
            tensor<int32, [4]> var_38801_begin_0 = const()[name = tensor<string, []>("op_38801_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_38801_end_0 = const()[name = tensor<string, []>("op_38801_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_38801_end_mask_0 = const()[name = tensor<string, []>("op_38801_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_38801_cast_fp16 = slice_by_index(begin = var_38801_begin_0, end = var_38801_end_0, end_mask = var_38801_end_mask_0, x = k_347_cast_fp16)[name = tensor<string, []>("op_38801_cast_fp16")];
            tensor<int32, [4]> var_38805_begin_0 = const()[name = tensor<string, []>("op_38805_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_38805_end_0 = const()[name = tensor<string, []>("op_38805_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_38805_end_mask_0 = const()[name = tensor<string, []>("op_38805_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_38805_cast_fp16 = slice_by_index(begin = var_38805_begin_0, end = var_38805_end_0, end_mask = var_38805_end_mask_0, x = k_347_cast_fp16)[name = tensor<string, []>("op_38805_cast_fp16")];
            tensor<int32, [4]> var_38809_begin_0 = const()[name = tensor<string, []>("op_38809_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_38809_end_0 = const()[name = tensor<string, []>("op_38809_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_38809_end_mask_0 = const()[name = tensor<string, []>("op_38809_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_38809_cast_fp16 = slice_by_index(begin = var_38809_begin_0, end = var_38809_end_0, end_mask = var_38809_end_mask_0, x = k_347_cast_fp16)[name = tensor<string, []>("op_38809_cast_fp16")];
            tensor<int32, [4]> var_38813_begin_0 = const()[name = tensor<string, []>("op_38813_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_38813_end_0 = const()[name = tensor<string, []>("op_38813_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_38813_end_mask_0 = const()[name = tensor<string, []>("op_38813_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_38813_cast_fp16 = slice_by_index(begin = var_38813_begin_0, end = var_38813_end_0, end_mask = var_38813_end_mask_0, x = k_347_cast_fp16)[name = tensor<string, []>("op_38813_cast_fp16")];
            tensor<int32, [4]> var_38817_begin_0 = const()[name = tensor<string, []>("op_38817_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_38817_end_0 = const()[name = tensor<string, []>("op_38817_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_38817_end_mask_0 = const()[name = tensor<string, []>("op_38817_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_38817_cast_fp16 = slice_by_index(begin = var_38817_begin_0, end = var_38817_end_0, end_mask = var_38817_end_mask_0, x = k_347_cast_fp16)[name = tensor<string, []>("op_38817_cast_fp16")];
            tensor<int32, [4]> var_38821_begin_0 = const()[name = tensor<string, []>("op_38821_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_38821_end_0 = const()[name = tensor<string, []>("op_38821_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_38821_end_mask_0 = const()[name = tensor<string, []>("op_38821_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_38821_cast_fp16 = slice_by_index(begin = var_38821_begin_0, end = var_38821_end_0, end_mask = var_38821_end_mask_0, x = k_347_cast_fp16)[name = tensor<string, []>("op_38821_cast_fp16")];
            tensor<int32, [4]> var_38825_begin_0 = const()[name = tensor<string, []>("op_38825_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_38825_end_0 = const()[name = tensor<string, []>("op_38825_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_38825_end_mask_0 = const()[name = tensor<string, []>("op_38825_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_38825_cast_fp16 = slice_by_index(begin = var_38825_begin_0, end = var_38825_end_0, end_mask = var_38825_end_mask_0, x = k_347_cast_fp16)[name = tensor<string, []>("op_38825_cast_fp16")];
            tensor<int32, [4]> var_38829_begin_0 = const()[name = tensor<string, []>("op_38829_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_38829_end_0 = const()[name = tensor<string, []>("op_38829_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_38829_end_mask_0 = const()[name = tensor<string, []>("op_38829_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_38829_cast_fp16 = slice_by_index(begin = var_38829_begin_0, end = var_38829_end_0, end_mask = var_38829_end_mask_0, x = k_347_cast_fp16)[name = tensor<string, []>("op_38829_cast_fp16")];
            tensor<int32, [4]> var_38833_begin_0 = const()[name = tensor<string, []>("op_38833_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_38833_end_0 = const()[name = tensor<string, []>("op_38833_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_38833_end_mask_0 = const()[name = tensor<string, []>("op_38833_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_38833_cast_fp16 = slice_by_index(begin = var_38833_begin_0, end = var_38833_end_0, end_mask = var_38833_end_mask_0, x = k_347_cast_fp16)[name = tensor<string, []>("op_38833_cast_fp16")];
            tensor<int32, [4]> var_38837_begin_0 = const()[name = tensor<string, []>("op_38837_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_38837_end_0 = const()[name = tensor<string, []>("op_38837_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_38837_end_mask_0 = const()[name = tensor<string, []>("op_38837_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_38837_cast_fp16 = slice_by_index(begin = var_38837_begin_0, end = var_38837_end_0, end_mask = var_38837_end_mask_0, x = k_347_cast_fp16)[name = tensor<string, []>("op_38837_cast_fp16")];
            tensor<int32, [4]> var_38841_begin_0 = const()[name = tensor<string, []>("op_38841_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_38841_end_0 = const()[name = tensor<string, []>("op_38841_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_38841_end_mask_0 = const()[name = tensor<string, []>("op_38841_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_38841_cast_fp16 = slice_by_index(begin = var_38841_begin_0, end = var_38841_end_0, end_mask = var_38841_end_mask_0, x = k_347_cast_fp16)[name = tensor<string, []>("op_38841_cast_fp16")];
            tensor<int32, [4]> var_38845_begin_0 = const()[name = tensor<string, []>("op_38845_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_38845_end_0 = const()[name = tensor<string, []>("op_38845_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_38845_end_mask_0 = const()[name = tensor<string, []>("op_38845_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_38845_cast_fp16 = slice_by_index(begin = var_38845_begin_0, end = var_38845_end_0, end_mask = var_38845_end_mask_0, x = k_347_cast_fp16)[name = tensor<string, []>("op_38845_cast_fp16")];
            tensor<int32, [4]> var_38849_begin_0 = const()[name = tensor<string, []>("op_38849_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_38849_end_0 = const()[name = tensor<string, []>("op_38849_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_38849_end_mask_0 = const()[name = tensor<string, []>("op_38849_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_38849_cast_fp16 = slice_by_index(begin = var_38849_begin_0, end = var_38849_end_0, end_mask = var_38849_end_mask_0, x = k_347_cast_fp16)[name = tensor<string, []>("op_38849_cast_fp16")];
            tensor<int32, [4]> var_38853_begin_0 = const()[name = tensor<string, []>("op_38853_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_38853_end_0 = const()[name = tensor<string, []>("op_38853_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_38853_end_mask_0 = const()[name = tensor<string, []>("op_38853_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_38853_cast_fp16 = slice_by_index(begin = var_38853_begin_0, end = var_38853_end_0, end_mask = var_38853_end_mask_0, x = k_347_cast_fp16)[name = tensor<string, []>("op_38853_cast_fp16")];
            tensor<int32, [4]> var_38857_begin_0 = const()[name = tensor<string, []>("op_38857_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_38857_end_0 = const()[name = tensor<string, []>("op_38857_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_38857_end_mask_0 = const()[name = tensor<string, []>("op_38857_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_38857_cast_fp16 = slice_by_index(begin = var_38857_begin_0, end = var_38857_end_0, end_mask = var_38857_end_mask_0, x = k_347_cast_fp16)[name = tensor<string, []>("op_38857_cast_fp16")];
            tensor<int32, [4]> var_38861_begin_0 = const()[name = tensor<string, []>("op_38861_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_38861_end_0 = const()[name = tensor<string, []>("op_38861_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_38861_end_mask_0 = const()[name = tensor<string, []>("op_38861_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_38861_cast_fp16 = slice_by_index(begin = var_38861_begin_0, end = var_38861_end_0, end_mask = var_38861_end_mask_0, x = k_347_cast_fp16)[name = tensor<string, []>("op_38861_cast_fp16")];
            tensor<int32, [4]> var_38865_begin_0 = const()[name = tensor<string, []>("op_38865_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_38865_end_0 = const()[name = tensor<string, []>("op_38865_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_38865_end_mask_0 = const()[name = tensor<string, []>("op_38865_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_38865_cast_fp16 = slice_by_index(begin = var_38865_begin_0, end = var_38865_end_0, end_mask = var_38865_end_mask_0, x = k_347_cast_fp16)[name = tensor<string, []>("op_38865_cast_fp16")];
            tensor<int32, [4]> var_38867_begin_0 = const()[name = tensor<string, []>("op_38867_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_38867_end_0 = const()[name = tensor<string, []>("op_38867_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_38867_end_mask_0 = const()[name = tensor<string, []>("op_38867_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38867_cast_fp16 = slice_by_index(begin = var_38867_begin_0, end = var_38867_end_0, end_mask = var_38867_end_mask_0, x = v_173_cast_fp16)[name = tensor<string, []>("op_38867_cast_fp16")];
            tensor<int32, [4]> var_38871_begin_0 = const()[name = tensor<string, []>("op_38871_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_38871_end_0 = const()[name = tensor<string, []>("op_38871_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_38871_end_mask_0 = const()[name = tensor<string, []>("op_38871_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38871_cast_fp16 = slice_by_index(begin = var_38871_begin_0, end = var_38871_end_0, end_mask = var_38871_end_mask_0, x = v_173_cast_fp16)[name = tensor<string, []>("op_38871_cast_fp16")];
            tensor<int32, [4]> var_38875_begin_0 = const()[name = tensor<string, []>("op_38875_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_38875_end_0 = const()[name = tensor<string, []>("op_38875_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_38875_end_mask_0 = const()[name = tensor<string, []>("op_38875_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38875_cast_fp16 = slice_by_index(begin = var_38875_begin_0, end = var_38875_end_0, end_mask = var_38875_end_mask_0, x = v_173_cast_fp16)[name = tensor<string, []>("op_38875_cast_fp16")];
            tensor<int32, [4]> var_38879_begin_0 = const()[name = tensor<string, []>("op_38879_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_38879_end_0 = const()[name = tensor<string, []>("op_38879_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_38879_end_mask_0 = const()[name = tensor<string, []>("op_38879_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38879_cast_fp16 = slice_by_index(begin = var_38879_begin_0, end = var_38879_end_0, end_mask = var_38879_end_mask_0, x = v_173_cast_fp16)[name = tensor<string, []>("op_38879_cast_fp16")];
            tensor<int32, [4]> var_38883_begin_0 = const()[name = tensor<string, []>("op_38883_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_38883_end_0 = const()[name = tensor<string, []>("op_38883_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_38883_end_mask_0 = const()[name = tensor<string, []>("op_38883_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38883_cast_fp16 = slice_by_index(begin = var_38883_begin_0, end = var_38883_end_0, end_mask = var_38883_end_mask_0, x = v_173_cast_fp16)[name = tensor<string, []>("op_38883_cast_fp16")];
            tensor<int32, [4]> var_38887_begin_0 = const()[name = tensor<string, []>("op_38887_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_38887_end_0 = const()[name = tensor<string, []>("op_38887_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_38887_end_mask_0 = const()[name = tensor<string, []>("op_38887_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38887_cast_fp16 = slice_by_index(begin = var_38887_begin_0, end = var_38887_end_0, end_mask = var_38887_end_mask_0, x = v_173_cast_fp16)[name = tensor<string, []>("op_38887_cast_fp16")];
            tensor<int32, [4]> var_38891_begin_0 = const()[name = tensor<string, []>("op_38891_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_38891_end_0 = const()[name = tensor<string, []>("op_38891_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_38891_end_mask_0 = const()[name = tensor<string, []>("op_38891_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38891_cast_fp16 = slice_by_index(begin = var_38891_begin_0, end = var_38891_end_0, end_mask = var_38891_end_mask_0, x = v_173_cast_fp16)[name = tensor<string, []>("op_38891_cast_fp16")];
            tensor<int32, [4]> var_38895_begin_0 = const()[name = tensor<string, []>("op_38895_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_38895_end_0 = const()[name = tensor<string, []>("op_38895_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_38895_end_mask_0 = const()[name = tensor<string, []>("op_38895_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38895_cast_fp16 = slice_by_index(begin = var_38895_begin_0, end = var_38895_end_0, end_mask = var_38895_end_mask_0, x = v_173_cast_fp16)[name = tensor<string, []>("op_38895_cast_fp16")];
            tensor<int32, [4]> var_38899_begin_0 = const()[name = tensor<string, []>("op_38899_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_38899_end_0 = const()[name = tensor<string, []>("op_38899_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_38899_end_mask_0 = const()[name = tensor<string, []>("op_38899_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38899_cast_fp16 = slice_by_index(begin = var_38899_begin_0, end = var_38899_end_0, end_mask = var_38899_end_mask_0, x = v_173_cast_fp16)[name = tensor<string, []>("op_38899_cast_fp16")];
            tensor<int32, [4]> var_38903_begin_0 = const()[name = tensor<string, []>("op_38903_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_38903_end_0 = const()[name = tensor<string, []>("op_38903_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_38903_end_mask_0 = const()[name = tensor<string, []>("op_38903_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38903_cast_fp16 = slice_by_index(begin = var_38903_begin_0, end = var_38903_end_0, end_mask = var_38903_end_mask_0, x = v_173_cast_fp16)[name = tensor<string, []>("op_38903_cast_fp16")];
            tensor<int32, [4]> var_38907_begin_0 = const()[name = tensor<string, []>("op_38907_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_38907_end_0 = const()[name = tensor<string, []>("op_38907_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_38907_end_mask_0 = const()[name = tensor<string, []>("op_38907_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38907_cast_fp16 = slice_by_index(begin = var_38907_begin_0, end = var_38907_end_0, end_mask = var_38907_end_mask_0, x = v_173_cast_fp16)[name = tensor<string, []>("op_38907_cast_fp16")];
            tensor<int32, [4]> var_38911_begin_0 = const()[name = tensor<string, []>("op_38911_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_38911_end_0 = const()[name = tensor<string, []>("op_38911_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_38911_end_mask_0 = const()[name = tensor<string, []>("op_38911_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38911_cast_fp16 = slice_by_index(begin = var_38911_begin_0, end = var_38911_end_0, end_mask = var_38911_end_mask_0, x = v_173_cast_fp16)[name = tensor<string, []>("op_38911_cast_fp16")];
            tensor<int32, [4]> var_38915_begin_0 = const()[name = tensor<string, []>("op_38915_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_38915_end_0 = const()[name = tensor<string, []>("op_38915_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_38915_end_mask_0 = const()[name = tensor<string, []>("op_38915_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38915_cast_fp16 = slice_by_index(begin = var_38915_begin_0, end = var_38915_end_0, end_mask = var_38915_end_mask_0, x = v_173_cast_fp16)[name = tensor<string, []>("op_38915_cast_fp16")];
            tensor<int32, [4]> var_38919_begin_0 = const()[name = tensor<string, []>("op_38919_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_38919_end_0 = const()[name = tensor<string, []>("op_38919_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_38919_end_mask_0 = const()[name = tensor<string, []>("op_38919_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38919_cast_fp16 = slice_by_index(begin = var_38919_begin_0, end = var_38919_end_0, end_mask = var_38919_end_mask_0, x = v_173_cast_fp16)[name = tensor<string, []>("op_38919_cast_fp16")];
            tensor<int32, [4]> var_38923_begin_0 = const()[name = tensor<string, []>("op_38923_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_38923_end_0 = const()[name = tensor<string, []>("op_38923_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_38923_end_mask_0 = const()[name = tensor<string, []>("op_38923_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38923_cast_fp16 = slice_by_index(begin = var_38923_begin_0, end = var_38923_end_0, end_mask = var_38923_end_mask_0, x = v_173_cast_fp16)[name = tensor<string, []>("op_38923_cast_fp16")];
            tensor<int32, [4]> var_38927_begin_0 = const()[name = tensor<string, []>("op_38927_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_38927_end_0 = const()[name = tensor<string, []>("op_38927_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_38927_end_mask_0 = const()[name = tensor<string, []>("op_38927_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38927_cast_fp16 = slice_by_index(begin = var_38927_begin_0, end = var_38927_end_0, end_mask = var_38927_end_mask_0, x = v_173_cast_fp16)[name = tensor<string, []>("op_38927_cast_fp16")];
            tensor<int32, [4]> var_38931_begin_0 = const()[name = tensor<string, []>("op_38931_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_38931_end_0 = const()[name = tensor<string, []>("op_38931_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_38931_end_mask_0 = const()[name = tensor<string, []>("op_38931_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38931_cast_fp16 = slice_by_index(begin = var_38931_begin_0, end = var_38931_end_0, end_mask = var_38931_end_mask_0, x = v_173_cast_fp16)[name = tensor<string, []>("op_38931_cast_fp16")];
            tensor<int32, [4]> var_38935_begin_0 = const()[name = tensor<string, []>("op_38935_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_38935_end_0 = const()[name = tensor<string, []>("op_38935_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_38935_end_mask_0 = const()[name = tensor<string, []>("op_38935_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38935_cast_fp16 = slice_by_index(begin = var_38935_begin_0, end = var_38935_end_0, end_mask = var_38935_end_mask_0, x = v_173_cast_fp16)[name = tensor<string, []>("op_38935_cast_fp16")];
            tensor<int32, [4]> var_38939_begin_0 = const()[name = tensor<string, []>("op_38939_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_38939_end_0 = const()[name = tensor<string, []>("op_38939_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_38939_end_mask_0 = const()[name = tensor<string, []>("op_38939_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38939_cast_fp16 = slice_by_index(begin = var_38939_begin_0, end = var_38939_end_0, end_mask = var_38939_end_mask_0, x = v_173_cast_fp16)[name = tensor<string, []>("op_38939_cast_fp16")];
            tensor<int32, [4]> var_38943_begin_0 = const()[name = tensor<string, []>("op_38943_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_38943_end_0 = const()[name = tensor<string, []>("op_38943_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_38943_end_mask_0 = const()[name = tensor<string, []>("op_38943_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38943_cast_fp16 = slice_by_index(begin = var_38943_begin_0, end = var_38943_end_0, end_mask = var_38943_end_mask_0, x = v_173_cast_fp16)[name = tensor<string, []>("op_38943_cast_fp16")];
            tensor<string, []> var_38947_equation_0 = const()[name = tensor<string, []>("op_38947_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38947_cast_fp16 = einsum(equation = var_38947_equation_0, values = (var_38789_cast_fp16, var_38706_cast_fp16))[name = tensor<string, []>("op_38947_cast_fp16")];
            tensor<fp16, []> var_38948_to_fp16 = const()[name = tensor<string, []>("op_38948_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3281_cast_fp16 = mul(x = var_38947_cast_fp16, y = var_38948_to_fp16)[name = tensor<string, []>("aw_3281_cast_fp16")];
            tensor<string, []> var_38951_equation_0 = const()[name = tensor<string, []>("op_38951_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38951_cast_fp16 = einsum(equation = var_38951_equation_0, values = (var_38793_cast_fp16, var_38710_cast_fp16))[name = tensor<string, []>("op_38951_cast_fp16")];
            tensor<fp16, []> var_38952_to_fp16 = const()[name = tensor<string, []>("op_38952_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3283_cast_fp16 = mul(x = var_38951_cast_fp16, y = var_38952_to_fp16)[name = tensor<string, []>("aw_3283_cast_fp16")];
            tensor<string, []> var_38955_equation_0 = const()[name = tensor<string, []>("op_38955_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38955_cast_fp16 = einsum(equation = var_38955_equation_0, values = (var_38797_cast_fp16, var_38714_cast_fp16))[name = tensor<string, []>("op_38955_cast_fp16")];
            tensor<fp16, []> var_38956_to_fp16 = const()[name = tensor<string, []>("op_38956_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3285_cast_fp16 = mul(x = var_38955_cast_fp16, y = var_38956_to_fp16)[name = tensor<string, []>("aw_3285_cast_fp16")];
            tensor<string, []> var_38959_equation_0 = const()[name = tensor<string, []>("op_38959_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38959_cast_fp16 = einsum(equation = var_38959_equation_0, values = (var_38801_cast_fp16, var_38718_cast_fp16))[name = tensor<string, []>("op_38959_cast_fp16")];
            tensor<fp16, []> var_38960_to_fp16 = const()[name = tensor<string, []>("op_38960_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3287_cast_fp16 = mul(x = var_38959_cast_fp16, y = var_38960_to_fp16)[name = tensor<string, []>("aw_3287_cast_fp16")];
            tensor<string, []> var_38963_equation_0 = const()[name = tensor<string, []>("op_38963_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38963_cast_fp16 = einsum(equation = var_38963_equation_0, values = (var_38805_cast_fp16, var_38722_cast_fp16))[name = tensor<string, []>("op_38963_cast_fp16")];
            tensor<fp16, []> var_38964_to_fp16 = const()[name = tensor<string, []>("op_38964_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3289_cast_fp16 = mul(x = var_38963_cast_fp16, y = var_38964_to_fp16)[name = tensor<string, []>("aw_3289_cast_fp16")];
            tensor<string, []> var_38967_equation_0 = const()[name = tensor<string, []>("op_38967_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38967_cast_fp16 = einsum(equation = var_38967_equation_0, values = (var_38809_cast_fp16, var_38726_cast_fp16))[name = tensor<string, []>("op_38967_cast_fp16")];
            tensor<fp16, []> var_38968_to_fp16 = const()[name = tensor<string, []>("op_38968_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3291_cast_fp16 = mul(x = var_38967_cast_fp16, y = var_38968_to_fp16)[name = tensor<string, []>("aw_3291_cast_fp16")];
            tensor<string, []> var_38971_equation_0 = const()[name = tensor<string, []>("op_38971_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38971_cast_fp16 = einsum(equation = var_38971_equation_0, values = (var_38813_cast_fp16, var_38730_cast_fp16))[name = tensor<string, []>("op_38971_cast_fp16")];
            tensor<fp16, []> var_38972_to_fp16 = const()[name = tensor<string, []>("op_38972_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3293_cast_fp16 = mul(x = var_38971_cast_fp16, y = var_38972_to_fp16)[name = tensor<string, []>("aw_3293_cast_fp16")];
            tensor<string, []> var_38975_equation_0 = const()[name = tensor<string, []>("op_38975_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38975_cast_fp16 = einsum(equation = var_38975_equation_0, values = (var_38817_cast_fp16, var_38734_cast_fp16))[name = tensor<string, []>("op_38975_cast_fp16")];
            tensor<fp16, []> var_38976_to_fp16 = const()[name = tensor<string, []>("op_38976_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3295_cast_fp16 = mul(x = var_38975_cast_fp16, y = var_38976_to_fp16)[name = tensor<string, []>("aw_3295_cast_fp16")];
            tensor<string, []> var_38979_equation_0 = const()[name = tensor<string, []>("op_38979_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38979_cast_fp16 = einsum(equation = var_38979_equation_0, values = (var_38821_cast_fp16, var_38738_cast_fp16))[name = tensor<string, []>("op_38979_cast_fp16")];
            tensor<fp16, []> var_38980_to_fp16 = const()[name = tensor<string, []>("op_38980_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3297_cast_fp16 = mul(x = var_38979_cast_fp16, y = var_38980_to_fp16)[name = tensor<string, []>("aw_3297_cast_fp16")];
            tensor<string, []> var_38983_equation_0 = const()[name = tensor<string, []>("op_38983_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38983_cast_fp16 = einsum(equation = var_38983_equation_0, values = (var_38825_cast_fp16, var_38742_cast_fp16))[name = tensor<string, []>("op_38983_cast_fp16")];
            tensor<fp16, []> var_38984_to_fp16 = const()[name = tensor<string, []>("op_38984_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3299_cast_fp16 = mul(x = var_38983_cast_fp16, y = var_38984_to_fp16)[name = tensor<string, []>("aw_3299_cast_fp16")];
            tensor<string, []> var_38987_equation_0 = const()[name = tensor<string, []>("op_38987_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38987_cast_fp16 = einsum(equation = var_38987_equation_0, values = (var_38829_cast_fp16, var_38746_cast_fp16))[name = tensor<string, []>("op_38987_cast_fp16")];
            tensor<fp16, []> var_38988_to_fp16 = const()[name = tensor<string, []>("op_38988_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3301_cast_fp16 = mul(x = var_38987_cast_fp16, y = var_38988_to_fp16)[name = tensor<string, []>("aw_3301_cast_fp16")];
            tensor<string, []> var_38991_equation_0 = const()[name = tensor<string, []>("op_38991_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38991_cast_fp16 = einsum(equation = var_38991_equation_0, values = (var_38833_cast_fp16, var_38750_cast_fp16))[name = tensor<string, []>("op_38991_cast_fp16")];
            tensor<fp16, []> var_38992_to_fp16 = const()[name = tensor<string, []>("op_38992_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3303_cast_fp16 = mul(x = var_38991_cast_fp16, y = var_38992_to_fp16)[name = tensor<string, []>("aw_3303_cast_fp16")];
            tensor<string, []> var_38995_equation_0 = const()[name = tensor<string, []>("op_38995_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38995_cast_fp16 = einsum(equation = var_38995_equation_0, values = (var_38837_cast_fp16, var_38754_cast_fp16))[name = tensor<string, []>("op_38995_cast_fp16")];
            tensor<fp16, []> var_38996_to_fp16 = const()[name = tensor<string, []>("op_38996_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3305_cast_fp16 = mul(x = var_38995_cast_fp16, y = var_38996_to_fp16)[name = tensor<string, []>("aw_3305_cast_fp16")];
            tensor<string, []> var_38999_equation_0 = const()[name = tensor<string, []>("op_38999_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38999_cast_fp16 = einsum(equation = var_38999_equation_0, values = (var_38841_cast_fp16, var_38758_cast_fp16))[name = tensor<string, []>("op_38999_cast_fp16")];
            tensor<fp16, []> var_39000_to_fp16 = const()[name = tensor<string, []>("op_39000_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3307_cast_fp16 = mul(x = var_38999_cast_fp16, y = var_39000_to_fp16)[name = tensor<string, []>("aw_3307_cast_fp16")];
            tensor<string, []> var_39003_equation_0 = const()[name = tensor<string, []>("op_39003_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39003_cast_fp16 = einsum(equation = var_39003_equation_0, values = (var_38845_cast_fp16, var_38762_cast_fp16))[name = tensor<string, []>("op_39003_cast_fp16")];
            tensor<fp16, []> var_39004_to_fp16 = const()[name = tensor<string, []>("op_39004_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3309_cast_fp16 = mul(x = var_39003_cast_fp16, y = var_39004_to_fp16)[name = tensor<string, []>("aw_3309_cast_fp16")];
            tensor<string, []> var_39007_equation_0 = const()[name = tensor<string, []>("op_39007_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39007_cast_fp16 = einsum(equation = var_39007_equation_0, values = (var_38849_cast_fp16, var_38766_cast_fp16))[name = tensor<string, []>("op_39007_cast_fp16")];
            tensor<fp16, []> var_39008_to_fp16 = const()[name = tensor<string, []>("op_39008_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3311_cast_fp16 = mul(x = var_39007_cast_fp16, y = var_39008_to_fp16)[name = tensor<string, []>("aw_3311_cast_fp16")];
            tensor<string, []> var_39011_equation_0 = const()[name = tensor<string, []>("op_39011_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39011_cast_fp16 = einsum(equation = var_39011_equation_0, values = (var_38853_cast_fp16, var_38770_cast_fp16))[name = tensor<string, []>("op_39011_cast_fp16")];
            tensor<fp16, []> var_39012_to_fp16 = const()[name = tensor<string, []>("op_39012_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3313_cast_fp16 = mul(x = var_39011_cast_fp16, y = var_39012_to_fp16)[name = tensor<string, []>("aw_3313_cast_fp16")];
            tensor<string, []> var_39015_equation_0 = const()[name = tensor<string, []>("op_39015_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39015_cast_fp16 = einsum(equation = var_39015_equation_0, values = (var_38857_cast_fp16, var_38774_cast_fp16))[name = tensor<string, []>("op_39015_cast_fp16")];
            tensor<fp16, []> var_39016_to_fp16 = const()[name = tensor<string, []>("op_39016_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3315_cast_fp16 = mul(x = var_39015_cast_fp16, y = var_39016_to_fp16)[name = tensor<string, []>("aw_3315_cast_fp16")];
            tensor<string, []> var_39019_equation_0 = const()[name = tensor<string, []>("op_39019_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39019_cast_fp16 = einsum(equation = var_39019_equation_0, values = (var_38861_cast_fp16, var_38778_cast_fp16))[name = tensor<string, []>("op_39019_cast_fp16")];
            tensor<fp16, []> var_39020_to_fp16 = const()[name = tensor<string, []>("op_39020_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3317_cast_fp16 = mul(x = var_39019_cast_fp16, y = var_39020_to_fp16)[name = tensor<string, []>("aw_3317_cast_fp16")];
            tensor<string, []> var_39023_equation_0 = const()[name = tensor<string, []>("op_39023_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39023_cast_fp16 = einsum(equation = var_39023_equation_0, values = (var_38865_cast_fp16, var_38782_cast_fp16))[name = tensor<string, []>("op_39023_cast_fp16")];
            tensor<fp16, []> var_39024_to_fp16 = const()[name = tensor<string, []>("op_39024_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3319_cast_fp16 = mul(x = var_39023_cast_fp16, y = var_39024_to_fp16)[name = tensor<string, []>("aw_3319_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39026_cast_fp16 = softmax(axis = var_30355, x = aw_3281_cast_fp16)[name = tensor<string, []>("op_39026_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39027_cast_fp16 = softmax(axis = var_30355, x = aw_3283_cast_fp16)[name = tensor<string, []>("op_39027_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39028_cast_fp16 = softmax(axis = var_30355, x = aw_3285_cast_fp16)[name = tensor<string, []>("op_39028_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39029_cast_fp16 = softmax(axis = var_30355, x = aw_3287_cast_fp16)[name = tensor<string, []>("op_39029_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39030_cast_fp16 = softmax(axis = var_30355, x = aw_3289_cast_fp16)[name = tensor<string, []>("op_39030_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39031_cast_fp16 = softmax(axis = var_30355, x = aw_3291_cast_fp16)[name = tensor<string, []>("op_39031_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39032_cast_fp16 = softmax(axis = var_30355, x = aw_3293_cast_fp16)[name = tensor<string, []>("op_39032_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39033_cast_fp16 = softmax(axis = var_30355, x = aw_3295_cast_fp16)[name = tensor<string, []>("op_39033_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39034_cast_fp16 = softmax(axis = var_30355, x = aw_3297_cast_fp16)[name = tensor<string, []>("op_39034_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39035_cast_fp16 = softmax(axis = var_30355, x = aw_3299_cast_fp16)[name = tensor<string, []>("op_39035_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39036_cast_fp16 = softmax(axis = var_30355, x = aw_3301_cast_fp16)[name = tensor<string, []>("op_39036_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39037_cast_fp16 = softmax(axis = var_30355, x = aw_3303_cast_fp16)[name = tensor<string, []>("op_39037_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39038_cast_fp16 = softmax(axis = var_30355, x = aw_3305_cast_fp16)[name = tensor<string, []>("op_39038_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39039_cast_fp16 = softmax(axis = var_30355, x = aw_3307_cast_fp16)[name = tensor<string, []>("op_39039_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39040_cast_fp16 = softmax(axis = var_30355, x = aw_3309_cast_fp16)[name = tensor<string, []>("op_39040_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39041_cast_fp16 = softmax(axis = var_30355, x = aw_3311_cast_fp16)[name = tensor<string, []>("op_39041_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39042_cast_fp16 = softmax(axis = var_30355, x = aw_3313_cast_fp16)[name = tensor<string, []>("op_39042_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39043_cast_fp16 = softmax(axis = var_30355, x = aw_3315_cast_fp16)[name = tensor<string, []>("op_39043_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39044_cast_fp16 = softmax(axis = var_30355, x = aw_3317_cast_fp16)[name = tensor<string, []>("op_39044_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39045_cast_fp16 = softmax(axis = var_30355, x = aw_3319_cast_fp16)[name = tensor<string, []>("op_39045_cast_fp16")];
            tensor<string, []> var_39047_equation_0 = const()[name = tensor<string, []>("op_39047_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39047_cast_fp16 = einsum(equation = var_39047_equation_0, values = (var_38867_cast_fp16, var_39026_cast_fp16))[name = tensor<string, []>("op_39047_cast_fp16")];
            tensor<string, []> var_39049_equation_0 = const()[name = tensor<string, []>("op_39049_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39049_cast_fp16 = einsum(equation = var_39049_equation_0, values = (var_38871_cast_fp16, var_39027_cast_fp16))[name = tensor<string, []>("op_39049_cast_fp16")];
            tensor<string, []> var_39051_equation_0 = const()[name = tensor<string, []>("op_39051_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39051_cast_fp16 = einsum(equation = var_39051_equation_0, values = (var_38875_cast_fp16, var_39028_cast_fp16))[name = tensor<string, []>("op_39051_cast_fp16")];
            tensor<string, []> var_39053_equation_0 = const()[name = tensor<string, []>("op_39053_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39053_cast_fp16 = einsum(equation = var_39053_equation_0, values = (var_38879_cast_fp16, var_39029_cast_fp16))[name = tensor<string, []>("op_39053_cast_fp16")];
            tensor<string, []> var_39055_equation_0 = const()[name = tensor<string, []>("op_39055_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39055_cast_fp16 = einsum(equation = var_39055_equation_0, values = (var_38883_cast_fp16, var_39030_cast_fp16))[name = tensor<string, []>("op_39055_cast_fp16")];
            tensor<string, []> var_39057_equation_0 = const()[name = tensor<string, []>("op_39057_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39057_cast_fp16 = einsum(equation = var_39057_equation_0, values = (var_38887_cast_fp16, var_39031_cast_fp16))[name = tensor<string, []>("op_39057_cast_fp16")];
            tensor<string, []> var_39059_equation_0 = const()[name = tensor<string, []>("op_39059_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39059_cast_fp16 = einsum(equation = var_39059_equation_0, values = (var_38891_cast_fp16, var_39032_cast_fp16))[name = tensor<string, []>("op_39059_cast_fp16")];
            tensor<string, []> var_39061_equation_0 = const()[name = tensor<string, []>("op_39061_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39061_cast_fp16 = einsum(equation = var_39061_equation_0, values = (var_38895_cast_fp16, var_39033_cast_fp16))[name = tensor<string, []>("op_39061_cast_fp16")];
            tensor<string, []> var_39063_equation_0 = const()[name = tensor<string, []>("op_39063_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39063_cast_fp16 = einsum(equation = var_39063_equation_0, values = (var_38899_cast_fp16, var_39034_cast_fp16))[name = tensor<string, []>("op_39063_cast_fp16")];
            tensor<string, []> var_39065_equation_0 = const()[name = tensor<string, []>("op_39065_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39065_cast_fp16 = einsum(equation = var_39065_equation_0, values = (var_38903_cast_fp16, var_39035_cast_fp16))[name = tensor<string, []>("op_39065_cast_fp16")];
            tensor<string, []> var_39067_equation_0 = const()[name = tensor<string, []>("op_39067_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39067_cast_fp16 = einsum(equation = var_39067_equation_0, values = (var_38907_cast_fp16, var_39036_cast_fp16))[name = tensor<string, []>("op_39067_cast_fp16")];
            tensor<string, []> var_39069_equation_0 = const()[name = tensor<string, []>("op_39069_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39069_cast_fp16 = einsum(equation = var_39069_equation_0, values = (var_38911_cast_fp16, var_39037_cast_fp16))[name = tensor<string, []>("op_39069_cast_fp16")];
            tensor<string, []> var_39071_equation_0 = const()[name = tensor<string, []>("op_39071_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39071_cast_fp16 = einsum(equation = var_39071_equation_0, values = (var_38915_cast_fp16, var_39038_cast_fp16))[name = tensor<string, []>("op_39071_cast_fp16")];
            tensor<string, []> var_39073_equation_0 = const()[name = tensor<string, []>("op_39073_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39073_cast_fp16 = einsum(equation = var_39073_equation_0, values = (var_38919_cast_fp16, var_39039_cast_fp16))[name = tensor<string, []>("op_39073_cast_fp16")];
            tensor<string, []> var_39075_equation_0 = const()[name = tensor<string, []>("op_39075_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39075_cast_fp16 = einsum(equation = var_39075_equation_0, values = (var_38923_cast_fp16, var_39040_cast_fp16))[name = tensor<string, []>("op_39075_cast_fp16")];
            tensor<string, []> var_39077_equation_0 = const()[name = tensor<string, []>("op_39077_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39077_cast_fp16 = einsum(equation = var_39077_equation_0, values = (var_38927_cast_fp16, var_39041_cast_fp16))[name = tensor<string, []>("op_39077_cast_fp16")];
            tensor<string, []> var_39079_equation_0 = const()[name = tensor<string, []>("op_39079_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39079_cast_fp16 = einsum(equation = var_39079_equation_0, values = (var_38931_cast_fp16, var_39042_cast_fp16))[name = tensor<string, []>("op_39079_cast_fp16")];
            tensor<string, []> var_39081_equation_0 = const()[name = tensor<string, []>("op_39081_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39081_cast_fp16 = einsum(equation = var_39081_equation_0, values = (var_38935_cast_fp16, var_39043_cast_fp16))[name = tensor<string, []>("op_39081_cast_fp16")];
            tensor<string, []> var_39083_equation_0 = const()[name = tensor<string, []>("op_39083_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39083_cast_fp16 = einsum(equation = var_39083_equation_0, values = (var_38939_cast_fp16, var_39044_cast_fp16))[name = tensor<string, []>("op_39083_cast_fp16")];
            tensor<string, []> var_39085_equation_0 = const()[name = tensor<string, []>("op_39085_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39085_cast_fp16 = einsum(equation = var_39085_equation_0, values = (var_38943_cast_fp16, var_39045_cast_fp16))[name = tensor<string, []>("op_39085_cast_fp16")];
            tensor<bool, []> input_511_interleave_0 = const()[name = tensor<string, []>("input_511_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_511_cast_fp16 = concat(axis = var_30355, interleave = input_511_interleave_0, values = (var_39047_cast_fp16, var_39049_cast_fp16, var_39051_cast_fp16, var_39053_cast_fp16, var_39055_cast_fp16, var_39057_cast_fp16, var_39059_cast_fp16, var_39061_cast_fp16, var_39063_cast_fp16, var_39065_cast_fp16, var_39067_cast_fp16, var_39069_cast_fp16, var_39071_cast_fp16, var_39073_cast_fp16, var_39075_cast_fp16, var_39077_cast_fp16, var_39079_cast_fp16, var_39081_cast_fp16, var_39083_cast_fp16, var_39085_cast_fp16))[name = tensor<string, []>("input_511_cast_fp16")];
            tensor<string, []> var_39095_pad_type_0 = const()[name = tensor<string, []>("op_39095_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_39095_strides_0 = const()[name = tensor<string, []>("op_39095_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_39095_pad_0 = const()[name = tensor<string, []>("op_39095_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_39095_dilations_0 = const()[name = tensor<string, []>("op_39095_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_39095_groups_0 = const()[name = tensor<string, []>("op_39095_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_9_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(242285120))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(243513984))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_9_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_9_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_9_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(243514176)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_39095_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_9_attn1_to_out_0_bias_to_fp16, dilations = var_39095_dilations_0, groups = var_39095_groups_0, pad = var_39095_pad_0, pad_type = var_39095_pad_type_0, strides = var_39095_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_9_attn1_to_out_0_weight_to_fp16_palettized, x = input_511_cast_fp16)[name = tensor<string, []>("op_39095_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_261_cast_fp16 = add(x = var_39095_cast_fp16, y = inputs_259_cast_fp16)[name = tensor<string, []>("inputs_261_cast_fp16")];
            tensor<int32, [1]> hidden_states_343_axes_0 = const()[name = tensor<string, []>("hidden_states_343_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_343_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_343_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(243516800)))];
            tensor<fp16, [1280]> hidden_states_343_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_343_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(243519424)))];
            tensor<fp16, []> var_39105_to_fp16 = const()[name = tensor<string, []>("op_39105_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_343_cast_fp16 = layer_norm(axes = hidden_states_343_axes_0, beta = hidden_states_343_beta_0_to_fp16, epsilon = var_39105_to_fp16, gamma = hidden_states_343_gamma_0_to_fp16, x = inputs_261_cast_fp16)[name = tensor<string, []>("hidden_states_343_cast_fp16")];
            tensor<string, []> q_175_pad_type_0 = const()[name = tensor<string, []>("q_175_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_175_strides_0 = const()[name = tensor<string, []>("q_175_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_175_pad_0 = const()[name = tensor<string, []>("q_175_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_175_dilations_0 = const()[name = tensor<string, []>("q_175_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_175_groups_0 = const()[name = tensor<string, []>("q_175_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_9_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(243522048))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(244750912))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_9_attn2_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_175_cast_fp16 = conv(dilations = q_175_dilations_0, groups = q_175_groups_0, pad = q_175_pad_0, pad_type = q_175_pad_type_0, strides = q_175_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_9_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_343_cast_fp16)[name = tensor<string, []>("q_175_cast_fp16")];
            tensor<string, []> k_349_pad_type_0 = const()[name = tensor<string, []>("k_349_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_349_strides_0 = const()[name = tensor<string, []>("k_349_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_349_pad_0 = const()[name = tensor<string, []>("k_349_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_349_dilations_0 = const()[name = tensor<string, []>("k_349_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_349_groups_0 = const()[name = tensor<string, []>("k_349_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_9_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(244751104))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(246717248))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_9_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> k_349_cast_fp16 = conv(dilations = k_349_dilations_0, groups = k_349_groups_0, pad = k_349_pad_0, pad_type = k_349_pad_type_0, strides = k_349_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_9_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("k_349_cast_fp16")];
            tensor<string, []> v_175_pad_type_0 = const()[name = tensor<string, []>("v_175_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_175_strides_0 = const()[name = tensor<string, []>("v_175_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_175_pad_0 = const()[name = tensor<string, []>("v_175_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_175_dilations_0 = const()[name = tensor<string, []>("v_175_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_175_groups_0 = const()[name = tensor<string, []>("v_175_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_9_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(246717440))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(248683584))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_9_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> v_175_cast_fp16 = conv(dilations = v_175_dilations_0, groups = v_175_groups_0, pad = v_175_pad_0, pad_type = v_175_pad_type_0, strides = v_175_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_9_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("v_175_cast_fp16")];
            tensor<int32, [4]> var_39138_begin_0 = const()[name = tensor<string, []>("op_39138_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_39138_end_0 = const()[name = tensor<string, []>("op_39138_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_39138_end_mask_0 = const()[name = tensor<string, []>("op_39138_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39138_cast_fp16 = slice_by_index(begin = var_39138_begin_0, end = var_39138_end_0, end_mask = var_39138_end_mask_0, x = q_175_cast_fp16)[name = tensor<string, []>("op_39138_cast_fp16")];
            tensor<int32, [4]> var_39142_begin_0 = const()[name = tensor<string, []>("op_39142_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_39142_end_0 = const()[name = tensor<string, []>("op_39142_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_39142_end_mask_0 = const()[name = tensor<string, []>("op_39142_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39142_cast_fp16 = slice_by_index(begin = var_39142_begin_0, end = var_39142_end_0, end_mask = var_39142_end_mask_0, x = q_175_cast_fp16)[name = tensor<string, []>("op_39142_cast_fp16")];
            tensor<int32, [4]> var_39146_begin_0 = const()[name = tensor<string, []>("op_39146_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_39146_end_0 = const()[name = tensor<string, []>("op_39146_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_39146_end_mask_0 = const()[name = tensor<string, []>("op_39146_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39146_cast_fp16 = slice_by_index(begin = var_39146_begin_0, end = var_39146_end_0, end_mask = var_39146_end_mask_0, x = q_175_cast_fp16)[name = tensor<string, []>("op_39146_cast_fp16")];
            tensor<int32, [4]> var_39150_begin_0 = const()[name = tensor<string, []>("op_39150_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_39150_end_0 = const()[name = tensor<string, []>("op_39150_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_39150_end_mask_0 = const()[name = tensor<string, []>("op_39150_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39150_cast_fp16 = slice_by_index(begin = var_39150_begin_0, end = var_39150_end_0, end_mask = var_39150_end_mask_0, x = q_175_cast_fp16)[name = tensor<string, []>("op_39150_cast_fp16")];
            tensor<int32, [4]> var_39154_begin_0 = const()[name = tensor<string, []>("op_39154_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_39154_end_0 = const()[name = tensor<string, []>("op_39154_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_39154_end_mask_0 = const()[name = tensor<string, []>("op_39154_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39154_cast_fp16 = slice_by_index(begin = var_39154_begin_0, end = var_39154_end_0, end_mask = var_39154_end_mask_0, x = q_175_cast_fp16)[name = tensor<string, []>("op_39154_cast_fp16")];
            tensor<int32, [4]> var_39158_begin_0 = const()[name = tensor<string, []>("op_39158_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_39158_end_0 = const()[name = tensor<string, []>("op_39158_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_39158_end_mask_0 = const()[name = tensor<string, []>("op_39158_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39158_cast_fp16 = slice_by_index(begin = var_39158_begin_0, end = var_39158_end_0, end_mask = var_39158_end_mask_0, x = q_175_cast_fp16)[name = tensor<string, []>("op_39158_cast_fp16")];
            tensor<int32, [4]> var_39162_begin_0 = const()[name = tensor<string, []>("op_39162_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_39162_end_0 = const()[name = tensor<string, []>("op_39162_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_39162_end_mask_0 = const()[name = tensor<string, []>("op_39162_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39162_cast_fp16 = slice_by_index(begin = var_39162_begin_0, end = var_39162_end_0, end_mask = var_39162_end_mask_0, x = q_175_cast_fp16)[name = tensor<string, []>("op_39162_cast_fp16")];
            tensor<int32, [4]> var_39166_begin_0 = const()[name = tensor<string, []>("op_39166_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_39166_end_0 = const()[name = tensor<string, []>("op_39166_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_39166_end_mask_0 = const()[name = tensor<string, []>("op_39166_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39166_cast_fp16 = slice_by_index(begin = var_39166_begin_0, end = var_39166_end_0, end_mask = var_39166_end_mask_0, x = q_175_cast_fp16)[name = tensor<string, []>("op_39166_cast_fp16")];
            tensor<int32, [4]> var_39170_begin_0 = const()[name = tensor<string, []>("op_39170_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_39170_end_0 = const()[name = tensor<string, []>("op_39170_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_39170_end_mask_0 = const()[name = tensor<string, []>("op_39170_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39170_cast_fp16 = slice_by_index(begin = var_39170_begin_0, end = var_39170_end_0, end_mask = var_39170_end_mask_0, x = q_175_cast_fp16)[name = tensor<string, []>("op_39170_cast_fp16")];
            tensor<int32, [4]> var_39174_begin_0 = const()[name = tensor<string, []>("op_39174_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_39174_end_0 = const()[name = tensor<string, []>("op_39174_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_39174_end_mask_0 = const()[name = tensor<string, []>("op_39174_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39174_cast_fp16 = slice_by_index(begin = var_39174_begin_0, end = var_39174_end_0, end_mask = var_39174_end_mask_0, x = q_175_cast_fp16)[name = tensor<string, []>("op_39174_cast_fp16")];
            tensor<int32, [4]> var_39178_begin_0 = const()[name = tensor<string, []>("op_39178_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_39178_end_0 = const()[name = tensor<string, []>("op_39178_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_39178_end_mask_0 = const()[name = tensor<string, []>("op_39178_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39178_cast_fp16 = slice_by_index(begin = var_39178_begin_0, end = var_39178_end_0, end_mask = var_39178_end_mask_0, x = q_175_cast_fp16)[name = tensor<string, []>("op_39178_cast_fp16")];
            tensor<int32, [4]> var_39182_begin_0 = const()[name = tensor<string, []>("op_39182_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_39182_end_0 = const()[name = tensor<string, []>("op_39182_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_39182_end_mask_0 = const()[name = tensor<string, []>("op_39182_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39182_cast_fp16 = slice_by_index(begin = var_39182_begin_0, end = var_39182_end_0, end_mask = var_39182_end_mask_0, x = q_175_cast_fp16)[name = tensor<string, []>("op_39182_cast_fp16")];
            tensor<int32, [4]> var_39186_begin_0 = const()[name = tensor<string, []>("op_39186_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_39186_end_0 = const()[name = tensor<string, []>("op_39186_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_39186_end_mask_0 = const()[name = tensor<string, []>("op_39186_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39186_cast_fp16 = slice_by_index(begin = var_39186_begin_0, end = var_39186_end_0, end_mask = var_39186_end_mask_0, x = q_175_cast_fp16)[name = tensor<string, []>("op_39186_cast_fp16")];
            tensor<int32, [4]> var_39190_begin_0 = const()[name = tensor<string, []>("op_39190_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_39190_end_0 = const()[name = tensor<string, []>("op_39190_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_39190_end_mask_0 = const()[name = tensor<string, []>("op_39190_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39190_cast_fp16 = slice_by_index(begin = var_39190_begin_0, end = var_39190_end_0, end_mask = var_39190_end_mask_0, x = q_175_cast_fp16)[name = tensor<string, []>("op_39190_cast_fp16")];
            tensor<int32, [4]> var_39194_begin_0 = const()[name = tensor<string, []>("op_39194_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_39194_end_0 = const()[name = tensor<string, []>("op_39194_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_39194_end_mask_0 = const()[name = tensor<string, []>("op_39194_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39194_cast_fp16 = slice_by_index(begin = var_39194_begin_0, end = var_39194_end_0, end_mask = var_39194_end_mask_0, x = q_175_cast_fp16)[name = tensor<string, []>("op_39194_cast_fp16")];
            tensor<int32, [4]> var_39198_begin_0 = const()[name = tensor<string, []>("op_39198_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_39198_end_0 = const()[name = tensor<string, []>("op_39198_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_39198_end_mask_0 = const()[name = tensor<string, []>("op_39198_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39198_cast_fp16 = slice_by_index(begin = var_39198_begin_0, end = var_39198_end_0, end_mask = var_39198_end_mask_0, x = q_175_cast_fp16)[name = tensor<string, []>("op_39198_cast_fp16")];
            tensor<int32, [4]> var_39202_begin_0 = const()[name = tensor<string, []>("op_39202_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_39202_end_0 = const()[name = tensor<string, []>("op_39202_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_39202_end_mask_0 = const()[name = tensor<string, []>("op_39202_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39202_cast_fp16 = slice_by_index(begin = var_39202_begin_0, end = var_39202_end_0, end_mask = var_39202_end_mask_0, x = q_175_cast_fp16)[name = tensor<string, []>("op_39202_cast_fp16")];
            tensor<int32, [4]> var_39206_begin_0 = const()[name = tensor<string, []>("op_39206_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_39206_end_0 = const()[name = tensor<string, []>("op_39206_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_39206_end_mask_0 = const()[name = tensor<string, []>("op_39206_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39206_cast_fp16 = slice_by_index(begin = var_39206_begin_0, end = var_39206_end_0, end_mask = var_39206_end_mask_0, x = q_175_cast_fp16)[name = tensor<string, []>("op_39206_cast_fp16")];
            tensor<int32, [4]> var_39210_begin_0 = const()[name = tensor<string, []>("op_39210_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_39210_end_0 = const()[name = tensor<string, []>("op_39210_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_39210_end_mask_0 = const()[name = tensor<string, []>("op_39210_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39210_cast_fp16 = slice_by_index(begin = var_39210_begin_0, end = var_39210_end_0, end_mask = var_39210_end_mask_0, x = q_175_cast_fp16)[name = tensor<string, []>("op_39210_cast_fp16")];
            tensor<int32, [4]> var_39214_begin_0 = const()[name = tensor<string, []>("op_39214_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_39214_end_0 = const()[name = tensor<string, []>("op_39214_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_39214_end_mask_0 = const()[name = tensor<string, []>("op_39214_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39214_cast_fp16 = slice_by_index(begin = var_39214_begin_0, end = var_39214_end_0, end_mask = var_39214_end_mask_0, x = q_175_cast_fp16)[name = tensor<string, []>("op_39214_cast_fp16")];
            tensor<int32, [4]> k_351_perm_0 = const()[name = tensor<string, []>("k_351_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_39221_begin_0 = const()[name = tensor<string, []>("op_39221_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_39221_end_0 = const()[name = tensor<string, []>("op_39221_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_39221_end_mask_0 = const()[name = tensor<string, []>("op_39221_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> k_351_cast_fp16 = transpose(perm = k_351_perm_0, x = k_349_cast_fp16)[name = tensor<string, []>("transpose_52")];
            tensor<fp16, [2, 77, 1, 64]> var_39221_cast_fp16 = slice_by_index(begin = var_39221_begin_0, end = var_39221_end_0, end_mask = var_39221_end_mask_0, x = k_351_cast_fp16)[name = tensor<string, []>("op_39221_cast_fp16")];
            tensor<int32, [4]> var_39225_begin_0 = const()[name = tensor<string, []>("op_39225_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_39225_end_0 = const()[name = tensor<string, []>("op_39225_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_39225_end_mask_0 = const()[name = tensor<string, []>("op_39225_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_39225_cast_fp16 = slice_by_index(begin = var_39225_begin_0, end = var_39225_end_0, end_mask = var_39225_end_mask_0, x = k_351_cast_fp16)[name = tensor<string, []>("op_39225_cast_fp16")];
            tensor<int32, [4]> var_39229_begin_0 = const()[name = tensor<string, []>("op_39229_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_39229_end_0 = const()[name = tensor<string, []>("op_39229_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_39229_end_mask_0 = const()[name = tensor<string, []>("op_39229_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_39229_cast_fp16 = slice_by_index(begin = var_39229_begin_0, end = var_39229_end_0, end_mask = var_39229_end_mask_0, x = k_351_cast_fp16)[name = tensor<string, []>("op_39229_cast_fp16")];
            tensor<int32, [4]> var_39233_begin_0 = const()[name = tensor<string, []>("op_39233_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_39233_end_0 = const()[name = tensor<string, []>("op_39233_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_39233_end_mask_0 = const()[name = tensor<string, []>("op_39233_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_39233_cast_fp16 = slice_by_index(begin = var_39233_begin_0, end = var_39233_end_0, end_mask = var_39233_end_mask_0, x = k_351_cast_fp16)[name = tensor<string, []>("op_39233_cast_fp16")];
            tensor<int32, [4]> var_39237_begin_0 = const()[name = tensor<string, []>("op_39237_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_39237_end_0 = const()[name = tensor<string, []>("op_39237_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_39237_end_mask_0 = const()[name = tensor<string, []>("op_39237_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_39237_cast_fp16 = slice_by_index(begin = var_39237_begin_0, end = var_39237_end_0, end_mask = var_39237_end_mask_0, x = k_351_cast_fp16)[name = tensor<string, []>("op_39237_cast_fp16")];
            tensor<int32, [4]> var_39241_begin_0 = const()[name = tensor<string, []>("op_39241_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_39241_end_0 = const()[name = tensor<string, []>("op_39241_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_39241_end_mask_0 = const()[name = tensor<string, []>("op_39241_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_39241_cast_fp16 = slice_by_index(begin = var_39241_begin_0, end = var_39241_end_0, end_mask = var_39241_end_mask_0, x = k_351_cast_fp16)[name = tensor<string, []>("op_39241_cast_fp16")];
            tensor<int32, [4]> var_39245_begin_0 = const()[name = tensor<string, []>("op_39245_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_39245_end_0 = const()[name = tensor<string, []>("op_39245_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_39245_end_mask_0 = const()[name = tensor<string, []>("op_39245_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_39245_cast_fp16 = slice_by_index(begin = var_39245_begin_0, end = var_39245_end_0, end_mask = var_39245_end_mask_0, x = k_351_cast_fp16)[name = tensor<string, []>("op_39245_cast_fp16")];
            tensor<int32, [4]> var_39249_begin_0 = const()[name = tensor<string, []>("op_39249_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_39249_end_0 = const()[name = tensor<string, []>("op_39249_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_39249_end_mask_0 = const()[name = tensor<string, []>("op_39249_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_39249_cast_fp16 = slice_by_index(begin = var_39249_begin_0, end = var_39249_end_0, end_mask = var_39249_end_mask_0, x = k_351_cast_fp16)[name = tensor<string, []>("op_39249_cast_fp16")];
            tensor<int32, [4]> var_39253_begin_0 = const()[name = tensor<string, []>("op_39253_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_39253_end_0 = const()[name = tensor<string, []>("op_39253_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_39253_end_mask_0 = const()[name = tensor<string, []>("op_39253_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_39253_cast_fp16 = slice_by_index(begin = var_39253_begin_0, end = var_39253_end_0, end_mask = var_39253_end_mask_0, x = k_351_cast_fp16)[name = tensor<string, []>("op_39253_cast_fp16")];
            tensor<int32, [4]> var_39257_begin_0 = const()[name = tensor<string, []>("op_39257_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_39257_end_0 = const()[name = tensor<string, []>("op_39257_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_39257_end_mask_0 = const()[name = tensor<string, []>("op_39257_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_39257_cast_fp16 = slice_by_index(begin = var_39257_begin_0, end = var_39257_end_0, end_mask = var_39257_end_mask_0, x = k_351_cast_fp16)[name = tensor<string, []>("op_39257_cast_fp16")];
            tensor<int32, [4]> var_39261_begin_0 = const()[name = tensor<string, []>("op_39261_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_39261_end_0 = const()[name = tensor<string, []>("op_39261_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_39261_end_mask_0 = const()[name = tensor<string, []>("op_39261_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_39261_cast_fp16 = slice_by_index(begin = var_39261_begin_0, end = var_39261_end_0, end_mask = var_39261_end_mask_0, x = k_351_cast_fp16)[name = tensor<string, []>("op_39261_cast_fp16")];
            tensor<int32, [4]> var_39265_begin_0 = const()[name = tensor<string, []>("op_39265_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_39265_end_0 = const()[name = tensor<string, []>("op_39265_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_39265_end_mask_0 = const()[name = tensor<string, []>("op_39265_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_39265_cast_fp16 = slice_by_index(begin = var_39265_begin_0, end = var_39265_end_0, end_mask = var_39265_end_mask_0, x = k_351_cast_fp16)[name = tensor<string, []>("op_39265_cast_fp16")];
            tensor<int32, [4]> var_39269_begin_0 = const()[name = tensor<string, []>("op_39269_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_39269_end_0 = const()[name = tensor<string, []>("op_39269_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_39269_end_mask_0 = const()[name = tensor<string, []>("op_39269_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_39269_cast_fp16 = slice_by_index(begin = var_39269_begin_0, end = var_39269_end_0, end_mask = var_39269_end_mask_0, x = k_351_cast_fp16)[name = tensor<string, []>("op_39269_cast_fp16")];
            tensor<int32, [4]> var_39273_begin_0 = const()[name = tensor<string, []>("op_39273_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_39273_end_0 = const()[name = tensor<string, []>("op_39273_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_39273_end_mask_0 = const()[name = tensor<string, []>("op_39273_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_39273_cast_fp16 = slice_by_index(begin = var_39273_begin_0, end = var_39273_end_0, end_mask = var_39273_end_mask_0, x = k_351_cast_fp16)[name = tensor<string, []>("op_39273_cast_fp16")];
            tensor<int32, [4]> var_39277_begin_0 = const()[name = tensor<string, []>("op_39277_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_39277_end_0 = const()[name = tensor<string, []>("op_39277_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_39277_end_mask_0 = const()[name = tensor<string, []>("op_39277_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_39277_cast_fp16 = slice_by_index(begin = var_39277_begin_0, end = var_39277_end_0, end_mask = var_39277_end_mask_0, x = k_351_cast_fp16)[name = tensor<string, []>("op_39277_cast_fp16")];
            tensor<int32, [4]> var_39281_begin_0 = const()[name = tensor<string, []>("op_39281_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_39281_end_0 = const()[name = tensor<string, []>("op_39281_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_39281_end_mask_0 = const()[name = tensor<string, []>("op_39281_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_39281_cast_fp16 = slice_by_index(begin = var_39281_begin_0, end = var_39281_end_0, end_mask = var_39281_end_mask_0, x = k_351_cast_fp16)[name = tensor<string, []>("op_39281_cast_fp16")];
            tensor<int32, [4]> var_39285_begin_0 = const()[name = tensor<string, []>("op_39285_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_39285_end_0 = const()[name = tensor<string, []>("op_39285_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_39285_end_mask_0 = const()[name = tensor<string, []>("op_39285_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_39285_cast_fp16 = slice_by_index(begin = var_39285_begin_0, end = var_39285_end_0, end_mask = var_39285_end_mask_0, x = k_351_cast_fp16)[name = tensor<string, []>("op_39285_cast_fp16")];
            tensor<int32, [4]> var_39289_begin_0 = const()[name = tensor<string, []>("op_39289_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_39289_end_0 = const()[name = tensor<string, []>("op_39289_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_39289_end_mask_0 = const()[name = tensor<string, []>("op_39289_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_39289_cast_fp16 = slice_by_index(begin = var_39289_begin_0, end = var_39289_end_0, end_mask = var_39289_end_mask_0, x = k_351_cast_fp16)[name = tensor<string, []>("op_39289_cast_fp16")];
            tensor<int32, [4]> var_39293_begin_0 = const()[name = tensor<string, []>("op_39293_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_39293_end_0 = const()[name = tensor<string, []>("op_39293_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_39293_end_mask_0 = const()[name = tensor<string, []>("op_39293_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_39293_cast_fp16 = slice_by_index(begin = var_39293_begin_0, end = var_39293_end_0, end_mask = var_39293_end_mask_0, x = k_351_cast_fp16)[name = tensor<string, []>("op_39293_cast_fp16")];
            tensor<int32, [4]> var_39297_begin_0 = const()[name = tensor<string, []>("op_39297_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_39297_end_0 = const()[name = tensor<string, []>("op_39297_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_39297_end_mask_0 = const()[name = tensor<string, []>("op_39297_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_39297_cast_fp16 = slice_by_index(begin = var_39297_begin_0, end = var_39297_end_0, end_mask = var_39297_end_mask_0, x = k_351_cast_fp16)[name = tensor<string, []>("op_39297_cast_fp16")];
            tensor<int32, [4]> var_39299_begin_0 = const()[name = tensor<string, []>("op_39299_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_39299_end_0 = const()[name = tensor<string, []>("op_39299_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_39299_end_mask_0 = const()[name = tensor<string, []>("op_39299_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_39299_cast_fp16 = slice_by_index(begin = var_39299_begin_0, end = var_39299_end_0, end_mask = var_39299_end_mask_0, x = v_175_cast_fp16)[name = tensor<string, []>("op_39299_cast_fp16")];
            tensor<int32, [4]> var_39303_begin_0 = const()[name = tensor<string, []>("op_39303_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_39303_end_0 = const()[name = tensor<string, []>("op_39303_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_39303_end_mask_0 = const()[name = tensor<string, []>("op_39303_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_39303_cast_fp16 = slice_by_index(begin = var_39303_begin_0, end = var_39303_end_0, end_mask = var_39303_end_mask_0, x = v_175_cast_fp16)[name = tensor<string, []>("op_39303_cast_fp16")];
            tensor<int32, [4]> var_39307_begin_0 = const()[name = tensor<string, []>("op_39307_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_39307_end_0 = const()[name = tensor<string, []>("op_39307_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_39307_end_mask_0 = const()[name = tensor<string, []>("op_39307_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_39307_cast_fp16 = slice_by_index(begin = var_39307_begin_0, end = var_39307_end_0, end_mask = var_39307_end_mask_0, x = v_175_cast_fp16)[name = tensor<string, []>("op_39307_cast_fp16")];
            tensor<int32, [4]> var_39311_begin_0 = const()[name = tensor<string, []>("op_39311_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_39311_end_0 = const()[name = tensor<string, []>("op_39311_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_39311_end_mask_0 = const()[name = tensor<string, []>("op_39311_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_39311_cast_fp16 = slice_by_index(begin = var_39311_begin_0, end = var_39311_end_0, end_mask = var_39311_end_mask_0, x = v_175_cast_fp16)[name = tensor<string, []>("op_39311_cast_fp16")];
            tensor<int32, [4]> var_39315_begin_0 = const()[name = tensor<string, []>("op_39315_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_39315_end_0 = const()[name = tensor<string, []>("op_39315_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_39315_end_mask_0 = const()[name = tensor<string, []>("op_39315_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_39315_cast_fp16 = slice_by_index(begin = var_39315_begin_0, end = var_39315_end_0, end_mask = var_39315_end_mask_0, x = v_175_cast_fp16)[name = tensor<string, []>("op_39315_cast_fp16")];
            tensor<int32, [4]> var_39319_begin_0 = const()[name = tensor<string, []>("op_39319_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_39319_end_0 = const()[name = tensor<string, []>("op_39319_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_39319_end_mask_0 = const()[name = tensor<string, []>("op_39319_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_39319_cast_fp16 = slice_by_index(begin = var_39319_begin_0, end = var_39319_end_0, end_mask = var_39319_end_mask_0, x = v_175_cast_fp16)[name = tensor<string, []>("op_39319_cast_fp16")];
            tensor<int32, [4]> var_39323_begin_0 = const()[name = tensor<string, []>("op_39323_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_39323_end_0 = const()[name = tensor<string, []>("op_39323_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_39323_end_mask_0 = const()[name = tensor<string, []>("op_39323_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_39323_cast_fp16 = slice_by_index(begin = var_39323_begin_0, end = var_39323_end_0, end_mask = var_39323_end_mask_0, x = v_175_cast_fp16)[name = tensor<string, []>("op_39323_cast_fp16")];
            tensor<int32, [4]> var_39327_begin_0 = const()[name = tensor<string, []>("op_39327_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_39327_end_0 = const()[name = tensor<string, []>("op_39327_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_39327_end_mask_0 = const()[name = tensor<string, []>("op_39327_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_39327_cast_fp16 = slice_by_index(begin = var_39327_begin_0, end = var_39327_end_0, end_mask = var_39327_end_mask_0, x = v_175_cast_fp16)[name = tensor<string, []>("op_39327_cast_fp16")];
            tensor<int32, [4]> var_39331_begin_0 = const()[name = tensor<string, []>("op_39331_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_39331_end_0 = const()[name = tensor<string, []>("op_39331_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_39331_end_mask_0 = const()[name = tensor<string, []>("op_39331_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_39331_cast_fp16 = slice_by_index(begin = var_39331_begin_0, end = var_39331_end_0, end_mask = var_39331_end_mask_0, x = v_175_cast_fp16)[name = tensor<string, []>("op_39331_cast_fp16")];
            tensor<int32, [4]> var_39335_begin_0 = const()[name = tensor<string, []>("op_39335_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_39335_end_0 = const()[name = tensor<string, []>("op_39335_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_39335_end_mask_0 = const()[name = tensor<string, []>("op_39335_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_39335_cast_fp16 = slice_by_index(begin = var_39335_begin_0, end = var_39335_end_0, end_mask = var_39335_end_mask_0, x = v_175_cast_fp16)[name = tensor<string, []>("op_39335_cast_fp16")];
            tensor<int32, [4]> var_39339_begin_0 = const()[name = tensor<string, []>("op_39339_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_39339_end_0 = const()[name = tensor<string, []>("op_39339_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_39339_end_mask_0 = const()[name = tensor<string, []>("op_39339_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_39339_cast_fp16 = slice_by_index(begin = var_39339_begin_0, end = var_39339_end_0, end_mask = var_39339_end_mask_0, x = v_175_cast_fp16)[name = tensor<string, []>("op_39339_cast_fp16")];
            tensor<int32, [4]> var_39343_begin_0 = const()[name = tensor<string, []>("op_39343_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_39343_end_0 = const()[name = tensor<string, []>("op_39343_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_39343_end_mask_0 = const()[name = tensor<string, []>("op_39343_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_39343_cast_fp16 = slice_by_index(begin = var_39343_begin_0, end = var_39343_end_0, end_mask = var_39343_end_mask_0, x = v_175_cast_fp16)[name = tensor<string, []>("op_39343_cast_fp16")];
            tensor<int32, [4]> var_39347_begin_0 = const()[name = tensor<string, []>("op_39347_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_39347_end_0 = const()[name = tensor<string, []>("op_39347_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_39347_end_mask_0 = const()[name = tensor<string, []>("op_39347_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_39347_cast_fp16 = slice_by_index(begin = var_39347_begin_0, end = var_39347_end_0, end_mask = var_39347_end_mask_0, x = v_175_cast_fp16)[name = tensor<string, []>("op_39347_cast_fp16")];
            tensor<int32, [4]> var_39351_begin_0 = const()[name = tensor<string, []>("op_39351_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_39351_end_0 = const()[name = tensor<string, []>("op_39351_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_39351_end_mask_0 = const()[name = tensor<string, []>("op_39351_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_39351_cast_fp16 = slice_by_index(begin = var_39351_begin_0, end = var_39351_end_0, end_mask = var_39351_end_mask_0, x = v_175_cast_fp16)[name = tensor<string, []>("op_39351_cast_fp16")];
            tensor<int32, [4]> var_39355_begin_0 = const()[name = tensor<string, []>("op_39355_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_39355_end_0 = const()[name = tensor<string, []>("op_39355_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_39355_end_mask_0 = const()[name = tensor<string, []>("op_39355_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_39355_cast_fp16 = slice_by_index(begin = var_39355_begin_0, end = var_39355_end_0, end_mask = var_39355_end_mask_0, x = v_175_cast_fp16)[name = tensor<string, []>("op_39355_cast_fp16")];
            tensor<int32, [4]> var_39359_begin_0 = const()[name = tensor<string, []>("op_39359_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_39359_end_0 = const()[name = tensor<string, []>("op_39359_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_39359_end_mask_0 = const()[name = tensor<string, []>("op_39359_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_39359_cast_fp16 = slice_by_index(begin = var_39359_begin_0, end = var_39359_end_0, end_mask = var_39359_end_mask_0, x = v_175_cast_fp16)[name = tensor<string, []>("op_39359_cast_fp16")];
            tensor<int32, [4]> var_39363_begin_0 = const()[name = tensor<string, []>("op_39363_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_39363_end_0 = const()[name = tensor<string, []>("op_39363_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_39363_end_mask_0 = const()[name = tensor<string, []>("op_39363_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_39363_cast_fp16 = slice_by_index(begin = var_39363_begin_0, end = var_39363_end_0, end_mask = var_39363_end_mask_0, x = v_175_cast_fp16)[name = tensor<string, []>("op_39363_cast_fp16")];
            tensor<int32, [4]> var_39367_begin_0 = const()[name = tensor<string, []>("op_39367_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_39367_end_0 = const()[name = tensor<string, []>("op_39367_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_39367_end_mask_0 = const()[name = tensor<string, []>("op_39367_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_39367_cast_fp16 = slice_by_index(begin = var_39367_begin_0, end = var_39367_end_0, end_mask = var_39367_end_mask_0, x = v_175_cast_fp16)[name = tensor<string, []>("op_39367_cast_fp16")];
            tensor<int32, [4]> var_39371_begin_0 = const()[name = tensor<string, []>("op_39371_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_39371_end_0 = const()[name = tensor<string, []>("op_39371_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_39371_end_mask_0 = const()[name = tensor<string, []>("op_39371_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_39371_cast_fp16 = slice_by_index(begin = var_39371_begin_0, end = var_39371_end_0, end_mask = var_39371_end_mask_0, x = v_175_cast_fp16)[name = tensor<string, []>("op_39371_cast_fp16")];
            tensor<int32, [4]> var_39375_begin_0 = const()[name = tensor<string, []>("op_39375_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_39375_end_0 = const()[name = tensor<string, []>("op_39375_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_39375_end_mask_0 = const()[name = tensor<string, []>("op_39375_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_39375_cast_fp16 = slice_by_index(begin = var_39375_begin_0, end = var_39375_end_0, end_mask = var_39375_end_mask_0, x = v_175_cast_fp16)[name = tensor<string, []>("op_39375_cast_fp16")];
            tensor<string, []> var_39379_equation_0 = const()[name = tensor<string, []>("op_39379_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_39379_cast_fp16 = einsum(equation = var_39379_equation_0, values = (var_39221_cast_fp16, var_39138_cast_fp16))[name = tensor<string, []>("op_39379_cast_fp16")];
            tensor<fp16, []> var_39380_to_fp16 = const()[name = tensor<string, []>("op_39380_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3321_cast_fp16 = mul(x = var_39379_cast_fp16, y = var_39380_to_fp16)[name = tensor<string, []>("aw_3321_cast_fp16")];
            tensor<string, []> var_39383_equation_0 = const()[name = tensor<string, []>("op_39383_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_39383_cast_fp16 = einsum(equation = var_39383_equation_0, values = (var_39225_cast_fp16, var_39142_cast_fp16))[name = tensor<string, []>("op_39383_cast_fp16")];
            tensor<fp16, []> var_39384_to_fp16 = const()[name = tensor<string, []>("op_39384_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3323_cast_fp16 = mul(x = var_39383_cast_fp16, y = var_39384_to_fp16)[name = tensor<string, []>("aw_3323_cast_fp16")];
            tensor<string, []> var_39387_equation_0 = const()[name = tensor<string, []>("op_39387_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_39387_cast_fp16 = einsum(equation = var_39387_equation_0, values = (var_39229_cast_fp16, var_39146_cast_fp16))[name = tensor<string, []>("op_39387_cast_fp16")];
            tensor<fp16, []> var_39388_to_fp16 = const()[name = tensor<string, []>("op_39388_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3325_cast_fp16 = mul(x = var_39387_cast_fp16, y = var_39388_to_fp16)[name = tensor<string, []>("aw_3325_cast_fp16")];
            tensor<string, []> var_39391_equation_0 = const()[name = tensor<string, []>("op_39391_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_39391_cast_fp16 = einsum(equation = var_39391_equation_0, values = (var_39233_cast_fp16, var_39150_cast_fp16))[name = tensor<string, []>("op_39391_cast_fp16")];
            tensor<fp16, []> var_39392_to_fp16 = const()[name = tensor<string, []>("op_39392_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3327_cast_fp16 = mul(x = var_39391_cast_fp16, y = var_39392_to_fp16)[name = tensor<string, []>("aw_3327_cast_fp16")];
            tensor<string, []> var_39395_equation_0 = const()[name = tensor<string, []>("op_39395_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_39395_cast_fp16 = einsum(equation = var_39395_equation_0, values = (var_39237_cast_fp16, var_39154_cast_fp16))[name = tensor<string, []>("op_39395_cast_fp16")];
            tensor<fp16, []> var_39396_to_fp16 = const()[name = tensor<string, []>("op_39396_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3329_cast_fp16 = mul(x = var_39395_cast_fp16, y = var_39396_to_fp16)[name = tensor<string, []>("aw_3329_cast_fp16")];
            tensor<string, []> var_39399_equation_0 = const()[name = tensor<string, []>("op_39399_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_39399_cast_fp16 = einsum(equation = var_39399_equation_0, values = (var_39241_cast_fp16, var_39158_cast_fp16))[name = tensor<string, []>("op_39399_cast_fp16")];
            tensor<fp16, []> var_39400_to_fp16 = const()[name = tensor<string, []>("op_39400_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3331_cast_fp16 = mul(x = var_39399_cast_fp16, y = var_39400_to_fp16)[name = tensor<string, []>("aw_3331_cast_fp16")];
            tensor<string, []> var_39403_equation_0 = const()[name = tensor<string, []>("op_39403_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_39403_cast_fp16 = einsum(equation = var_39403_equation_0, values = (var_39245_cast_fp16, var_39162_cast_fp16))[name = tensor<string, []>("op_39403_cast_fp16")];
            tensor<fp16, []> var_39404_to_fp16 = const()[name = tensor<string, []>("op_39404_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3333_cast_fp16 = mul(x = var_39403_cast_fp16, y = var_39404_to_fp16)[name = tensor<string, []>("aw_3333_cast_fp16")];
            tensor<string, []> var_39407_equation_0 = const()[name = tensor<string, []>("op_39407_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_39407_cast_fp16 = einsum(equation = var_39407_equation_0, values = (var_39249_cast_fp16, var_39166_cast_fp16))[name = tensor<string, []>("op_39407_cast_fp16")];
            tensor<fp16, []> var_39408_to_fp16 = const()[name = tensor<string, []>("op_39408_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3335_cast_fp16 = mul(x = var_39407_cast_fp16, y = var_39408_to_fp16)[name = tensor<string, []>("aw_3335_cast_fp16")];
            tensor<string, []> var_39411_equation_0 = const()[name = tensor<string, []>("op_39411_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_39411_cast_fp16 = einsum(equation = var_39411_equation_0, values = (var_39253_cast_fp16, var_39170_cast_fp16))[name = tensor<string, []>("op_39411_cast_fp16")];
            tensor<fp16, []> var_39412_to_fp16 = const()[name = tensor<string, []>("op_39412_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3337_cast_fp16 = mul(x = var_39411_cast_fp16, y = var_39412_to_fp16)[name = tensor<string, []>("aw_3337_cast_fp16")];
            tensor<string, []> var_39415_equation_0 = const()[name = tensor<string, []>("op_39415_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_39415_cast_fp16 = einsum(equation = var_39415_equation_0, values = (var_39257_cast_fp16, var_39174_cast_fp16))[name = tensor<string, []>("op_39415_cast_fp16")];
            tensor<fp16, []> var_39416_to_fp16 = const()[name = tensor<string, []>("op_39416_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3339_cast_fp16 = mul(x = var_39415_cast_fp16, y = var_39416_to_fp16)[name = tensor<string, []>("aw_3339_cast_fp16")];
            tensor<string, []> var_39419_equation_0 = const()[name = tensor<string, []>("op_39419_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_39419_cast_fp16 = einsum(equation = var_39419_equation_0, values = (var_39261_cast_fp16, var_39178_cast_fp16))[name = tensor<string, []>("op_39419_cast_fp16")];
            tensor<fp16, []> var_39420_to_fp16 = const()[name = tensor<string, []>("op_39420_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3341_cast_fp16 = mul(x = var_39419_cast_fp16, y = var_39420_to_fp16)[name = tensor<string, []>("aw_3341_cast_fp16")];
            tensor<string, []> var_39423_equation_0 = const()[name = tensor<string, []>("op_39423_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_39423_cast_fp16 = einsum(equation = var_39423_equation_0, values = (var_39265_cast_fp16, var_39182_cast_fp16))[name = tensor<string, []>("op_39423_cast_fp16")];
            tensor<fp16, []> var_39424_to_fp16 = const()[name = tensor<string, []>("op_39424_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3343_cast_fp16 = mul(x = var_39423_cast_fp16, y = var_39424_to_fp16)[name = tensor<string, []>("aw_3343_cast_fp16")];
            tensor<string, []> var_39427_equation_0 = const()[name = tensor<string, []>("op_39427_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_39427_cast_fp16 = einsum(equation = var_39427_equation_0, values = (var_39269_cast_fp16, var_39186_cast_fp16))[name = tensor<string, []>("op_39427_cast_fp16")];
            tensor<fp16, []> var_39428_to_fp16 = const()[name = tensor<string, []>("op_39428_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3345_cast_fp16 = mul(x = var_39427_cast_fp16, y = var_39428_to_fp16)[name = tensor<string, []>("aw_3345_cast_fp16")];
            tensor<string, []> var_39431_equation_0 = const()[name = tensor<string, []>("op_39431_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_39431_cast_fp16 = einsum(equation = var_39431_equation_0, values = (var_39273_cast_fp16, var_39190_cast_fp16))[name = tensor<string, []>("op_39431_cast_fp16")];
            tensor<fp16, []> var_39432_to_fp16 = const()[name = tensor<string, []>("op_39432_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3347_cast_fp16 = mul(x = var_39431_cast_fp16, y = var_39432_to_fp16)[name = tensor<string, []>("aw_3347_cast_fp16")];
            tensor<string, []> var_39435_equation_0 = const()[name = tensor<string, []>("op_39435_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_39435_cast_fp16 = einsum(equation = var_39435_equation_0, values = (var_39277_cast_fp16, var_39194_cast_fp16))[name = tensor<string, []>("op_39435_cast_fp16")];
            tensor<fp16, []> var_39436_to_fp16 = const()[name = tensor<string, []>("op_39436_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3349_cast_fp16 = mul(x = var_39435_cast_fp16, y = var_39436_to_fp16)[name = tensor<string, []>("aw_3349_cast_fp16")];
            tensor<string, []> var_39439_equation_0 = const()[name = tensor<string, []>("op_39439_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_39439_cast_fp16 = einsum(equation = var_39439_equation_0, values = (var_39281_cast_fp16, var_39198_cast_fp16))[name = tensor<string, []>("op_39439_cast_fp16")];
            tensor<fp16, []> var_39440_to_fp16 = const()[name = tensor<string, []>("op_39440_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3351_cast_fp16 = mul(x = var_39439_cast_fp16, y = var_39440_to_fp16)[name = tensor<string, []>("aw_3351_cast_fp16")];
            tensor<string, []> var_39443_equation_0 = const()[name = tensor<string, []>("op_39443_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_39443_cast_fp16 = einsum(equation = var_39443_equation_0, values = (var_39285_cast_fp16, var_39202_cast_fp16))[name = tensor<string, []>("op_39443_cast_fp16")];
            tensor<fp16, []> var_39444_to_fp16 = const()[name = tensor<string, []>("op_39444_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3353_cast_fp16 = mul(x = var_39443_cast_fp16, y = var_39444_to_fp16)[name = tensor<string, []>("aw_3353_cast_fp16")];
            tensor<string, []> var_39447_equation_0 = const()[name = tensor<string, []>("op_39447_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_39447_cast_fp16 = einsum(equation = var_39447_equation_0, values = (var_39289_cast_fp16, var_39206_cast_fp16))[name = tensor<string, []>("op_39447_cast_fp16")];
            tensor<fp16, []> var_39448_to_fp16 = const()[name = tensor<string, []>("op_39448_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3355_cast_fp16 = mul(x = var_39447_cast_fp16, y = var_39448_to_fp16)[name = tensor<string, []>("aw_3355_cast_fp16")];
            tensor<string, []> var_39451_equation_0 = const()[name = tensor<string, []>("op_39451_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_39451_cast_fp16 = einsum(equation = var_39451_equation_0, values = (var_39293_cast_fp16, var_39210_cast_fp16))[name = tensor<string, []>("op_39451_cast_fp16")];
            tensor<fp16, []> var_39452_to_fp16 = const()[name = tensor<string, []>("op_39452_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3357_cast_fp16 = mul(x = var_39451_cast_fp16, y = var_39452_to_fp16)[name = tensor<string, []>("aw_3357_cast_fp16")];
            tensor<string, []> var_39455_equation_0 = const()[name = tensor<string, []>("op_39455_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_39455_cast_fp16 = einsum(equation = var_39455_equation_0, values = (var_39297_cast_fp16, var_39214_cast_fp16))[name = tensor<string, []>("op_39455_cast_fp16")];
            tensor<fp16, []> var_39456_to_fp16 = const()[name = tensor<string, []>("op_39456_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3359_cast_fp16 = mul(x = var_39455_cast_fp16, y = var_39456_to_fp16)[name = tensor<string, []>("aw_3359_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_39458_cast_fp16 = softmax(axis = var_30355, x = aw_3321_cast_fp16)[name = tensor<string, []>("op_39458_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_39459_cast_fp16 = softmax(axis = var_30355, x = aw_3323_cast_fp16)[name = tensor<string, []>("op_39459_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_39460_cast_fp16 = softmax(axis = var_30355, x = aw_3325_cast_fp16)[name = tensor<string, []>("op_39460_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_39461_cast_fp16 = softmax(axis = var_30355, x = aw_3327_cast_fp16)[name = tensor<string, []>("op_39461_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_39462_cast_fp16 = softmax(axis = var_30355, x = aw_3329_cast_fp16)[name = tensor<string, []>("op_39462_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_39463_cast_fp16 = softmax(axis = var_30355, x = aw_3331_cast_fp16)[name = tensor<string, []>("op_39463_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_39464_cast_fp16 = softmax(axis = var_30355, x = aw_3333_cast_fp16)[name = tensor<string, []>("op_39464_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_39465_cast_fp16 = softmax(axis = var_30355, x = aw_3335_cast_fp16)[name = tensor<string, []>("op_39465_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_39466_cast_fp16 = softmax(axis = var_30355, x = aw_3337_cast_fp16)[name = tensor<string, []>("op_39466_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_39467_cast_fp16 = softmax(axis = var_30355, x = aw_3339_cast_fp16)[name = tensor<string, []>("op_39467_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_39468_cast_fp16 = softmax(axis = var_30355, x = aw_3341_cast_fp16)[name = tensor<string, []>("op_39468_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_39469_cast_fp16 = softmax(axis = var_30355, x = aw_3343_cast_fp16)[name = tensor<string, []>("op_39469_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_39470_cast_fp16 = softmax(axis = var_30355, x = aw_3345_cast_fp16)[name = tensor<string, []>("op_39470_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_39471_cast_fp16 = softmax(axis = var_30355, x = aw_3347_cast_fp16)[name = tensor<string, []>("op_39471_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_39472_cast_fp16 = softmax(axis = var_30355, x = aw_3349_cast_fp16)[name = tensor<string, []>("op_39472_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_39473_cast_fp16 = softmax(axis = var_30355, x = aw_3351_cast_fp16)[name = tensor<string, []>("op_39473_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_39474_cast_fp16 = softmax(axis = var_30355, x = aw_3353_cast_fp16)[name = tensor<string, []>("op_39474_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_39475_cast_fp16 = softmax(axis = var_30355, x = aw_3355_cast_fp16)[name = tensor<string, []>("op_39475_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_39476_cast_fp16 = softmax(axis = var_30355, x = aw_3357_cast_fp16)[name = tensor<string, []>("op_39476_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_39477_cast_fp16 = softmax(axis = var_30355, x = aw_3359_cast_fp16)[name = tensor<string, []>("op_39477_cast_fp16")];
            tensor<string, []> var_39479_equation_0 = const()[name = tensor<string, []>("op_39479_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39479_cast_fp16 = einsum(equation = var_39479_equation_0, values = (var_39299_cast_fp16, var_39458_cast_fp16))[name = tensor<string, []>("op_39479_cast_fp16")];
            tensor<string, []> var_39481_equation_0 = const()[name = tensor<string, []>("op_39481_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39481_cast_fp16 = einsum(equation = var_39481_equation_0, values = (var_39303_cast_fp16, var_39459_cast_fp16))[name = tensor<string, []>("op_39481_cast_fp16")];
            tensor<string, []> var_39483_equation_0 = const()[name = tensor<string, []>("op_39483_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39483_cast_fp16 = einsum(equation = var_39483_equation_0, values = (var_39307_cast_fp16, var_39460_cast_fp16))[name = tensor<string, []>("op_39483_cast_fp16")];
            tensor<string, []> var_39485_equation_0 = const()[name = tensor<string, []>("op_39485_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39485_cast_fp16 = einsum(equation = var_39485_equation_0, values = (var_39311_cast_fp16, var_39461_cast_fp16))[name = tensor<string, []>("op_39485_cast_fp16")];
            tensor<string, []> var_39487_equation_0 = const()[name = tensor<string, []>("op_39487_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39487_cast_fp16 = einsum(equation = var_39487_equation_0, values = (var_39315_cast_fp16, var_39462_cast_fp16))[name = tensor<string, []>("op_39487_cast_fp16")];
            tensor<string, []> var_39489_equation_0 = const()[name = tensor<string, []>("op_39489_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39489_cast_fp16 = einsum(equation = var_39489_equation_0, values = (var_39319_cast_fp16, var_39463_cast_fp16))[name = tensor<string, []>("op_39489_cast_fp16")];
            tensor<string, []> var_39491_equation_0 = const()[name = tensor<string, []>("op_39491_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39491_cast_fp16 = einsum(equation = var_39491_equation_0, values = (var_39323_cast_fp16, var_39464_cast_fp16))[name = tensor<string, []>("op_39491_cast_fp16")];
            tensor<string, []> var_39493_equation_0 = const()[name = tensor<string, []>("op_39493_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39493_cast_fp16 = einsum(equation = var_39493_equation_0, values = (var_39327_cast_fp16, var_39465_cast_fp16))[name = tensor<string, []>("op_39493_cast_fp16")];
            tensor<string, []> var_39495_equation_0 = const()[name = tensor<string, []>("op_39495_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39495_cast_fp16 = einsum(equation = var_39495_equation_0, values = (var_39331_cast_fp16, var_39466_cast_fp16))[name = tensor<string, []>("op_39495_cast_fp16")];
            tensor<string, []> var_39497_equation_0 = const()[name = tensor<string, []>("op_39497_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39497_cast_fp16 = einsum(equation = var_39497_equation_0, values = (var_39335_cast_fp16, var_39467_cast_fp16))[name = tensor<string, []>("op_39497_cast_fp16")];
            tensor<string, []> var_39499_equation_0 = const()[name = tensor<string, []>("op_39499_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39499_cast_fp16 = einsum(equation = var_39499_equation_0, values = (var_39339_cast_fp16, var_39468_cast_fp16))[name = tensor<string, []>("op_39499_cast_fp16")];
            tensor<string, []> var_39501_equation_0 = const()[name = tensor<string, []>("op_39501_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39501_cast_fp16 = einsum(equation = var_39501_equation_0, values = (var_39343_cast_fp16, var_39469_cast_fp16))[name = tensor<string, []>("op_39501_cast_fp16")];
            tensor<string, []> var_39503_equation_0 = const()[name = tensor<string, []>("op_39503_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39503_cast_fp16 = einsum(equation = var_39503_equation_0, values = (var_39347_cast_fp16, var_39470_cast_fp16))[name = tensor<string, []>("op_39503_cast_fp16")];
            tensor<string, []> var_39505_equation_0 = const()[name = tensor<string, []>("op_39505_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39505_cast_fp16 = einsum(equation = var_39505_equation_0, values = (var_39351_cast_fp16, var_39471_cast_fp16))[name = tensor<string, []>("op_39505_cast_fp16")];
            tensor<string, []> var_39507_equation_0 = const()[name = tensor<string, []>("op_39507_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39507_cast_fp16 = einsum(equation = var_39507_equation_0, values = (var_39355_cast_fp16, var_39472_cast_fp16))[name = tensor<string, []>("op_39507_cast_fp16")];
            tensor<string, []> var_39509_equation_0 = const()[name = tensor<string, []>("op_39509_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39509_cast_fp16 = einsum(equation = var_39509_equation_0, values = (var_39359_cast_fp16, var_39473_cast_fp16))[name = tensor<string, []>("op_39509_cast_fp16")];
            tensor<string, []> var_39511_equation_0 = const()[name = tensor<string, []>("op_39511_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39511_cast_fp16 = einsum(equation = var_39511_equation_0, values = (var_39363_cast_fp16, var_39474_cast_fp16))[name = tensor<string, []>("op_39511_cast_fp16")];
            tensor<string, []> var_39513_equation_0 = const()[name = tensor<string, []>("op_39513_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39513_cast_fp16 = einsum(equation = var_39513_equation_0, values = (var_39367_cast_fp16, var_39475_cast_fp16))[name = tensor<string, []>("op_39513_cast_fp16")];
            tensor<string, []> var_39515_equation_0 = const()[name = tensor<string, []>("op_39515_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39515_cast_fp16 = einsum(equation = var_39515_equation_0, values = (var_39371_cast_fp16, var_39476_cast_fp16))[name = tensor<string, []>("op_39515_cast_fp16")];
            tensor<string, []> var_39517_equation_0 = const()[name = tensor<string, []>("op_39517_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39517_cast_fp16 = einsum(equation = var_39517_equation_0, values = (var_39375_cast_fp16, var_39477_cast_fp16))[name = tensor<string, []>("op_39517_cast_fp16")];
            tensor<bool, []> input_513_interleave_0 = const()[name = tensor<string, []>("input_513_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_513_cast_fp16 = concat(axis = var_30355, interleave = input_513_interleave_0, values = (var_39479_cast_fp16, var_39481_cast_fp16, var_39483_cast_fp16, var_39485_cast_fp16, var_39487_cast_fp16, var_39489_cast_fp16, var_39491_cast_fp16, var_39493_cast_fp16, var_39495_cast_fp16, var_39497_cast_fp16, var_39499_cast_fp16, var_39501_cast_fp16, var_39503_cast_fp16, var_39505_cast_fp16, var_39507_cast_fp16, var_39509_cast_fp16, var_39511_cast_fp16, var_39513_cast_fp16, var_39515_cast_fp16, var_39517_cast_fp16))[name = tensor<string, []>("input_513_cast_fp16")];
            tensor<string, []> var_39527_pad_type_0 = const()[name = tensor<string, []>("op_39527_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_39527_strides_0 = const()[name = tensor<string, []>("op_39527_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_39527_pad_0 = const()[name = tensor<string, []>("op_39527_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_39527_dilations_0 = const()[name = tensor<string, []>("op_39527_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_39527_groups_0 = const()[name = tensor<string, []>("op_39527_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_9_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(248683776))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(249912640))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_9_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_9_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_9_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(249912832)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_39527_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_9_attn2_to_out_0_bias_to_fp16, dilations = var_39527_dilations_0, groups = var_39527_groups_0, pad = var_39527_pad_0, pad_type = var_39527_pad_type_0, strides = var_39527_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_9_attn2_to_out_0_weight_to_fp16_palettized, x = input_513_cast_fp16)[name = tensor<string, []>("op_39527_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_263_cast_fp16 = add(x = var_39527_cast_fp16, y = inputs_261_cast_fp16)[name = tensor<string, []>("inputs_263_cast_fp16")];
            tensor<int32, [1]> input_515_axes_0 = const()[name = tensor<string, []>("input_515_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> input_515_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_515_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(249915456)))];
            tensor<fp16, [1280]> input_515_beta_0_to_fp16 = const()[name = tensor<string, []>("input_515_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(249918080)))];
            tensor<fp16, []> var_39537_to_fp16 = const()[name = tensor<string, []>("op_39537_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> input_515_cast_fp16 = layer_norm(axes = input_515_axes_0, beta = input_515_beta_0_to_fp16, epsilon = var_39537_to_fp16, gamma = input_515_gamma_0_to_fp16, x = inputs_263_cast_fp16)[name = tensor<string, []>("input_515_cast_fp16")];
            tensor<string, []> var_39557_pad_type_0 = const()[name = tensor<string, []>("op_39557_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_39557_strides_0 = const()[name = tensor<string, []>("op_39557_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_39557_pad_0 = const()[name = tensor<string, []>("op_39557_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_39557_dilations_0 = const()[name = tensor<string, []>("op_39557_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_39557_groups_0 = const()[name = tensor<string, []>("op_39557_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_9_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [9830400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(249920704))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(259751168))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_9_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([10240, 1280, 1, 1])];
            tensor<fp16, [10240]> up_blocks_0_attentions_0_transformer_blocks_9_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_9_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(259751360)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_39557_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_9_ff_net_0_proj_bias_to_fp16, dilations = var_39557_dilations_0, groups = var_39557_groups_0, pad = var_39557_pad_0, pad_type = var_39557_pad_type_0, strides = var_39557_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_9_ff_net_0_proj_weight_to_fp16_palettized, x = input_515_cast_fp16)[name = tensor<string, []>("op_39557_cast_fp16")];
            tensor<int32, [2]> var_39558_split_sizes_0 = const()[name = tensor<string, []>("op_39558_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_39558_axis_0 = const()[name = tensor<string, []>("op_39558_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_39558_cast_fp16_0, tensor<fp16, [2, 5120, 1, 1024]> var_39558_cast_fp16_1 = split(axis = var_39558_axis_0, split_sizes = var_39558_split_sizes_0, x = var_39557_cast_fp16)[name = tensor<string, []>("op_39558_cast_fp16")];
            tensor<string, []> var_39560_mode_0 = const()[name = tensor<string, []>("op_39560_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_39560_cast_fp16 = gelu(mode = var_39560_mode_0, x = var_39558_cast_fp16_1)[name = tensor<string, []>("op_39560_cast_fp16")];
            tensor<fp16, [2, 5120, 1, 1024]> input_517_cast_fp16 = mul(x = var_39558_cast_fp16_0, y = var_39560_cast_fp16)[name = tensor<string, []>("input_517_cast_fp16")];
            tensor<string, []> var_39568_pad_type_0 = const()[name = tensor<string, []>("op_39568_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_39568_strides_0 = const()[name = tensor<string, []>("op_39568_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_39568_pad_0 = const()[name = tensor<string, []>("op_39568_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_39568_dilations_0 = const()[name = tensor<string, []>("op_39568_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_39568_groups_0 = const()[name = tensor<string, []>("op_39568_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_9_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [4915200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(259771904))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(264687168))), name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_9_ff_net_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 5120, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_9_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_9_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(264687360)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_39568_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_9_ff_net_2_bias_to_fp16, dilations = var_39568_dilations_0, groups = var_39568_groups_0, pad = var_39568_pad_0, pad_type = var_39568_pad_type_0, strides = var_39568_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_9_ff_net_2_weight_to_fp16_palettized, x = input_517_cast_fp16)[name = tensor<string, []>("op_39568_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_347_cast_fp16 = add(x = var_39568_cast_fp16, y = inputs_263_cast_fp16)[name = tensor<string, []>("hidden_states_347_cast_fp16")];
            tensor<int32, [4]> var_39570 = const()[name = tensor<string, []>("op_39570"), val = tensor<int32, [4]>([2, 1280, 32, 32])];
            tensor<fp16, [2, 1280, 32, 32]> input_519_cast_fp16 = reshape(shape = var_39570, x = hidden_states_347_cast_fp16)[name = tensor<string, []>("input_519_cast_fp16")];
            tensor<string, []> hidden_states_349_pad_type_0 = const()[name = tensor<string, []>("hidden_states_349_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> hidden_states_349_strides_0 = const()[name = tensor<string, []>("hidden_states_349_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_349_pad_0 = const()[name = tensor<string, []>("hidden_states_349_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_349_dilations_0 = const()[name = tensor<string, []>("hidden_states_349_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_349_groups_0 = const()[name = tensor<string, []>("hidden_states_349_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_proj_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(264689984))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(265918848))), name = tensor<string, []>("up_blocks_0_attentions_0_proj_out_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_proj_out_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_proj_out_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(265919040)))];
            tensor<fp16, [2, 1280, 32, 32]> hidden_states_349_cast_fp16 = conv(bias = up_blocks_0_attentions_0_proj_out_bias_to_fp16, dilations = hidden_states_349_dilations_0, groups = hidden_states_349_groups_0, pad = hidden_states_349_pad_0, pad_type = hidden_states_349_pad_type_0, strides = hidden_states_349_strides_0, weight = up_blocks_0_attentions_0_proj_out_weight_to_fp16_palettized, x = input_519_cast_fp16)[name = tensor<string, []>("hidden_states_349_cast_fp16")];
            tensor<fp16, [2, 1280, 32, 32]> hidden_states_351_cast_fp16 = add(x = hidden_states_349_cast_fp16, y = hidden_states_283_cast_fp16)[name = tensor<string, []>("hidden_states_351_cast_fp16")];
            tensor<bool, []> input_521_interleave_0 = const()[name = tensor<string, []>("input_521_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 32, 32]> cast_6 = cast(dtype = cast_6_dtype_0, x = input_213_cast_fp16)[name = tensor<string, []>("cast_3")];
            tensor<fp16, [2, 2560, 32, 32]> input_521_cast_fp16 = concat(axis = var_30355, interleave = input_521_interleave_0, values = (hidden_states_351_cast_fp16, cast_6))[name = tensor<string, []>("input_521_cast_fp16")];
            tensor<int32, [5]> reshape_96_shape_0 = const()[name = tensor<string, []>("reshape_96_shape_0"), val = tensor<int32, [5]>([2, 32, 80, 32, 32])];
            tensor<fp16, [2, 32, 80, 32, 32]> reshape_96_cast_fp16 = reshape(shape = reshape_96_shape_0, x = input_521_cast_fp16)[name = tensor<string, []>("reshape_96_cast_fp16")];
            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_fp16 = reduce_mean(axes = reduce_mean_72_axes_0, keep_dims = reduce_mean_72_keep_dims_0, x = reshape_96_cast_fp16)[name = tensor<string, []>("reduce_mean_72_cast_fp16")];
            tensor<fp16, [2, 32, 80, 32, 32]> sub_48_cast_fp16 = sub(x = reshape_96_cast_fp16, y = reduce_mean_72_cast_fp16)[name = tensor<string, []>("sub_48_cast_fp16")];
            tensor<fp16, [2, 32, 80, 32, 32]> square_24_cast_fp16 = square(x = sub_48_cast_fp16)[name = tensor<string, []>("square_24_cast_fp16")];
            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_fp16 = reduce_mean(axes = reduce_mean_74_axes_0, keep_dims = reduce_mean_74_keep_dims_0, x = square_24_cast_fp16)[name = tensor<string, []>("reduce_mean_74_cast_fp16")];
            tensor<fp16, []> add_48_y_0_to_fp16 = const()[name = tensor<string, []>("add_48_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_48_cast_fp16 = add(x = reduce_mean_74_cast_fp16, y = add_48_y_0_to_fp16)[name = tensor<string, []>("add_48_cast_fp16")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_24_cast_fp16 = sqrt(x = add_48_cast_fp16)[name = tensor<string, []>("sqrt_24_cast_fp16")];
            tensor<fp16, [2, 32, 80, 32, 32]> real_div_24_cast_fp16 = real_div(x = sub_48_cast_fp16, y = sqrt_24_cast_fp16)[name = tensor<string, []>("real_div_24_cast_fp16")];
            tensor<int32, [4]> reshape_97_shape_0 = const()[name = tensor<string, []>("reshape_97_shape_0"), val = tensor<int32, [4]>([2, 2560, 32, 32])];
            tensor<fp16, [2, 2560, 32, 32]> reshape_97_cast_fp16 = reshape(shape = reshape_97_shape_0, x = real_div_24_cast_fp16)[name = tensor<string, []>("reshape_97_cast_fp16")];
            tensor<fp16, [2560]> add_49_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_49_gamma_0_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(265921664)))];
            tensor<fp16, [2560]> add_49_beta_0_to_fp16 = const()[name = tensor<string, []>("add_49_beta_0_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(265926848)))];
            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, 2560, 32, 32]> add_49_cast_fp16 = 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_43_mean_0_to_fp16, variance = add_43_variance_0_to_fp16, x = reshape_97_cast_fp16)[name = tensor<string, []>("add_49_cast_fp16")];
            tensor<fp16, [2, 2560, 32, 32]> input_525_cast_fp16 = silu(x = add_49_cast_fp16)[name = tensor<string, []>("input_525_cast_fp16")];
            tensor<string, []> hidden_states_353_pad_type_0 = const()[name = tensor<string, []>("hidden_states_353_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_353_pad_0 = const()[name = tensor<string, []>("hidden_states_353_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> hidden_states_353_strides_0 = const()[name = tensor<string, []>("hidden_states_353_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> hidden_states_353_dilations_0 = const()[name = tensor<string, []>("hidden_states_353_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_353_groups_0 = const()[name = tensor<string, []>("hidden_states_353_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2560, 3, 3]> up_blocks_0_resnets_1_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [22118400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(265932032))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(288050496))), 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, []>(288050688)))];
            tensor<fp16, [2, 1280, 32, 32]> hidden_states_353_cast_fp16 = conv(bias = up_blocks_0_resnets_1_conv1_bias_to_fp16, dilations = hidden_states_353_dilations_0, groups = hidden_states_353_groups_0, pad = hidden_states_353_pad_0, pad_type = hidden_states_353_pad_type_0, strides = hidden_states_353_strides_0, weight = up_blocks_0_resnets_1_conv1_weight_to_fp16_palettized, x = input_525_cast_fp16)[name = tensor<string, []>("hidden_states_353_cast_fp16")];
            tensor<string, []> temb_19_pad_type_0 = const()[name = tensor<string, []>("temb_19_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> temb_19_strides_0 = const()[name = tensor<string, []>("temb_19_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> temb_19_pad_0 = const()[name = tensor<string, []>("temb_19_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> temb_19_dilations_0 = const()[name = tensor<string, []>("temb_19_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> temb_19_groups_0 = const()[name = tensor<string, []>("temb_19_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_resnets_1_time_emb_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(288053312))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(289282176))), 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, []>(289282368)))];
            tensor<fp16, [2, 1280, 1, 1]> cast_2 = cast(dtype = cast_2_dtype_0, x = input_21_cast_fp16)[name = tensor<string, []>("cast_10")];
            tensor<fp16, [2, 1280, 1, 1]> temb_19_cast_fp16 = conv(bias = up_blocks_0_resnets_1_time_emb_proj_bias_to_fp16, dilations = temb_19_dilations_0, groups = temb_19_groups_0, pad = temb_19_pad_0, pad_type = temb_19_pad_type_0, strides = temb_19_strides_0, weight = up_blocks_0_resnets_1_time_emb_proj_weight_to_fp16_palettized, x = cast_2)[name = tensor<string, []>("temb_19_cast_fp16")];
            tensor<fp16, [2, 1280, 32, 32]> input_529_cast_fp16 = add(x = hidden_states_353_cast_fp16, y = temb_19_cast_fp16)[name = tensor<string, []>("input_529_cast_fp16")];
            tensor<int32, [5]> reshape_100_shape_0 = const()[name = tensor<string, []>("reshape_100_shape_0"), val = tensor<int32, [5]>([2, 32, 40, 32, 32])];
            tensor<fp16, [2, 32, 40, 32, 32]> reshape_100_cast_fp16 = reshape(shape = reshape_100_shape_0, x = input_529_cast_fp16)[name = tensor<string, []>("reshape_100_cast_fp16")];
            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_fp16 = reduce_mean(axes = reduce_mean_75_axes_0, keep_dims = reduce_mean_75_keep_dims_0, x = reshape_100_cast_fp16)[name = tensor<string, []>("reduce_mean_75_cast_fp16")];
            tensor<fp16, [2, 32, 40, 32, 32]> sub_50_cast_fp16 = sub(x = reshape_100_cast_fp16, y = reduce_mean_75_cast_fp16)[name = tensor<string, []>("sub_50_cast_fp16")];
            tensor<fp16, [2, 32, 40, 32, 32]> square_25_cast_fp16 = square(x = sub_50_cast_fp16)[name = tensor<string, []>("square_25_cast_fp16")];
            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_fp16 = reduce_mean(axes = reduce_mean_77_axes_0, keep_dims = reduce_mean_77_keep_dims_0, x = square_25_cast_fp16)[name = tensor<string, []>("reduce_mean_77_cast_fp16")];
            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_fp16 = add(x = reduce_mean_77_cast_fp16, y = add_50_y_0_to_fp16)[name = tensor<string, []>("add_50_cast_fp16")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_25_cast_fp16 = sqrt(x = add_50_cast_fp16)[name = tensor<string, []>("sqrt_25_cast_fp16")];
            tensor<fp16, [2, 32, 40, 32, 32]> real_div_25_cast_fp16 = real_div(x = sub_50_cast_fp16, y = sqrt_25_cast_fp16)[name = tensor<string, []>("real_div_25_cast_fp16")];
            tensor<int32, [4]> reshape_101_shape_0 = const()[name = tensor<string, []>("reshape_101_shape_0"), val = tensor<int32, [4]>([2, 1280, 32, 32])];
            tensor<fp16, [2, 1280, 32, 32]> reshape_101_cast_fp16 = reshape(shape = reshape_101_shape_0, x = real_div_25_cast_fp16)[name = tensor<string, []>("reshape_101_cast_fp16")];
            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, []>(289284992)))];
            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, []>(289287616)))];
            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, 32, 32]> add_51_cast_fp16 = 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_23_mean_0_to_fp16, variance = add_23_variance_0_to_fp16, x = reshape_101_cast_fp16)[name = tensor<string, []>("add_51_cast_fp16")];
            tensor<fp16, [2, 1280, 32, 32]> input_533_cast_fp16 = silu(x = add_51_cast_fp16)[name = tensor<string, []>("input_533_cast_fp16")];
            tensor<string, []> hidden_states_355_pad_type_0 = const()[name = tensor<string, []>("hidden_states_355_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_355_pad_0 = const()[name = tensor<string, []>("hidden_states_355_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> hidden_states_355_strides_0 = const()[name = tensor<string, []>("hidden_states_355_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> hidden_states_355_dilations_0 = const()[name = tensor<string, []>("hidden_states_355_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_355_groups_0 = const()[name = tensor<string, []>("hidden_states_355_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 3, 3]> up_blocks_0_resnets_1_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [11059200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(289290240))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(300349504))), 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, []>(300349696)))];
            tensor<fp16, [2, 1280, 32, 32]> hidden_states_355_cast_fp16 = conv(bias = up_blocks_0_resnets_1_conv2_bias_to_fp16, dilations = hidden_states_355_dilations_0, groups = hidden_states_355_groups_0, pad = hidden_states_355_pad_0, pad_type = hidden_states_355_pad_type_0, strides = hidden_states_355_strides_0, weight = up_blocks_0_resnets_1_conv2_weight_to_fp16_palettized, x = input_533_cast_fp16)[name = tensor<string, []>("hidden_states_355_cast_fp16")];
            tensor<string, []> x_7_pad_type_0 = const()[name = tensor<string, []>("x_7_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> x_7_strides_0 = const()[name = tensor<string, []>("x_7_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> x_7_pad_0 = const()[name = tensor<string, []>("x_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> x_7_dilations_0 = const()[name = tensor<string, []>("x_7_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> x_7_groups_0 = const()[name = tensor<string, []>("x_7_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2560, 1, 1]> up_blocks_0_resnets_1_conv_shortcut_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2457600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(300352320))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(302809984))), 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, []>(302810176)))];
            tensor<fp16, [2, 1280, 32, 32]> x_7_cast_fp16 = conv(bias = up_blocks_0_resnets_1_conv_shortcut_bias_to_fp16, dilations = x_7_dilations_0, groups = x_7_groups_0, pad = x_7_pad_0, pad_type = x_7_pad_type_0, strides = x_7_strides_0, weight = up_blocks_0_resnets_1_conv_shortcut_weight_to_fp16_palettized, x = input_521_cast_fp16)[name = tensor<string, []>("x_7_cast_fp16")];
            tensor<fp16, [2, 1280, 32, 32]> hidden_states_357_cast_fp16 = add(x = x_7_cast_fp16, y = hidden_states_355_cast_fp16)[name = tensor<string, []>("hidden_states_357_cast_fp16")];
            tensor<int32, [5]> reshape_104_shape_0 = const()[name = tensor<string, []>("reshape_104_shape_0"), val = tensor<int32, [5]>([2, 32, 40, 32, 32])];
            tensor<fp16, [2, 32, 40, 32, 32]> reshape_104_cast_fp16 = reshape(shape = reshape_104_shape_0, x = hidden_states_357_cast_fp16)[name = tensor<string, []>("reshape_104_cast_fp16")];
            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_fp16 = reduce_mean(axes = reduce_mean_78_axes_0, keep_dims = reduce_mean_78_keep_dims_0, x = reshape_104_cast_fp16)[name = tensor<string, []>("reduce_mean_78_cast_fp16")];
            tensor<fp16, [2, 32, 40, 32, 32]> sub_52_cast_fp16 = sub(x = reshape_104_cast_fp16, y = reduce_mean_78_cast_fp16)[name = tensor<string, []>("sub_52_cast_fp16")];
            tensor<fp16, [2, 32, 40, 32, 32]> square_26_cast_fp16 = square(x = sub_52_cast_fp16)[name = tensor<string, []>("square_26_cast_fp16")];
            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_fp16 = reduce_mean(axes = reduce_mean_80_axes_0, keep_dims = reduce_mean_80_keep_dims_0, x = square_26_cast_fp16)[name = tensor<string, []>("reduce_mean_80_cast_fp16")];
            tensor<fp16, []> add_52_y_0_to_fp16 = const()[name = tensor<string, []>("add_52_y_0_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_52_cast_fp16 = add(x = reduce_mean_80_cast_fp16, y = add_52_y_0_to_fp16)[name = tensor<string, []>("add_52_cast_fp16")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_26_cast_fp16 = sqrt(x = add_52_cast_fp16)[name = tensor<string, []>("sqrt_26_cast_fp16")];
            tensor<fp16, [2, 32, 40, 32, 32]> real_div_26_cast_fp16 = real_div(x = sub_52_cast_fp16, y = sqrt_26_cast_fp16)[name = tensor<string, []>("real_div_26_cast_fp16")];
            tensor<int32, [4]> reshape_105_shape_0 = const()[name = tensor<string, []>("reshape_105_shape_0"), val = tensor<int32, [4]>([2, 1280, 32, 32])];
            tensor<fp16, [2, 1280, 32, 32]> reshape_105_cast_fp16 = reshape(shape = reshape_105_shape_0, x = real_div_26_cast_fp16)[name = tensor<string, []>("reshape_105_cast_fp16")];
            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, []>(302812800)))];
            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, []>(302815424)))];
            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, 32, 32]> add_53_cast_fp16 = 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_23_mean_0_to_fp16, variance = add_23_variance_0_to_fp16, x = reshape_105_cast_fp16)[name = tensor<string, []>("add_53_cast_fp16")];
            tensor<string, []> hidden_states_359_pad_type_0 = const()[name = tensor<string, []>("hidden_states_359_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> hidden_states_359_strides_0 = const()[name = tensor<string, []>("hidden_states_359_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_359_pad_0 = const()[name = tensor<string, []>("hidden_states_359_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_359_dilations_0 = const()[name = tensor<string, []>("hidden_states_359_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_359_groups_0 = const()[name = tensor<string, []>("hidden_states_359_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_proj_in_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(302818048))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(304046912))), name = tensor<string, []>("up_blocks_0_attentions_1_proj_in_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_proj_in_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_proj_in_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(304047104)))];
            tensor<fp16, [2, 1280, 32, 32]> hidden_states_359_cast_fp16 = conv(bias = up_blocks_0_attentions_1_proj_in_bias_to_fp16, dilations = hidden_states_359_dilations_0, groups = hidden_states_359_groups_0, pad = hidden_states_359_pad_0, pad_type = hidden_states_359_pad_type_0, strides = hidden_states_359_strides_0, weight = up_blocks_0_attentions_1_proj_in_weight_to_fp16_palettized, x = add_53_cast_fp16)[name = tensor<string, []>("hidden_states_359_cast_fp16")];
            tensor<int32, [4]> var_39668 = const()[name = tensor<string, []>("op_39668"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_265_cast_fp16 = reshape(shape = var_39668, x = hidden_states_359_cast_fp16)[name = tensor<string, []>("inputs_265_cast_fp16")];
            tensor<int32, [1]> hidden_states_361_axes_0 = const()[name = tensor<string, []>("hidden_states_361_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_361_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_361_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(304049728)))];
            tensor<fp16, [1280]> hidden_states_361_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_361_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(304052352)))];
            tensor<fp16, []> var_39684_to_fp16 = const()[name = tensor<string, []>("op_39684_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_361_cast_fp16 = layer_norm(axes = hidden_states_361_axes_0, beta = hidden_states_361_beta_0_to_fp16, epsilon = var_39684_to_fp16, gamma = hidden_states_361_gamma_0_to_fp16, x = inputs_265_cast_fp16)[name = tensor<string, []>("hidden_states_361_cast_fp16")];
            tensor<string, []> q_177_pad_type_0 = const()[name = tensor<string, []>("q_177_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_177_strides_0 = const()[name = tensor<string, []>("q_177_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_177_pad_0 = const()[name = tensor<string, []>("q_177_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_177_dilations_0 = const()[name = tensor<string, []>("q_177_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_177_groups_0 = const()[name = tensor<string, []>("q_177_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(304054976))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(305283840))), name = tensor<string, []>("up_blocks_0_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, 1024]> q_177_cast_fp16 = conv(dilations = q_177_dilations_0, groups = q_177_groups_0, pad = q_177_pad_0, pad_type = q_177_pad_type_0, strides = q_177_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_361_cast_fp16)[name = tensor<string, []>("q_177_cast_fp16")];
            tensor<string, []> k_353_pad_type_0 = const()[name = tensor<string, []>("k_353_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_353_strides_0 = const()[name = tensor<string, []>("k_353_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_353_pad_0 = const()[name = tensor<string, []>("k_353_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_353_dilations_0 = const()[name = tensor<string, []>("k_353_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_353_groups_0 = const()[name = tensor<string, []>("k_353_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(305284032))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(306512896))), name = tensor<string, []>("up_blocks_0_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, 1024]> k_353_cast_fp16 = conv(dilations = k_353_dilations_0, groups = k_353_groups_0, pad = k_353_pad_0, pad_type = k_353_pad_type_0, strides = k_353_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_361_cast_fp16)[name = tensor<string, []>("k_353_cast_fp16")];
            tensor<string, []> v_177_pad_type_0 = const()[name = tensor<string, []>("v_177_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_177_strides_0 = const()[name = tensor<string, []>("v_177_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_177_pad_0 = const()[name = tensor<string, []>("v_177_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_177_dilations_0 = const()[name = tensor<string, []>("v_177_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_177_groups_0 = const()[name = tensor<string, []>("v_177_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(306513088))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(307741952))), name = tensor<string, []>("up_blocks_0_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, 1024]> v_177_cast_fp16 = conv(dilations = v_177_dilations_0, groups = v_177_groups_0, pad = v_177_pad_0, pad_type = v_177_pad_type_0, strides = v_177_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_361_cast_fp16)[name = tensor<string, []>("v_177_cast_fp16")];
            tensor<int32, [4]> var_39717_begin_0 = const()[name = tensor<string, []>("op_39717_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_39717_end_0 = const()[name = tensor<string, []>("op_39717_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_39717_end_mask_0 = const()[name = tensor<string, []>("op_39717_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39717_cast_fp16 = slice_by_index(begin = var_39717_begin_0, end = var_39717_end_0, end_mask = var_39717_end_mask_0, x = q_177_cast_fp16)[name = tensor<string, []>("op_39717_cast_fp16")];
            tensor<int32, [4]> var_39721_begin_0 = const()[name = tensor<string, []>("op_39721_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_39721_end_0 = const()[name = tensor<string, []>("op_39721_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_39721_end_mask_0 = const()[name = tensor<string, []>("op_39721_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39721_cast_fp16 = slice_by_index(begin = var_39721_begin_0, end = var_39721_end_0, end_mask = var_39721_end_mask_0, x = q_177_cast_fp16)[name = tensor<string, []>("op_39721_cast_fp16")];
            tensor<int32, [4]> var_39725_begin_0 = const()[name = tensor<string, []>("op_39725_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_39725_end_0 = const()[name = tensor<string, []>("op_39725_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_39725_end_mask_0 = const()[name = tensor<string, []>("op_39725_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39725_cast_fp16 = slice_by_index(begin = var_39725_begin_0, end = var_39725_end_0, end_mask = var_39725_end_mask_0, x = q_177_cast_fp16)[name = tensor<string, []>("op_39725_cast_fp16")];
            tensor<int32, [4]> var_39729_begin_0 = const()[name = tensor<string, []>("op_39729_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_39729_end_0 = const()[name = tensor<string, []>("op_39729_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_39729_end_mask_0 = const()[name = tensor<string, []>("op_39729_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39729_cast_fp16 = slice_by_index(begin = var_39729_begin_0, end = var_39729_end_0, end_mask = var_39729_end_mask_0, x = q_177_cast_fp16)[name = tensor<string, []>("op_39729_cast_fp16")];
            tensor<int32, [4]> var_39733_begin_0 = const()[name = tensor<string, []>("op_39733_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_39733_end_0 = const()[name = tensor<string, []>("op_39733_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_39733_end_mask_0 = const()[name = tensor<string, []>("op_39733_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39733_cast_fp16 = slice_by_index(begin = var_39733_begin_0, end = var_39733_end_0, end_mask = var_39733_end_mask_0, x = q_177_cast_fp16)[name = tensor<string, []>("op_39733_cast_fp16")];
            tensor<int32, [4]> var_39737_begin_0 = const()[name = tensor<string, []>("op_39737_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_39737_end_0 = const()[name = tensor<string, []>("op_39737_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_39737_end_mask_0 = const()[name = tensor<string, []>("op_39737_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39737_cast_fp16 = slice_by_index(begin = var_39737_begin_0, end = var_39737_end_0, end_mask = var_39737_end_mask_0, x = q_177_cast_fp16)[name = tensor<string, []>("op_39737_cast_fp16")];
            tensor<int32, [4]> var_39741_begin_0 = const()[name = tensor<string, []>("op_39741_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_39741_end_0 = const()[name = tensor<string, []>("op_39741_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_39741_end_mask_0 = const()[name = tensor<string, []>("op_39741_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39741_cast_fp16 = slice_by_index(begin = var_39741_begin_0, end = var_39741_end_0, end_mask = var_39741_end_mask_0, x = q_177_cast_fp16)[name = tensor<string, []>("op_39741_cast_fp16")];
            tensor<int32, [4]> var_39745_begin_0 = const()[name = tensor<string, []>("op_39745_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_39745_end_0 = const()[name = tensor<string, []>("op_39745_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_39745_end_mask_0 = const()[name = tensor<string, []>("op_39745_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39745_cast_fp16 = slice_by_index(begin = var_39745_begin_0, end = var_39745_end_0, end_mask = var_39745_end_mask_0, x = q_177_cast_fp16)[name = tensor<string, []>("op_39745_cast_fp16")];
            tensor<int32, [4]> var_39749_begin_0 = const()[name = tensor<string, []>("op_39749_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_39749_end_0 = const()[name = tensor<string, []>("op_39749_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_39749_end_mask_0 = const()[name = tensor<string, []>("op_39749_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39749_cast_fp16 = slice_by_index(begin = var_39749_begin_0, end = var_39749_end_0, end_mask = var_39749_end_mask_0, x = q_177_cast_fp16)[name = tensor<string, []>("op_39749_cast_fp16")];
            tensor<int32, [4]> var_39753_begin_0 = const()[name = tensor<string, []>("op_39753_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_39753_end_0 = const()[name = tensor<string, []>("op_39753_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_39753_end_mask_0 = const()[name = tensor<string, []>("op_39753_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39753_cast_fp16 = slice_by_index(begin = var_39753_begin_0, end = var_39753_end_0, end_mask = var_39753_end_mask_0, x = q_177_cast_fp16)[name = tensor<string, []>("op_39753_cast_fp16")];
            tensor<int32, [4]> var_39757_begin_0 = const()[name = tensor<string, []>("op_39757_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_39757_end_0 = const()[name = tensor<string, []>("op_39757_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_39757_end_mask_0 = const()[name = tensor<string, []>("op_39757_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39757_cast_fp16 = slice_by_index(begin = var_39757_begin_0, end = var_39757_end_0, end_mask = var_39757_end_mask_0, x = q_177_cast_fp16)[name = tensor<string, []>("op_39757_cast_fp16")];
            tensor<int32, [4]> var_39761_begin_0 = const()[name = tensor<string, []>("op_39761_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_39761_end_0 = const()[name = tensor<string, []>("op_39761_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_39761_end_mask_0 = const()[name = tensor<string, []>("op_39761_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39761_cast_fp16 = slice_by_index(begin = var_39761_begin_0, end = var_39761_end_0, end_mask = var_39761_end_mask_0, x = q_177_cast_fp16)[name = tensor<string, []>("op_39761_cast_fp16")];
            tensor<int32, [4]> var_39765_begin_0 = const()[name = tensor<string, []>("op_39765_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_39765_end_0 = const()[name = tensor<string, []>("op_39765_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_39765_end_mask_0 = const()[name = tensor<string, []>("op_39765_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39765_cast_fp16 = slice_by_index(begin = var_39765_begin_0, end = var_39765_end_0, end_mask = var_39765_end_mask_0, x = q_177_cast_fp16)[name = tensor<string, []>("op_39765_cast_fp16")];
            tensor<int32, [4]> var_39769_begin_0 = const()[name = tensor<string, []>("op_39769_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_39769_end_0 = const()[name = tensor<string, []>("op_39769_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_39769_end_mask_0 = const()[name = tensor<string, []>("op_39769_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39769_cast_fp16 = slice_by_index(begin = var_39769_begin_0, end = var_39769_end_0, end_mask = var_39769_end_mask_0, x = q_177_cast_fp16)[name = tensor<string, []>("op_39769_cast_fp16")];
            tensor<int32, [4]> var_39773_begin_0 = const()[name = tensor<string, []>("op_39773_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_39773_end_0 = const()[name = tensor<string, []>("op_39773_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_39773_end_mask_0 = const()[name = tensor<string, []>("op_39773_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39773_cast_fp16 = slice_by_index(begin = var_39773_begin_0, end = var_39773_end_0, end_mask = var_39773_end_mask_0, x = q_177_cast_fp16)[name = tensor<string, []>("op_39773_cast_fp16")];
            tensor<int32, [4]> var_39777_begin_0 = const()[name = tensor<string, []>("op_39777_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_39777_end_0 = const()[name = tensor<string, []>("op_39777_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_39777_end_mask_0 = const()[name = tensor<string, []>("op_39777_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39777_cast_fp16 = slice_by_index(begin = var_39777_begin_0, end = var_39777_end_0, end_mask = var_39777_end_mask_0, x = q_177_cast_fp16)[name = tensor<string, []>("op_39777_cast_fp16")];
            tensor<int32, [4]> var_39781_begin_0 = const()[name = tensor<string, []>("op_39781_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_39781_end_0 = const()[name = tensor<string, []>("op_39781_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_39781_end_mask_0 = const()[name = tensor<string, []>("op_39781_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39781_cast_fp16 = slice_by_index(begin = var_39781_begin_0, end = var_39781_end_0, end_mask = var_39781_end_mask_0, x = q_177_cast_fp16)[name = tensor<string, []>("op_39781_cast_fp16")];
            tensor<int32, [4]> var_39785_begin_0 = const()[name = tensor<string, []>("op_39785_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_39785_end_0 = const()[name = tensor<string, []>("op_39785_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_39785_end_mask_0 = const()[name = tensor<string, []>("op_39785_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39785_cast_fp16 = slice_by_index(begin = var_39785_begin_0, end = var_39785_end_0, end_mask = var_39785_end_mask_0, x = q_177_cast_fp16)[name = tensor<string, []>("op_39785_cast_fp16")];
            tensor<int32, [4]> var_39789_begin_0 = const()[name = tensor<string, []>("op_39789_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_39789_end_0 = const()[name = tensor<string, []>("op_39789_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_39789_end_mask_0 = const()[name = tensor<string, []>("op_39789_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39789_cast_fp16 = slice_by_index(begin = var_39789_begin_0, end = var_39789_end_0, end_mask = var_39789_end_mask_0, x = q_177_cast_fp16)[name = tensor<string, []>("op_39789_cast_fp16")];
            tensor<int32, [4]> var_39793_begin_0 = const()[name = tensor<string, []>("op_39793_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_39793_end_0 = const()[name = tensor<string, []>("op_39793_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_39793_end_mask_0 = const()[name = tensor<string, []>("op_39793_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39793_cast_fp16 = slice_by_index(begin = var_39793_begin_0, end = var_39793_end_0, end_mask = var_39793_end_mask_0, x = q_177_cast_fp16)[name = tensor<string, []>("op_39793_cast_fp16")];
            tensor<int32, [4]> k_355_perm_0 = const()[name = tensor<string, []>("k_355_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_39800_begin_0 = const()[name = tensor<string, []>("op_39800_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_39800_end_0 = const()[name = tensor<string, []>("op_39800_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_39800_end_mask_0 = const()[name = tensor<string, []>("op_39800_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> k_355_cast_fp16 = transpose(perm = k_355_perm_0, x = k_353_cast_fp16)[name = tensor<string, []>("transpose_51")];
            tensor<fp16, [2, 1024, 1, 64]> var_39800_cast_fp16 = slice_by_index(begin = var_39800_begin_0, end = var_39800_end_0, end_mask = var_39800_end_mask_0, x = k_355_cast_fp16)[name = tensor<string, []>("op_39800_cast_fp16")];
            tensor<int32, [4]> var_39804_begin_0 = const()[name = tensor<string, []>("op_39804_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_39804_end_0 = const()[name = tensor<string, []>("op_39804_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_39804_end_mask_0 = const()[name = tensor<string, []>("op_39804_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_39804_cast_fp16 = slice_by_index(begin = var_39804_begin_0, end = var_39804_end_0, end_mask = var_39804_end_mask_0, x = k_355_cast_fp16)[name = tensor<string, []>("op_39804_cast_fp16")];
            tensor<int32, [4]> var_39808_begin_0 = const()[name = tensor<string, []>("op_39808_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_39808_end_0 = const()[name = tensor<string, []>("op_39808_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_39808_end_mask_0 = const()[name = tensor<string, []>("op_39808_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_39808_cast_fp16 = slice_by_index(begin = var_39808_begin_0, end = var_39808_end_0, end_mask = var_39808_end_mask_0, x = k_355_cast_fp16)[name = tensor<string, []>("op_39808_cast_fp16")];
            tensor<int32, [4]> var_39812_begin_0 = const()[name = tensor<string, []>("op_39812_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_39812_end_0 = const()[name = tensor<string, []>("op_39812_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_39812_end_mask_0 = const()[name = tensor<string, []>("op_39812_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_39812_cast_fp16 = slice_by_index(begin = var_39812_begin_0, end = var_39812_end_0, end_mask = var_39812_end_mask_0, x = k_355_cast_fp16)[name = tensor<string, []>("op_39812_cast_fp16")];
            tensor<int32, [4]> var_39816_begin_0 = const()[name = tensor<string, []>("op_39816_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_39816_end_0 = const()[name = tensor<string, []>("op_39816_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_39816_end_mask_0 = const()[name = tensor<string, []>("op_39816_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_39816_cast_fp16 = slice_by_index(begin = var_39816_begin_0, end = var_39816_end_0, end_mask = var_39816_end_mask_0, x = k_355_cast_fp16)[name = tensor<string, []>("op_39816_cast_fp16")];
            tensor<int32, [4]> var_39820_begin_0 = const()[name = tensor<string, []>("op_39820_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_39820_end_0 = const()[name = tensor<string, []>("op_39820_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_39820_end_mask_0 = const()[name = tensor<string, []>("op_39820_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_39820_cast_fp16 = slice_by_index(begin = var_39820_begin_0, end = var_39820_end_0, end_mask = var_39820_end_mask_0, x = k_355_cast_fp16)[name = tensor<string, []>("op_39820_cast_fp16")];
            tensor<int32, [4]> var_39824_begin_0 = const()[name = tensor<string, []>("op_39824_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_39824_end_0 = const()[name = tensor<string, []>("op_39824_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_39824_end_mask_0 = const()[name = tensor<string, []>("op_39824_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_39824_cast_fp16 = slice_by_index(begin = var_39824_begin_0, end = var_39824_end_0, end_mask = var_39824_end_mask_0, x = k_355_cast_fp16)[name = tensor<string, []>("op_39824_cast_fp16")];
            tensor<int32, [4]> var_39828_begin_0 = const()[name = tensor<string, []>("op_39828_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_39828_end_0 = const()[name = tensor<string, []>("op_39828_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_39828_end_mask_0 = const()[name = tensor<string, []>("op_39828_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_39828_cast_fp16 = slice_by_index(begin = var_39828_begin_0, end = var_39828_end_0, end_mask = var_39828_end_mask_0, x = k_355_cast_fp16)[name = tensor<string, []>("op_39828_cast_fp16")];
            tensor<int32, [4]> var_39832_begin_0 = const()[name = tensor<string, []>("op_39832_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_39832_end_0 = const()[name = tensor<string, []>("op_39832_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_39832_end_mask_0 = const()[name = tensor<string, []>("op_39832_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_39832_cast_fp16 = slice_by_index(begin = var_39832_begin_0, end = var_39832_end_0, end_mask = var_39832_end_mask_0, x = k_355_cast_fp16)[name = tensor<string, []>("op_39832_cast_fp16")];
            tensor<int32, [4]> var_39836_begin_0 = const()[name = tensor<string, []>("op_39836_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_39836_end_0 = const()[name = tensor<string, []>("op_39836_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_39836_end_mask_0 = const()[name = tensor<string, []>("op_39836_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_39836_cast_fp16 = slice_by_index(begin = var_39836_begin_0, end = var_39836_end_0, end_mask = var_39836_end_mask_0, x = k_355_cast_fp16)[name = tensor<string, []>("op_39836_cast_fp16")];
            tensor<int32, [4]> var_39840_begin_0 = const()[name = tensor<string, []>("op_39840_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_39840_end_0 = const()[name = tensor<string, []>("op_39840_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_39840_end_mask_0 = const()[name = tensor<string, []>("op_39840_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_39840_cast_fp16 = slice_by_index(begin = var_39840_begin_0, end = var_39840_end_0, end_mask = var_39840_end_mask_0, x = k_355_cast_fp16)[name = tensor<string, []>("op_39840_cast_fp16")];
            tensor<int32, [4]> var_39844_begin_0 = const()[name = tensor<string, []>("op_39844_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_39844_end_0 = const()[name = tensor<string, []>("op_39844_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_39844_end_mask_0 = const()[name = tensor<string, []>("op_39844_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_39844_cast_fp16 = slice_by_index(begin = var_39844_begin_0, end = var_39844_end_0, end_mask = var_39844_end_mask_0, x = k_355_cast_fp16)[name = tensor<string, []>("op_39844_cast_fp16")];
            tensor<int32, [4]> var_39848_begin_0 = const()[name = tensor<string, []>("op_39848_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_39848_end_0 = const()[name = tensor<string, []>("op_39848_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_39848_end_mask_0 = const()[name = tensor<string, []>("op_39848_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_39848_cast_fp16 = slice_by_index(begin = var_39848_begin_0, end = var_39848_end_0, end_mask = var_39848_end_mask_0, x = k_355_cast_fp16)[name = tensor<string, []>("op_39848_cast_fp16")];
            tensor<int32, [4]> var_39852_begin_0 = const()[name = tensor<string, []>("op_39852_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_39852_end_0 = const()[name = tensor<string, []>("op_39852_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_39852_end_mask_0 = const()[name = tensor<string, []>("op_39852_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_39852_cast_fp16 = slice_by_index(begin = var_39852_begin_0, end = var_39852_end_0, end_mask = var_39852_end_mask_0, x = k_355_cast_fp16)[name = tensor<string, []>("op_39852_cast_fp16")];
            tensor<int32, [4]> var_39856_begin_0 = const()[name = tensor<string, []>("op_39856_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_39856_end_0 = const()[name = tensor<string, []>("op_39856_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_39856_end_mask_0 = const()[name = tensor<string, []>("op_39856_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_39856_cast_fp16 = slice_by_index(begin = var_39856_begin_0, end = var_39856_end_0, end_mask = var_39856_end_mask_0, x = k_355_cast_fp16)[name = tensor<string, []>("op_39856_cast_fp16")];
            tensor<int32, [4]> var_39860_begin_0 = const()[name = tensor<string, []>("op_39860_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_39860_end_0 = const()[name = tensor<string, []>("op_39860_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_39860_end_mask_0 = const()[name = tensor<string, []>("op_39860_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_39860_cast_fp16 = slice_by_index(begin = var_39860_begin_0, end = var_39860_end_0, end_mask = var_39860_end_mask_0, x = k_355_cast_fp16)[name = tensor<string, []>("op_39860_cast_fp16")];
            tensor<int32, [4]> var_39864_begin_0 = const()[name = tensor<string, []>("op_39864_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_39864_end_0 = const()[name = tensor<string, []>("op_39864_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_39864_end_mask_0 = const()[name = tensor<string, []>("op_39864_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_39864_cast_fp16 = slice_by_index(begin = var_39864_begin_0, end = var_39864_end_0, end_mask = var_39864_end_mask_0, x = k_355_cast_fp16)[name = tensor<string, []>("op_39864_cast_fp16")];
            tensor<int32, [4]> var_39868_begin_0 = const()[name = tensor<string, []>("op_39868_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_39868_end_0 = const()[name = tensor<string, []>("op_39868_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_39868_end_mask_0 = const()[name = tensor<string, []>("op_39868_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_39868_cast_fp16 = slice_by_index(begin = var_39868_begin_0, end = var_39868_end_0, end_mask = var_39868_end_mask_0, x = k_355_cast_fp16)[name = tensor<string, []>("op_39868_cast_fp16")];
            tensor<int32, [4]> var_39872_begin_0 = const()[name = tensor<string, []>("op_39872_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_39872_end_0 = const()[name = tensor<string, []>("op_39872_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_39872_end_mask_0 = const()[name = tensor<string, []>("op_39872_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_39872_cast_fp16 = slice_by_index(begin = var_39872_begin_0, end = var_39872_end_0, end_mask = var_39872_end_mask_0, x = k_355_cast_fp16)[name = tensor<string, []>("op_39872_cast_fp16")];
            tensor<int32, [4]> var_39876_begin_0 = const()[name = tensor<string, []>("op_39876_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_39876_end_0 = const()[name = tensor<string, []>("op_39876_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_39876_end_mask_0 = const()[name = tensor<string, []>("op_39876_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_39876_cast_fp16 = slice_by_index(begin = var_39876_begin_0, end = var_39876_end_0, end_mask = var_39876_end_mask_0, x = k_355_cast_fp16)[name = tensor<string, []>("op_39876_cast_fp16")];
            tensor<int32, [4]> var_39878_begin_0 = const()[name = tensor<string, []>("op_39878_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_39878_end_0 = const()[name = tensor<string, []>("op_39878_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_39878_end_mask_0 = const()[name = tensor<string, []>("op_39878_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39878_cast_fp16 = slice_by_index(begin = var_39878_begin_0, end = var_39878_end_0, end_mask = var_39878_end_mask_0, x = v_177_cast_fp16)[name = tensor<string, []>("op_39878_cast_fp16")];
            tensor<int32, [4]> var_39882_begin_0 = const()[name = tensor<string, []>("op_39882_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_39882_end_0 = const()[name = tensor<string, []>("op_39882_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_39882_end_mask_0 = const()[name = tensor<string, []>("op_39882_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39882_cast_fp16 = slice_by_index(begin = var_39882_begin_0, end = var_39882_end_0, end_mask = var_39882_end_mask_0, x = v_177_cast_fp16)[name = tensor<string, []>("op_39882_cast_fp16")];
            tensor<int32, [4]> var_39886_begin_0 = const()[name = tensor<string, []>("op_39886_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_39886_end_0 = const()[name = tensor<string, []>("op_39886_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_39886_end_mask_0 = const()[name = tensor<string, []>("op_39886_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39886_cast_fp16 = slice_by_index(begin = var_39886_begin_0, end = var_39886_end_0, end_mask = var_39886_end_mask_0, x = v_177_cast_fp16)[name = tensor<string, []>("op_39886_cast_fp16")];
            tensor<int32, [4]> var_39890_begin_0 = const()[name = tensor<string, []>("op_39890_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_39890_end_0 = const()[name = tensor<string, []>("op_39890_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_39890_end_mask_0 = const()[name = tensor<string, []>("op_39890_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39890_cast_fp16 = slice_by_index(begin = var_39890_begin_0, end = var_39890_end_0, end_mask = var_39890_end_mask_0, x = v_177_cast_fp16)[name = tensor<string, []>("op_39890_cast_fp16")];
            tensor<int32, [4]> var_39894_begin_0 = const()[name = tensor<string, []>("op_39894_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_39894_end_0 = const()[name = tensor<string, []>("op_39894_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_39894_end_mask_0 = const()[name = tensor<string, []>("op_39894_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39894_cast_fp16 = slice_by_index(begin = var_39894_begin_0, end = var_39894_end_0, end_mask = var_39894_end_mask_0, x = v_177_cast_fp16)[name = tensor<string, []>("op_39894_cast_fp16")];
            tensor<int32, [4]> var_39898_begin_0 = const()[name = tensor<string, []>("op_39898_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_39898_end_0 = const()[name = tensor<string, []>("op_39898_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_39898_end_mask_0 = const()[name = tensor<string, []>("op_39898_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39898_cast_fp16 = slice_by_index(begin = var_39898_begin_0, end = var_39898_end_0, end_mask = var_39898_end_mask_0, x = v_177_cast_fp16)[name = tensor<string, []>("op_39898_cast_fp16")];
            tensor<int32, [4]> var_39902_begin_0 = const()[name = tensor<string, []>("op_39902_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_39902_end_0 = const()[name = tensor<string, []>("op_39902_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_39902_end_mask_0 = const()[name = tensor<string, []>("op_39902_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39902_cast_fp16 = slice_by_index(begin = var_39902_begin_0, end = var_39902_end_0, end_mask = var_39902_end_mask_0, x = v_177_cast_fp16)[name = tensor<string, []>("op_39902_cast_fp16")];
            tensor<int32, [4]> var_39906_begin_0 = const()[name = tensor<string, []>("op_39906_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_39906_end_0 = const()[name = tensor<string, []>("op_39906_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_39906_end_mask_0 = const()[name = tensor<string, []>("op_39906_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39906_cast_fp16 = slice_by_index(begin = var_39906_begin_0, end = var_39906_end_0, end_mask = var_39906_end_mask_0, x = v_177_cast_fp16)[name = tensor<string, []>("op_39906_cast_fp16")];
            tensor<int32, [4]> var_39910_begin_0 = const()[name = tensor<string, []>("op_39910_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_39910_end_0 = const()[name = tensor<string, []>("op_39910_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_39910_end_mask_0 = const()[name = tensor<string, []>("op_39910_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39910_cast_fp16 = slice_by_index(begin = var_39910_begin_0, end = var_39910_end_0, end_mask = var_39910_end_mask_0, x = v_177_cast_fp16)[name = tensor<string, []>("op_39910_cast_fp16")];
            tensor<int32, [4]> var_39914_begin_0 = const()[name = tensor<string, []>("op_39914_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_39914_end_0 = const()[name = tensor<string, []>("op_39914_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_39914_end_mask_0 = const()[name = tensor<string, []>("op_39914_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39914_cast_fp16 = slice_by_index(begin = var_39914_begin_0, end = var_39914_end_0, end_mask = var_39914_end_mask_0, x = v_177_cast_fp16)[name = tensor<string, []>("op_39914_cast_fp16")];
            tensor<int32, [4]> var_39918_begin_0 = const()[name = tensor<string, []>("op_39918_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_39918_end_0 = const()[name = tensor<string, []>("op_39918_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_39918_end_mask_0 = const()[name = tensor<string, []>("op_39918_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39918_cast_fp16 = slice_by_index(begin = var_39918_begin_0, end = var_39918_end_0, end_mask = var_39918_end_mask_0, x = v_177_cast_fp16)[name = tensor<string, []>("op_39918_cast_fp16")];
            tensor<int32, [4]> var_39922_begin_0 = const()[name = tensor<string, []>("op_39922_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_39922_end_0 = const()[name = tensor<string, []>("op_39922_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_39922_end_mask_0 = const()[name = tensor<string, []>("op_39922_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39922_cast_fp16 = slice_by_index(begin = var_39922_begin_0, end = var_39922_end_0, end_mask = var_39922_end_mask_0, x = v_177_cast_fp16)[name = tensor<string, []>("op_39922_cast_fp16")];
            tensor<int32, [4]> var_39926_begin_0 = const()[name = tensor<string, []>("op_39926_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_39926_end_0 = const()[name = tensor<string, []>("op_39926_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_39926_end_mask_0 = const()[name = tensor<string, []>("op_39926_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39926_cast_fp16 = slice_by_index(begin = var_39926_begin_0, end = var_39926_end_0, end_mask = var_39926_end_mask_0, x = v_177_cast_fp16)[name = tensor<string, []>("op_39926_cast_fp16")];
            tensor<int32, [4]> var_39930_begin_0 = const()[name = tensor<string, []>("op_39930_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_39930_end_0 = const()[name = tensor<string, []>("op_39930_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_39930_end_mask_0 = const()[name = tensor<string, []>("op_39930_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39930_cast_fp16 = slice_by_index(begin = var_39930_begin_0, end = var_39930_end_0, end_mask = var_39930_end_mask_0, x = v_177_cast_fp16)[name = tensor<string, []>("op_39930_cast_fp16")];
            tensor<int32, [4]> var_39934_begin_0 = const()[name = tensor<string, []>("op_39934_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_39934_end_0 = const()[name = tensor<string, []>("op_39934_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_39934_end_mask_0 = const()[name = tensor<string, []>("op_39934_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39934_cast_fp16 = slice_by_index(begin = var_39934_begin_0, end = var_39934_end_0, end_mask = var_39934_end_mask_0, x = v_177_cast_fp16)[name = tensor<string, []>("op_39934_cast_fp16")];
            tensor<int32, [4]> var_39938_begin_0 = const()[name = tensor<string, []>("op_39938_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_39938_end_0 = const()[name = tensor<string, []>("op_39938_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_39938_end_mask_0 = const()[name = tensor<string, []>("op_39938_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39938_cast_fp16 = slice_by_index(begin = var_39938_begin_0, end = var_39938_end_0, end_mask = var_39938_end_mask_0, x = v_177_cast_fp16)[name = tensor<string, []>("op_39938_cast_fp16")];
            tensor<int32, [4]> var_39942_begin_0 = const()[name = tensor<string, []>("op_39942_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_39942_end_0 = const()[name = tensor<string, []>("op_39942_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_39942_end_mask_0 = const()[name = tensor<string, []>("op_39942_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39942_cast_fp16 = slice_by_index(begin = var_39942_begin_0, end = var_39942_end_0, end_mask = var_39942_end_mask_0, x = v_177_cast_fp16)[name = tensor<string, []>("op_39942_cast_fp16")];
            tensor<int32, [4]> var_39946_begin_0 = const()[name = tensor<string, []>("op_39946_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_39946_end_0 = const()[name = tensor<string, []>("op_39946_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_39946_end_mask_0 = const()[name = tensor<string, []>("op_39946_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39946_cast_fp16 = slice_by_index(begin = var_39946_begin_0, end = var_39946_end_0, end_mask = var_39946_end_mask_0, x = v_177_cast_fp16)[name = tensor<string, []>("op_39946_cast_fp16")];
            tensor<int32, [4]> var_39950_begin_0 = const()[name = tensor<string, []>("op_39950_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_39950_end_0 = const()[name = tensor<string, []>("op_39950_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_39950_end_mask_0 = const()[name = tensor<string, []>("op_39950_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39950_cast_fp16 = slice_by_index(begin = var_39950_begin_0, end = var_39950_end_0, end_mask = var_39950_end_mask_0, x = v_177_cast_fp16)[name = tensor<string, []>("op_39950_cast_fp16")];
            tensor<int32, [4]> var_39954_begin_0 = const()[name = tensor<string, []>("op_39954_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_39954_end_0 = const()[name = tensor<string, []>("op_39954_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_39954_end_mask_0 = const()[name = tensor<string, []>("op_39954_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39954_cast_fp16 = slice_by_index(begin = var_39954_begin_0, end = var_39954_end_0, end_mask = var_39954_end_mask_0, x = v_177_cast_fp16)[name = tensor<string, []>("op_39954_cast_fp16")];
            tensor<string, []> var_39958_equation_0 = const()[name = tensor<string, []>("op_39958_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39958_cast_fp16 = einsum(equation = var_39958_equation_0, values = (var_39800_cast_fp16, var_39717_cast_fp16))[name = tensor<string, []>("op_39958_cast_fp16")];
            tensor<fp16, []> var_39959_to_fp16 = const()[name = tensor<string, []>("op_39959_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3361_cast_fp16 = mul(x = var_39958_cast_fp16, y = var_39959_to_fp16)[name = tensor<string, []>("aw_3361_cast_fp16")];
            tensor<string, []> var_39962_equation_0 = const()[name = tensor<string, []>("op_39962_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39962_cast_fp16 = einsum(equation = var_39962_equation_0, values = (var_39804_cast_fp16, var_39721_cast_fp16))[name = tensor<string, []>("op_39962_cast_fp16")];
            tensor<fp16, []> var_39963_to_fp16 = const()[name = tensor<string, []>("op_39963_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3363_cast_fp16 = mul(x = var_39962_cast_fp16, y = var_39963_to_fp16)[name = tensor<string, []>("aw_3363_cast_fp16")];
            tensor<string, []> var_39966_equation_0 = const()[name = tensor<string, []>("op_39966_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39966_cast_fp16 = einsum(equation = var_39966_equation_0, values = (var_39808_cast_fp16, var_39725_cast_fp16))[name = tensor<string, []>("op_39966_cast_fp16")];
            tensor<fp16, []> var_39967_to_fp16 = const()[name = tensor<string, []>("op_39967_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3365_cast_fp16 = mul(x = var_39966_cast_fp16, y = var_39967_to_fp16)[name = tensor<string, []>("aw_3365_cast_fp16")];
            tensor<string, []> var_39970_equation_0 = const()[name = tensor<string, []>("op_39970_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39970_cast_fp16 = einsum(equation = var_39970_equation_0, values = (var_39812_cast_fp16, var_39729_cast_fp16))[name = tensor<string, []>("op_39970_cast_fp16")];
            tensor<fp16, []> var_39971_to_fp16 = const()[name = tensor<string, []>("op_39971_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3367_cast_fp16 = mul(x = var_39970_cast_fp16, y = var_39971_to_fp16)[name = tensor<string, []>("aw_3367_cast_fp16")];
            tensor<string, []> var_39974_equation_0 = const()[name = tensor<string, []>("op_39974_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39974_cast_fp16 = einsum(equation = var_39974_equation_0, values = (var_39816_cast_fp16, var_39733_cast_fp16))[name = tensor<string, []>("op_39974_cast_fp16")];
            tensor<fp16, []> var_39975_to_fp16 = const()[name = tensor<string, []>("op_39975_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3369_cast_fp16 = mul(x = var_39974_cast_fp16, y = var_39975_to_fp16)[name = tensor<string, []>("aw_3369_cast_fp16")];
            tensor<string, []> var_39978_equation_0 = const()[name = tensor<string, []>("op_39978_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39978_cast_fp16 = einsum(equation = var_39978_equation_0, values = (var_39820_cast_fp16, var_39737_cast_fp16))[name = tensor<string, []>("op_39978_cast_fp16")];
            tensor<fp16, []> var_39979_to_fp16 = const()[name = tensor<string, []>("op_39979_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3371_cast_fp16 = mul(x = var_39978_cast_fp16, y = var_39979_to_fp16)[name = tensor<string, []>("aw_3371_cast_fp16")];
            tensor<string, []> var_39982_equation_0 = const()[name = tensor<string, []>("op_39982_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39982_cast_fp16 = einsum(equation = var_39982_equation_0, values = (var_39824_cast_fp16, var_39741_cast_fp16))[name = tensor<string, []>("op_39982_cast_fp16")];
            tensor<fp16, []> var_39983_to_fp16 = const()[name = tensor<string, []>("op_39983_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3373_cast_fp16 = mul(x = var_39982_cast_fp16, y = var_39983_to_fp16)[name = tensor<string, []>("aw_3373_cast_fp16")];
            tensor<string, []> var_39986_equation_0 = const()[name = tensor<string, []>("op_39986_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39986_cast_fp16 = einsum(equation = var_39986_equation_0, values = (var_39828_cast_fp16, var_39745_cast_fp16))[name = tensor<string, []>("op_39986_cast_fp16")];
            tensor<fp16, []> var_39987_to_fp16 = const()[name = tensor<string, []>("op_39987_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3375_cast_fp16 = mul(x = var_39986_cast_fp16, y = var_39987_to_fp16)[name = tensor<string, []>("aw_3375_cast_fp16")];
            tensor<string, []> var_39990_equation_0 = const()[name = tensor<string, []>("op_39990_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39990_cast_fp16 = einsum(equation = var_39990_equation_0, values = (var_39832_cast_fp16, var_39749_cast_fp16))[name = tensor<string, []>("op_39990_cast_fp16")];
            tensor<fp16, []> var_39991_to_fp16 = const()[name = tensor<string, []>("op_39991_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3377_cast_fp16 = mul(x = var_39990_cast_fp16, y = var_39991_to_fp16)[name = tensor<string, []>("aw_3377_cast_fp16")];
            tensor<string, []> var_39994_equation_0 = const()[name = tensor<string, []>("op_39994_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39994_cast_fp16 = einsum(equation = var_39994_equation_0, values = (var_39836_cast_fp16, var_39753_cast_fp16))[name = tensor<string, []>("op_39994_cast_fp16")];
            tensor<fp16, []> var_39995_to_fp16 = const()[name = tensor<string, []>("op_39995_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3379_cast_fp16 = mul(x = var_39994_cast_fp16, y = var_39995_to_fp16)[name = tensor<string, []>("aw_3379_cast_fp16")];
            tensor<string, []> var_39998_equation_0 = const()[name = tensor<string, []>("op_39998_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39998_cast_fp16 = einsum(equation = var_39998_equation_0, values = (var_39840_cast_fp16, var_39757_cast_fp16))[name = tensor<string, []>("op_39998_cast_fp16")];
            tensor<fp16, []> var_39999_to_fp16 = const()[name = tensor<string, []>("op_39999_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3381_cast_fp16 = mul(x = var_39998_cast_fp16, y = var_39999_to_fp16)[name = tensor<string, []>("aw_3381_cast_fp16")];
            tensor<string, []> var_40002_equation_0 = const()[name = tensor<string, []>("op_40002_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40002_cast_fp16 = einsum(equation = var_40002_equation_0, values = (var_39844_cast_fp16, var_39761_cast_fp16))[name = tensor<string, []>("op_40002_cast_fp16")];
            tensor<fp16, []> var_40003_to_fp16 = const()[name = tensor<string, []>("op_40003_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3383_cast_fp16 = mul(x = var_40002_cast_fp16, y = var_40003_to_fp16)[name = tensor<string, []>("aw_3383_cast_fp16")];
            tensor<string, []> var_40006_equation_0 = const()[name = tensor<string, []>("op_40006_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40006_cast_fp16 = einsum(equation = var_40006_equation_0, values = (var_39848_cast_fp16, var_39765_cast_fp16))[name = tensor<string, []>("op_40006_cast_fp16")];
            tensor<fp16, []> var_40007_to_fp16 = const()[name = tensor<string, []>("op_40007_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3385_cast_fp16 = mul(x = var_40006_cast_fp16, y = var_40007_to_fp16)[name = tensor<string, []>("aw_3385_cast_fp16")];
            tensor<string, []> var_40010_equation_0 = const()[name = tensor<string, []>("op_40010_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40010_cast_fp16 = einsum(equation = var_40010_equation_0, values = (var_39852_cast_fp16, var_39769_cast_fp16))[name = tensor<string, []>("op_40010_cast_fp16")];
            tensor<fp16, []> var_40011_to_fp16 = const()[name = tensor<string, []>("op_40011_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3387_cast_fp16 = mul(x = var_40010_cast_fp16, y = var_40011_to_fp16)[name = tensor<string, []>("aw_3387_cast_fp16")];
            tensor<string, []> var_40014_equation_0 = const()[name = tensor<string, []>("op_40014_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40014_cast_fp16 = einsum(equation = var_40014_equation_0, values = (var_39856_cast_fp16, var_39773_cast_fp16))[name = tensor<string, []>("op_40014_cast_fp16")];
            tensor<fp16, []> var_40015_to_fp16 = const()[name = tensor<string, []>("op_40015_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3389_cast_fp16 = mul(x = var_40014_cast_fp16, y = var_40015_to_fp16)[name = tensor<string, []>("aw_3389_cast_fp16")];
            tensor<string, []> var_40018_equation_0 = const()[name = tensor<string, []>("op_40018_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40018_cast_fp16 = einsum(equation = var_40018_equation_0, values = (var_39860_cast_fp16, var_39777_cast_fp16))[name = tensor<string, []>("op_40018_cast_fp16")];
            tensor<fp16, []> var_40019_to_fp16 = const()[name = tensor<string, []>("op_40019_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3391_cast_fp16 = mul(x = var_40018_cast_fp16, y = var_40019_to_fp16)[name = tensor<string, []>("aw_3391_cast_fp16")];
            tensor<string, []> var_40022_equation_0 = const()[name = tensor<string, []>("op_40022_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40022_cast_fp16 = einsum(equation = var_40022_equation_0, values = (var_39864_cast_fp16, var_39781_cast_fp16))[name = tensor<string, []>("op_40022_cast_fp16")];
            tensor<fp16, []> var_40023_to_fp16 = const()[name = tensor<string, []>("op_40023_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3393_cast_fp16 = mul(x = var_40022_cast_fp16, y = var_40023_to_fp16)[name = tensor<string, []>("aw_3393_cast_fp16")];
            tensor<string, []> var_40026_equation_0 = const()[name = tensor<string, []>("op_40026_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40026_cast_fp16 = einsum(equation = var_40026_equation_0, values = (var_39868_cast_fp16, var_39785_cast_fp16))[name = tensor<string, []>("op_40026_cast_fp16")];
            tensor<fp16, []> var_40027_to_fp16 = const()[name = tensor<string, []>("op_40027_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3395_cast_fp16 = mul(x = var_40026_cast_fp16, y = var_40027_to_fp16)[name = tensor<string, []>("aw_3395_cast_fp16")];
            tensor<string, []> var_40030_equation_0 = const()[name = tensor<string, []>("op_40030_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40030_cast_fp16 = einsum(equation = var_40030_equation_0, values = (var_39872_cast_fp16, var_39789_cast_fp16))[name = tensor<string, []>("op_40030_cast_fp16")];
            tensor<fp16, []> var_40031_to_fp16 = const()[name = tensor<string, []>("op_40031_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3397_cast_fp16 = mul(x = var_40030_cast_fp16, y = var_40031_to_fp16)[name = tensor<string, []>("aw_3397_cast_fp16")];
            tensor<string, []> var_40034_equation_0 = const()[name = tensor<string, []>("op_40034_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40034_cast_fp16 = einsum(equation = var_40034_equation_0, values = (var_39876_cast_fp16, var_39793_cast_fp16))[name = tensor<string, []>("op_40034_cast_fp16")];
            tensor<fp16, []> var_40035_to_fp16 = const()[name = tensor<string, []>("op_40035_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3399_cast_fp16 = mul(x = var_40034_cast_fp16, y = var_40035_to_fp16)[name = tensor<string, []>("aw_3399_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40037_cast_fp16 = softmax(axis = var_30355, x = aw_3361_cast_fp16)[name = tensor<string, []>("op_40037_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40038_cast_fp16 = softmax(axis = var_30355, x = aw_3363_cast_fp16)[name = tensor<string, []>("op_40038_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40039_cast_fp16 = softmax(axis = var_30355, x = aw_3365_cast_fp16)[name = tensor<string, []>("op_40039_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40040_cast_fp16 = softmax(axis = var_30355, x = aw_3367_cast_fp16)[name = tensor<string, []>("op_40040_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40041_cast_fp16 = softmax(axis = var_30355, x = aw_3369_cast_fp16)[name = tensor<string, []>("op_40041_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40042_cast_fp16 = softmax(axis = var_30355, x = aw_3371_cast_fp16)[name = tensor<string, []>("op_40042_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40043_cast_fp16 = softmax(axis = var_30355, x = aw_3373_cast_fp16)[name = tensor<string, []>("op_40043_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40044_cast_fp16 = softmax(axis = var_30355, x = aw_3375_cast_fp16)[name = tensor<string, []>("op_40044_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40045_cast_fp16 = softmax(axis = var_30355, x = aw_3377_cast_fp16)[name = tensor<string, []>("op_40045_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40046_cast_fp16 = softmax(axis = var_30355, x = aw_3379_cast_fp16)[name = tensor<string, []>("op_40046_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40047_cast_fp16 = softmax(axis = var_30355, x = aw_3381_cast_fp16)[name = tensor<string, []>("op_40047_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40048_cast_fp16 = softmax(axis = var_30355, x = aw_3383_cast_fp16)[name = tensor<string, []>("op_40048_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40049_cast_fp16 = softmax(axis = var_30355, x = aw_3385_cast_fp16)[name = tensor<string, []>("op_40049_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40050_cast_fp16 = softmax(axis = var_30355, x = aw_3387_cast_fp16)[name = tensor<string, []>("op_40050_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40051_cast_fp16 = softmax(axis = var_30355, x = aw_3389_cast_fp16)[name = tensor<string, []>("op_40051_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40052_cast_fp16 = softmax(axis = var_30355, x = aw_3391_cast_fp16)[name = tensor<string, []>("op_40052_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40053_cast_fp16 = softmax(axis = var_30355, x = aw_3393_cast_fp16)[name = tensor<string, []>("op_40053_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40054_cast_fp16 = softmax(axis = var_30355, x = aw_3395_cast_fp16)[name = tensor<string, []>("op_40054_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40055_cast_fp16 = softmax(axis = var_30355, x = aw_3397_cast_fp16)[name = tensor<string, []>("op_40055_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40056_cast_fp16 = softmax(axis = var_30355, x = aw_3399_cast_fp16)[name = tensor<string, []>("op_40056_cast_fp16")];
            tensor<string, []> var_40058_equation_0 = const()[name = tensor<string, []>("op_40058_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40058_cast_fp16 = einsum(equation = var_40058_equation_0, values = (var_39878_cast_fp16, var_40037_cast_fp16))[name = tensor<string, []>("op_40058_cast_fp16")];
            tensor<string, []> var_40060_equation_0 = const()[name = tensor<string, []>("op_40060_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40060_cast_fp16 = einsum(equation = var_40060_equation_0, values = (var_39882_cast_fp16, var_40038_cast_fp16))[name = tensor<string, []>("op_40060_cast_fp16")];
            tensor<string, []> var_40062_equation_0 = const()[name = tensor<string, []>("op_40062_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40062_cast_fp16 = einsum(equation = var_40062_equation_0, values = (var_39886_cast_fp16, var_40039_cast_fp16))[name = tensor<string, []>("op_40062_cast_fp16")];
            tensor<string, []> var_40064_equation_0 = const()[name = tensor<string, []>("op_40064_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40064_cast_fp16 = einsum(equation = var_40064_equation_0, values = (var_39890_cast_fp16, var_40040_cast_fp16))[name = tensor<string, []>("op_40064_cast_fp16")];
            tensor<string, []> var_40066_equation_0 = const()[name = tensor<string, []>("op_40066_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40066_cast_fp16 = einsum(equation = var_40066_equation_0, values = (var_39894_cast_fp16, var_40041_cast_fp16))[name = tensor<string, []>("op_40066_cast_fp16")];
            tensor<string, []> var_40068_equation_0 = const()[name = tensor<string, []>("op_40068_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40068_cast_fp16 = einsum(equation = var_40068_equation_0, values = (var_39898_cast_fp16, var_40042_cast_fp16))[name = tensor<string, []>("op_40068_cast_fp16")];
            tensor<string, []> var_40070_equation_0 = const()[name = tensor<string, []>("op_40070_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40070_cast_fp16 = einsum(equation = var_40070_equation_0, values = (var_39902_cast_fp16, var_40043_cast_fp16))[name = tensor<string, []>("op_40070_cast_fp16")];
            tensor<string, []> var_40072_equation_0 = const()[name = tensor<string, []>("op_40072_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40072_cast_fp16 = einsum(equation = var_40072_equation_0, values = (var_39906_cast_fp16, var_40044_cast_fp16))[name = tensor<string, []>("op_40072_cast_fp16")];
            tensor<string, []> var_40074_equation_0 = const()[name = tensor<string, []>("op_40074_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40074_cast_fp16 = einsum(equation = var_40074_equation_0, values = (var_39910_cast_fp16, var_40045_cast_fp16))[name = tensor<string, []>("op_40074_cast_fp16")];
            tensor<string, []> var_40076_equation_0 = const()[name = tensor<string, []>("op_40076_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40076_cast_fp16 = einsum(equation = var_40076_equation_0, values = (var_39914_cast_fp16, var_40046_cast_fp16))[name = tensor<string, []>("op_40076_cast_fp16")];
            tensor<string, []> var_40078_equation_0 = const()[name = tensor<string, []>("op_40078_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40078_cast_fp16 = einsum(equation = var_40078_equation_0, values = (var_39918_cast_fp16, var_40047_cast_fp16))[name = tensor<string, []>("op_40078_cast_fp16")];
            tensor<string, []> var_40080_equation_0 = const()[name = tensor<string, []>("op_40080_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40080_cast_fp16 = einsum(equation = var_40080_equation_0, values = (var_39922_cast_fp16, var_40048_cast_fp16))[name = tensor<string, []>("op_40080_cast_fp16")];
            tensor<string, []> var_40082_equation_0 = const()[name = tensor<string, []>("op_40082_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40082_cast_fp16 = einsum(equation = var_40082_equation_0, values = (var_39926_cast_fp16, var_40049_cast_fp16))[name = tensor<string, []>("op_40082_cast_fp16")];
            tensor<string, []> var_40084_equation_0 = const()[name = tensor<string, []>("op_40084_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40084_cast_fp16 = einsum(equation = var_40084_equation_0, values = (var_39930_cast_fp16, var_40050_cast_fp16))[name = tensor<string, []>("op_40084_cast_fp16")];
            tensor<string, []> var_40086_equation_0 = const()[name = tensor<string, []>("op_40086_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40086_cast_fp16 = einsum(equation = var_40086_equation_0, values = (var_39934_cast_fp16, var_40051_cast_fp16))[name = tensor<string, []>("op_40086_cast_fp16")];
            tensor<string, []> var_40088_equation_0 = const()[name = tensor<string, []>("op_40088_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40088_cast_fp16 = einsum(equation = var_40088_equation_0, values = (var_39938_cast_fp16, var_40052_cast_fp16))[name = tensor<string, []>("op_40088_cast_fp16")];
            tensor<string, []> var_40090_equation_0 = const()[name = tensor<string, []>("op_40090_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40090_cast_fp16 = einsum(equation = var_40090_equation_0, values = (var_39942_cast_fp16, var_40053_cast_fp16))[name = tensor<string, []>("op_40090_cast_fp16")];
            tensor<string, []> var_40092_equation_0 = const()[name = tensor<string, []>("op_40092_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40092_cast_fp16 = einsum(equation = var_40092_equation_0, values = (var_39946_cast_fp16, var_40054_cast_fp16))[name = tensor<string, []>("op_40092_cast_fp16")];
            tensor<string, []> var_40094_equation_0 = const()[name = tensor<string, []>("op_40094_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40094_cast_fp16 = einsum(equation = var_40094_equation_0, values = (var_39950_cast_fp16, var_40055_cast_fp16))[name = tensor<string, []>("op_40094_cast_fp16")];
            tensor<string, []> var_40096_equation_0 = const()[name = tensor<string, []>("op_40096_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40096_cast_fp16 = einsum(equation = var_40096_equation_0, values = (var_39954_cast_fp16, var_40056_cast_fp16))[name = tensor<string, []>("op_40096_cast_fp16")];
            tensor<bool, []> input_537_interleave_0 = const()[name = tensor<string, []>("input_537_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_537_cast_fp16 = concat(axis = var_30355, interleave = input_537_interleave_0, values = (var_40058_cast_fp16, var_40060_cast_fp16, var_40062_cast_fp16, var_40064_cast_fp16, var_40066_cast_fp16, var_40068_cast_fp16, var_40070_cast_fp16, var_40072_cast_fp16, var_40074_cast_fp16, var_40076_cast_fp16, var_40078_cast_fp16, var_40080_cast_fp16, var_40082_cast_fp16, var_40084_cast_fp16, var_40086_cast_fp16, var_40088_cast_fp16, var_40090_cast_fp16, var_40092_cast_fp16, var_40094_cast_fp16, var_40096_cast_fp16))[name = tensor<string, []>("input_537_cast_fp16")];
            tensor<string, []> var_40106_pad_type_0 = const()[name = tensor<string, []>("op_40106_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_40106_strides_0 = const()[name = tensor<string, []>("op_40106_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_40106_pad_0 = const()[name = tensor<string, []>("op_40106_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_40106_dilations_0 = const()[name = tensor<string, []>("op_40106_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_40106_groups_0 = const()[name = tensor<string, []>("op_40106_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(307742144))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(308971008))), name = tensor<string, []>("up_blocks_0_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_0_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_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, []>(308971200)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_40106_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_40106_dilations_0, groups = var_40106_groups_0, pad = var_40106_pad_0, pad_type = var_40106_pad_type_0, strides = var_40106_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized, x = input_537_cast_fp16)[name = tensor<string, []>("op_40106_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_267_cast_fp16 = add(x = var_40106_cast_fp16, y = inputs_265_cast_fp16)[name = tensor<string, []>("inputs_267_cast_fp16")];
            tensor<int32, [1]> hidden_states_363_axes_0 = const()[name = tensor<string, []>("hidden_states_363_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_363_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_363_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(308973824)))];
            tensor<fp16, [1280]> hidden_states_363_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_363_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(308976448)))];
            tensor<fp16, []> var_40116_to_fp16 = const()[name = tensor<string, []>("op_40116_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_363_cast_fp16 = layer_norm(axes = hidden_states_363_axes_0, beta = hidden_states_363_beta_0_to_fp16, epsilon = var_40116_to_fp16, gamma = hidden_states_363_gamma_0_to_fp16, x = inputs_267_cast_fp16)[name = tensor<string, []>("hidden_states_363_cast_fp16")];
            tensor<string, []> q_179_pad_type_0 = const()[name = tensor<string, []>("q_179_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_179_strides_0 = const()[name = tensor<string, []>("q_179_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_179_pad_0 = const()[name = tensor<string, []>("q_179_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_179_dilations_0 = const()[name = tensor<string, []>("q_179_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_179_groups_0 = const()[name = tensor<string, []>("q_179_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(308979072))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(310207936))), name = tensor<string, []>("up_blocks_0_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, 1024]> q_179_cast_fp16 = conv(dilations = q_179_dilations_0, groups = q_179_groups_0, pad = q_179_pad_0, pad_type = q_179_pad_type_0, strides = q_179_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_363_cast_fp16)[name = tensor<string, []>("q_179_cast_fp16")];
            tensor<string, []> k_357_pad_type_0 = const()[name = tensor<string, []>("k_357_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_357_strides_0 = const()[name = tensor<string, []>("k_357_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_357_pad_0 = const()[name = tensor<string, []>("k_357_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_357_dilations_0 = const()[name = tensor<string, []>("k_357_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_357_groups_0 = const()[name = tensor<string, []>("k_357_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(310208128))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(312174272))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> k_357_cast_fp16 = conv(dilations = k_357_dilations_0, groups = k_357_groups_0, pad = k_357_pad_0, pad_type = k_357_pad_type_0, strides = k_357_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("k_357_cast_fp16")];
            tensor<string, []> v_179_pad_type_0 = const()[name = tensor<string, []>("v_179_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_179_strides_0 = const()[name = tensor<string, []>("v_179_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_179_pad_0 = const()[name = tensor<string, []>("v_179_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_179_dilations_0 = const()[name = tensor<string, []>("v_179_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_179_groups_0 = const()[name = tensor<string, []>("v_179_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(312174464))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(314140608))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> v_179_cast_fp16 = conv(dilations = v_179_dilations_0, groups = v_179_groups_0, pad = v_179_pad_0, pad_type = v_179_pad_type_0, strides = v_179_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("v_179_cast_fp16")];
            tensor<int32, [4]> var_40149_begin_0 = const()[name = tensor<string, []>("op_40149_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_40149_end_0 = const()[name = tensor<string, []>("op_40149_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_40149_end_mask_0 = const()[name = tensor<string, []>("op_40149_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40149_cast_fp16 = slice_by_index(begin = var_40149_begin_0, end = var_40149_end_0, end_mask = var_40149_end_mask_0, x = q_179_cast_fp16)[name = tensor<string, []>("op_40149_cast_fp16")];
            tensor<int32, [4]> var_40153_begin_0 = const()[name = tensor<string, []>("op_40153_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_40153_end_0 = const()[name = tensor<string, []>("op_40153_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_40153_end_mask_0 = const()[name = tensor<string, []>("op_40153_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40153_cast_fp16 = slice_by_index(begin = var_40153_begin_0, end = var_40153_end_0, end_mask = var_40153_end_mask_0, x = q_179_cast_fp16)[name = tensor<string, []>("op_40153_cast_fp16")];
            tensor<int32, [4]> var_40157_begin_0 = const()[name = tensor<string, []>("op_40157_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_40157_end_0 = const()[name = tensor<string, []>("op_40157_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_40157_end_mask_0 = const()[name = tensor<string, []>("op_40157_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40157_cast_fp16 = slice_by_index(begin = var_40157_begin_0, end = var_40157_end_0, end_mask = var_40157_end_mask_0, x = q_179_cast_fp16)[name = tensor<string, []>("op_40157_cast_fp16")];
            tensor<int32, [4]> var_40161_begin_0 = const()[name = tensor<string, []>("op_40161_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_40161_end_0 = const()[name = tensor<string, []>("op_40161_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_40161_end_mask_0 = const()[name = tensor<string, []>("op_40161_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40161_cast_fp16 = slice_by_index(begin = var_40161_begin_0, end = var_40161_end_0, end_mask = var_40161_end_mask_0, x = q_179_cast_fp16)[name = tensor<string, []>("op_40161_cast_fp16")];
            tensor<int32, [4]> var_40165_begin_0 = const()[name = tensor<string, []>("op_40165_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_40165_end_0 = const()[name = tensor<string, []>("op_40165_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_40165_end_mask_0 = const()[name = tensor<string, []>("op_40165_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40165_cast_fp16 = slice_by_index(begin = var_40165_begin_0, end = var_40165_end_0, end_mask = var_40165_end_mask_0, x = q_179_cast_fp16)[name = tensor<string, []>("op_40165_cast_fp16")];
            tensor<int32, [4]> var_40169_begin_0 = const()[name = tensor<string, []>("op_40169_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_40169_end_0 = const()[name = tensor<string, []>("op_40169_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_40169_end_mask_0 = const()[name = tensor<string, []>("op_40169_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40169_cast_fp16 = slice_by_index(begin = var_40169_begin_0, end = var_40169_end_0, end_mask = var_40169_end_mask_0, x = q_179_cast_fp16)[name = tensor<string, []>("op_40169_cast_fp16")];
            tensor<int32, [4]> var_40173_begin_0 = const()[name = tensor<string, []>("op_40173_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_40173_end_0 = const()[name = tensor<string, []>("op_40173_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_40173_end_mask_0 = const()[name = tensor<string, []>("op_40173_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40173_cast_fp16 = slice_by_index(begin = var_40173_begin_0, end = var_40173_end_0, end_mask = var_40173_end_mask_0, x = q_179_cast_fp16)[name = tensor<string, []>("op_40173_cast_fp16")];
            tensor<int32, [4]> var_40177_begin_0 = const()[name = tensor<string, []>("op_40177_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_40177_end_0 = const()[name = tensor<string, []>("op_40177_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_40177_end_mask_0 = const()[name = tensor<string, []>("op_40177_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40177_cast_fp16 = slice_by_index(begin = var_40177_begin_0, end = var_40177_end_0, end_mask = var_40177_end_mask_0, x = q_179_cast_fp16)[name = tensor<string, []>("op_40177_cast_fp16")];
            tensor<int32, [4]> var_40181_begin_0 = const()[name = tensor<string, []>("op_40181_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_40181_end_0 = const()[name = tensor<string, []>("op_40181_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_40181_end_mask_0 = const()[name = tensor<string, []>("op_40181_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40181_cast_fp16 = slice_by_index(begin = var_40181_begin_0, end = var_40181_end_0, end_mask = var_40181_end_mask_0, x = q_179_cast_fp16)[name = tensor<string, []>("op_40181_cast_fp16")];
            tensor<int32, [4]> var_40185_begin_0 = const()[name = tensor<string, []>("op_40185_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_40185_end_0 = const()[name = tensor<string, []>("op_40185_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_40185_end_mask_0 = const()[name = tensor<string, []>("op_40185_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40185_cast_fp16 = slice_by_index(begin = var_40185_begin_0, end = var_40185_end_0, end_mask = var_40185_end_mask_0, x = q_179_cast_fp16)[name = tensor<string, []>("op_40185_cast_fp16")];
            tensor<int32, [4]> var_40189_begin_0 = const()[name = tensor<string, []>("op_40189_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_40189_end_0 = const()[name = tensor<string, []>("op_40189_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_40189_end_mask_0 = const()[name = tensor<string, []>("op_40189_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40189_cast_fp16 = slice_by_index(begin = var_40189_begin_0, end = var_40189_end_0, end_mask = var_40189_end_mask_0, x = q_179_cast_fp16)[name = tensor<string, []>("op_40189_cast_fp16")];
            tensor<int32, [4]> var_40193_begin_0 = const()[name = tensor<string, []>("op_40193_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_40193_end_0 = const()[name = tensor<string, []>("op_40193_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_40193_end_mask_0 = const()[name = tensor<string, []>("op_40193_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40193_cast_fp16 = slice_by_index(begin = var_40193_begin_0, end = var_40193_end_0, end_mask = var_40193_end_mask_0, x = q_179_cast_fp16)[name = tensor<string, []>("op_40193_cast_fp16")];
            tensor<int32, [4]> var_40197_begin_0 = const()[name = tensor<string, []>("op_40197_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_40197_end_0 = const()[name = tensor<string, []>("op_40197_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_40197_end_mask_0 = const()[name = tensor<string, []>("op_40197_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40197_cast_fp16 = slice_by_index(begin = var_40197_begin_0, end = var_40197_end_0, end_mask = var_40197_end_mask_0, x = q_179_cast_fp16)[name = tensor<string, []>("op_40197_cast_fp16")];
            tensor<int32, [4]> var_40201_begin_0 = const()[name = tensor<string, []>("op_40201_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_40201_end_0 = const()[name = tensor<string, []>("op_40201_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_40201_end_mask_0 = const()[name = tensor<string, []>("op_40201_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40201_cast_fp16 = slice_by_index(begin = var_40201_begin_0, end = var_40201_end_0, end_mask = var_40201_end_mask_0, x = q_179_cast_fp16)[name = tensor<string, []>("op_40201_cast_fp16")];
            tensor<int32, [4]> var_40205_begin_0 = const()[name = tensor<string, []>("op_40205_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_40205_end_0 = const()[name = tensor<string, []>("op_40205_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_40205_end_mask_0 = const()[name = tensor<string, []>("op_40205_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40205_cast_fp16 = slice_by_index(begin = var_40205_begin_0, end = var_40205_end_0, end_mask = var_40205_end_mask_0, x = q_179_cast_fp16)[name = tensor<string, []>("op_40205_cast_fp16")];
            tensor<int32, [4]> var_40209_begin_0 = const()[name = tensor<string, []>("op_40209_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_40209_end_0 = const()[name = tensor<string, []>("op_40209_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_40209_end_mask_0 = const()[name = tensor<string, []>("op_40209_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40209_cast_fp16 = slice_by_index(begin = var_40209_begin_0, end = var_40209_end_0, end_mask = var_40209_end_mask_0, x = q_179_cast_fp16)[name = tensor<string, []>("op_40209_cast_fp16")];
            tensor<int32, [4]> var_40213_begin_0 = const()[name = tensor<string, []>("op_40213_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_40213_end_0 = const()[name = tensor<string, []>("op_40213_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_40213_end_mask_0 = const()[name = tensor<string, []>("op_40213_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40213_cast_fp16 = slice_by_index(begin = var_40213_begin_0, end = var_40213_end_0, end_mask = var_40213_end_mask_0, x = q_179_cast_fp16)[name = tensor<string, []>("op_40213_cast_fp16")];
            tensor<int32, [4]> var_40217_begin_0 = const()[name = tensor<string, []>("op_40217_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_40217_end_0 = const()[name = tensor<string, []>("op_40217_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_40217_end_mask_0 = const()[name = tensor<string, []>("op_40217_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40217_cast_fp16 = slice_by_index(begin = var_40217_begin_0, end = var_40217_end_0, end_mask = var_40217_end_mask_0, x = q_179_cast_fp16)[name = tensor<string, []>("op_40217_cast_fp16")];
            tensor<int32, [4]> var_40221_begin_0 = const()[name = tensor<string, []>("op_40221_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_40221_end_0 = const()[name = tensor<string, []>("op_40221_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_40221_end_mask_0 = const()[name = tensor<string, []>("op_40221_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40221_cast_fp16 = slice_by_index(begin = var_40221_begin_0, end = var_40221_end_0, end_mask = var_40221_end_mask_0, x = q_179_cast_fp16)[name = tensor<string, []>("op_40221_cast_fp16")];
            tensor<int32, [4]> var_40225_begin_0 = const()[name = tensor<string, []>("op_40225_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_40225_end_0 = const()[name = tensor<string, []>("op_40225_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_40225_end_mask_0 = const()[name = tensor<string, []>("op_40225_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40225_cast_fp16 = slice_by_index(begin = var_40225_begin_0, end = var_40225_end_0, end_mask = var_40225_end_mask_0, x = q_179_cast_fp16)[name = tensor<string, []>("op_40225_cast_fp16")];
            tensor<int32, [4]> k_359_perm_0 = const()[name = tensor<string, []>("k_359_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_40232_begin_0 = const()[name = tensor<string, []>("op_40232_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_40232_end_0 = const()[name = tensor<string, []>("op_40232_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_40232_end_mask_0 = const()[name = tensor<string, []>("op_40232_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> k_359_cast_fp16 = transpose(perm = k_359_perm_0, x = k_357_cast_fp16)[name = tensor<string, []>("transpose_50")];
            tensor<fp16, [2, 77, 1, 64]> var_40232_cast_fp16 = slice_by_index(begin = var_40232_begin_0, end = var_40232_end_0, end_mask = var_40232_end_mask_0, x = k_359_cast_fp16)[name = tensor<string, []>("op_40232_cast_fp16")];
            tensor<int32, [4]> var_40236_begin_0 = const()[name = tensor<string, []>("op_40236_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_40236_end_0 = const()[name = tensor<string, []>("op_40236_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_40236_end_mask_0 = const()[name = tensor<string, []>("op_40236_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_40236_cast_fp16 = slice_by_index(begin = var_40236_begin_0, end = var_40236_end_0, end_mask = var_40236_end_mask_0, x = k_359_cast_fp16)[name = tensor<string, []>("op_40236_cast_fp16")];
            tensor<int32, [4]> var_40240_begin_0 = const()[name = tensor<string, []>("op_40240_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_40240_end_0 = const()[name = tensor<string, []>("op_40240_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_40240_end_mask_0 = const()[name = tensor<string, []>("op_40240_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_40240_cast_fp16 = slice_by_index(begin = var_40240_begin_0, end = var_40240_end_0, end_mask = var_40240_end_mask_0, x = k_359_cast_fp16)[name = tensor<string, []>("op_40240_cast_fp16")];
            tensor<int32, [4]> var_40244_begin_0 = const()[name = tensor<string, []>("op_40244_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_40244_end_0 = const()[name = tensor<string, []>("op_40244_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_40244_end_mask_0 = const()[name = tensor<string, []>("op_40244_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_40244_cast_fp16 = slice_by_index(begin = var_40244_begin_0, end = var_40244_end_0, end_mask = var_40244_end_mask_0, x = k_359_cast_fp16)[name = tensor<string, []>("op_40244_cast_fp16")];
            tensor<int32, [4]> var_40248_begin_0 = const()[name = tensor<string, []>("op_40248_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_40248_end_0 = const()[name = tensor<string, []>("op_40248_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_40248_end_mask_0 = const()[name = tensor<string, []>("op_40248_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_40248_cast_fp16 = slice_by_index(begin = var_40248_begin_0, end = var_40248_end_0, end_mask = var_40248_end_mask_0, x = k_359_cast_fp16)[name = tensor<string, []>("op_40248_cast_fp16")];
            tensor<int32, [4]> var_40252_begin_0 = const()[name = tensor<string, []>("op_40252_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_40252_end_0 = const()[name = tensor<string, []>("op_40252_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_40252_end_mask_0 = const()[name = tensor<string, []>("op_40252_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_40252_cast_fp16 = slice_by_index(begin = var_40252_begin_0, end = var_40252_end_0, end_mask = var_40252_end_mask_0, x = k_359_cast_fp16)[name = tensor<string, []>("op_40252_cast_fp16")];
            tensor<int32, [4]> var_40256_begin_0 = const()[name = tensor<string, []>("op_40256_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_40256_end_0 = const()[name = tensor<string, []>("op_40256_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_40256_end_mask_0 = const()[name = tensor<string, []>("op_40256_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_40256_cast_fp16 = slice_by_index(begin = var_40256_begin_0, end = var_40256_end_0, end_mask = var_40256_end_mask_0, x = k_359_cast_fp16)[name = tensor<string, []>("op_40256_cast_fp16")];
            tensor<int32, [4]> var_40260_begin_0 = const()[name = tensor<string, []>("op_40260_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_40260_end_0 = const()[name = tensor<string, []>("op_40260_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_40260_end_mask_0 = const()[name = tensor<string, []>("op_40260_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_40260_cast_fp16 = slice_by_index(begin = var_40260_begin_0, end = var_40260_end_0, end_mask = var_40260_end_mask_0, x = k_359_cast_fp16)[name = tensor<string, []>("op_40260_cast_fp16")];
            tensor<int32, [4]> var_40264_begin_0 = const()[name = tensor<string, []>("op_40264_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_40264_end_0 = const()[name = tensor<string, []>("op_40264_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_40264_end_mask_0 = const()[name = tensor<string, []>("op_40264_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_40264_cast_fp16 = slice_by_index(begin = var_40264_begin_0, end = var_40264_end_0, end_mask = var_40264_end_mask_0, x = k_359_cast_fp16)[name = tensor<string, []>("op_40264_cast_fp16")];
            tensor<int32, [4]> var_40268_begin_0 = const()[name = tensor<string, []>("op_40268_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_40268_end_0 = const()[name = tensor<string, []>("op_40268_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_40268_end_mask_0 = const()[name = tensor<string, []>("op_40268_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_40268_cast_fp16 = slice_by_index(begin = var_40268_begin_0, end = var_40268_end_0, end_mask = var_40268_end_mask_0, x = k_359_cast_fp16)[name = tensor<string, []>("op_40268_cast_fp16")];
            tensor<int32, [4]> var_40272_begin_0 = const()[name = tensor<string, []>("op_40272_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_40272_end_0 = const()[name = tensor<string, []>("op_40272_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_40272_end_mask_0 = const()[name = tensor<string, []>("op_40272_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_40272_cast_fp16 = slice_by_index(begin = var_40272_begin_0, end = var_40272_end_0, end_mask = var_40272_end_mask_0, x = k_359_cast_fp16)[name = tensor<string, []>("op_40272_cast_fp16")];
            tensor<int32, [4]> var_40276_begin_0 = const()[name = tensor<string, []>("op_40276_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_40276_end_0 = const()[name = tensor<string, []>("op_40276_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_40276_end_mask_0 = const()[name = tensor<string, []>("op_40276_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_40276_cast_fp16 = slice_by_index(begin = var_40276_begin_0, end = var_40276_end_0, end_mask = var_40276_end_mask_0, x = k_359_cast_fp16)[name = tensor<string, []>("op_40276_cast_fp16")];
            tensor<int32, [4]> var_40280_begin_0 = const()[name = tensor<string, []>("op_40280_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_40280_end_0 = const()[name = tensor<string, []>("op_40280_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_40280_end_mask_0 = const()[name = tensor<string, []>("op_40280_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_40280_cast_fp16 = slice_by_index(begin = var_40280_begin_0, end = var_40280_end_0, end_mask = var_40280_end_mask_0, x = k_359_cast_fp16)[name = tensor<string, []>("op_40280_cast_fp16")];
            tensor<int32, [4]> var_40284_begin_0 = const()[name = tensor<string, []>("op_40284_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_40284_end_0 = const()[name = tensor<string, []>("op_40284_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_40284_end_mask_0 = const()[name = tensor<string, []>("op_40284_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_40284_cast_fp16 = slice_by_index(begin = var_40284_begin_0, end = var_40284_end_0, end_mask = var_40284_end_mask_0, x = k_359_cast_fp16)[name = tensor<string, []>("op_40284_cast_fp16")];
            tensor<int32, [4]> var_40288_begin_0 = const()[name = tensor<string, []>("op_40288_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_40288_end_0 = const()[name = tensor<string, []>("op_40288_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_40288_end_mask_0 = const()[name = tensor<string, []>("op_40288_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_40288_cast_fp16 = slice_by_index(begin = var_40288_begin_0, end = var_40288_end_0, end_mask = var_40288_end_mask_0, x = k_359_cast_fp16)[name = tensor<string, []>("op_40288_cast_fp16")];
            tensor<int32, [4]> var_40292_begin_0 = const()[name = tensor<string, []>("op_40292_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_40292_end_0 = const()[name = tensor<string, []>("op_40292_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_40292_end_mask_0 = const()[name = tensor<string, []>("op_40292_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_40292_cast_fp16 = slice_by_index(begin = var_40292_begin_0, end = var_40292_end_0, end_mask = var_40292_end_mask_0, x = k_359_cast_fp16)[name = tensor<string, []>("op_40292_cast_fp16")];
            tensor<int32, [4]> var_40296_begin_0 = const()[name = tensor<string, []>("op_40296_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_40296_end_0 = const()[name = tensor<string, []>("op_40296_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_40296_end_mask_0 = const()[name = tensor<string, []>("op_40296_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_40296_cast_fp16 = slice_by_index(begin = var_40296_begin_0, end = var_40296_end_0, end_mask = var_40296_end_mask_0, x = k_359_cast_fp16)[name = tensor<string, []>("op_40296_cast_fp16")];
            tensor<int32, [4]> var_40300_begin_0 = const()[name = tensor<string, []>("op_40300_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_40300_end_0 = const()[name = tensor<string, []>("op_40300_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_40300_end_mask_0 = const()[name = tensor<string, []>("op_40300_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_40300_cast_fp16 = slice_by_index(begin = var_40300_begin_0, end = var_40300_end_0, end_mask = var_40300_end_mask_0, x = k_359_cast_fp16)[name = tensor<string, []>("op_40300_cast_fp16")];
            tensor<int32, [4]> var_40304_begin_0 = const()[name = tensor<string, []>("op_40304_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_40304_end_0 = const()[name = tensor<string, []>("op_40304_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_40304_end_mask_0 = const()[name = tensor<string, []>("op_40304_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_40304_cast_fp16 = slice_by_index(begin = var_40304_begin_0, end = var_40304_end_0, end_mask = var_40304_end_mask_0, x = k_359_cast_fp16)[name = tensor<string, []>("op_40304_cast_fp16")];
            tensor<int32, [4]> var_40308_begin_0 = const()[name = tensor<string, []>("op_40308_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_40308_end_0 = const()[name = tensor<string, []>("op_40308_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_40308_end_mask_0 = const()[name = tensor<string, []>("op_40308_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_40308_cast_fp16 = slice_by_index(begin = var_40308_begin_0, end = var_40308_end_0, end_mask = var_40308_end_mask_0, x = k_359_cast_fp16)[name = tensor<string, []>("op_40308_cast_fp16")];
            tensor<int32, [4]> var_40310_begin_0 = const()[name = tensor<string, []>("op_40310_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_40310_end_0 = const()[name = tensor<string, []>("op_40310_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_40310_end_mask_0 = const()[name = tensor<string, []>("op_40310_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_40310_cast_fp16 = slice_by_index(begin = var_40310_begin_0, end = var_40310_end_0, end_mask = var_40310_end_mask_0, x = v_179_cast_fp16)[name = tensor<string, []>("op_40310_cast_fp16")];
            tensor<int32, [4]> var_40314_begin_0 = const()[name = tensor<string, []>("op_40314_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_40314_end_0 = const()[name = tensor<string, []>("op_40314_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_40314_end_mask_0 = const()[name = tensor<string, []>("op_40314_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_40314_cast_fp16 = slice_by_index(begin = var_40314_begin_0, end = var_40314_end_0, end_mask = var_40314_end_mask_0, x = v_179_cast_fp16)[name = tensor<string, []>("op_40314_cast_fp16")];
            tensor<int32, [4]> var_40318_begin_0 = const()[name = tensor<string, []>("op_40318_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_40318_end_0 = const()[name = tensor<string, []>("op_40318_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_40318_end_mask_0 = const()[name = tensor<string, []>("op_40318_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_40318_cast_fp16 = slice_by_index(begin = var_40318_begin_0, end = var_40318_end_0, end_mask = var_40318_end_mask_0, x = v_179_cast_fp16)[name = tensor<string, []>("op_40318_cast_fp16")];
            tensor<int32, [4]> var_40322_begin_0 = const()[name = tensor<string, []>("op_40322_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_40322_end_0 = const()[name = tensor<string, []>("op_40322_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_40322_end_mask_0 = const()[name = tensor<string, []>("op_40322_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_40322_cast_fp16 = slice_by_index(begin = var_40322_begin_0, end = var_40322_end_0, end_mask = var_40322_end_mask_0, x = v_179_cast_fp16)[name = tensor<string, []>("op_40322_cast_fp16")];
            tensor<int32, [4]> var_40326_begin_0 = const()[name = tensor<string, []>("op_40326_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_40326_end_0 = const()[name = tensor<string, []>("op_40326_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_40326_end_mask_0 = const()[name = tensor<string, []>("op_40326_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_40326_cast_fp16 = slice_by_index(begin = var_40326_begin_0, end = var_40326_end_0, end_mask = var_40326_end_mask_0, x = v_179_cast_fp16)[name = tensor<string, []>("op_40326_cast_fp16")];
            tensor<int32, [4]> var_40330_begin_0 = const()[name = tensor<string, []>("op_40330_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_40330_end_0 = const()[name = tensor<string, []>("op_40330_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_40330_end_mask_0 = const()[name = tensor<string, []>("op_40330_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_40330_cast_fp16 = slice_by_index(begin = var_40330_begin_0, end = var_40330_end_0, end_mask = var_40330_end_mask_0, x = v_179_cast_fp16)[name = tensor<string, []>("op_40330_cast_fp16")];
            tensor<int32, [4]> var_40334_begin_0 = const()[name = tensor<string, []>("op_40334_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_40334_end_0 = const()[name = tensor<string, []>("op_40334_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_40334_end_mask_0 = const()[name = tensor<string, []>("op_40334_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_40334_cast_fp16 = slice_by_index(begin = var_40334_begin_0, end = var_40334_end_0, end_mask = var_40334_end_mask_0, x = v_179_cast_fp16)[name = tensor<string, []>("op_40334_cast_fp16")];
            tensor<int32, [4]> var_40338_begin_0 = const()[name = tensor<string, []>("op_40338_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_40338_end_0 = const()[name = tensor<string, []>("op_40338_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_40338_end_mask_0 = const()[name = tensor<string, []>("op_40338_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_40338_cast_fp16 = slice_by_index(begin = var_40338_begin_0, end = var_40338_end_0, end_mask = var_40338_end_mask_0, x = v_179_cast_fp16)[name = tensor<string, []>("op_40338_cast_fp16")];
            tensor<int32, [4]> var_40342_begin_0 = const()[name = tensor<string, []>("op_40342_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_40342_end_0 = const()[name = tensor<string, []>("op_40342_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_40342_end_mask_0 = const()[name = tensor<string, []>("op_40342_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_40342_cast_fp16 = slice_by_index(begin = var_40342_begin_0, end = var_40342_end_0, end_mask = var_40342_end_mask_0, x = v_179_cast_fp16)[name = tensor<string, []>("op_40342_cast_fp16")];
            tensor<int32, [4]> var_40346_begin_0 = const()[name = tensor<string, []>("op_40346_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_40346_end_0 = const()[name = tensor<string, []>("op_40346_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_40346_end_mask_0 = const()[name = tensor<string, []>("op_40346_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_40346_cast_fp16 = slice_by_index(begin = var_40346_begin_0, end = var_40346_end_0, end_mask = var_40346_end_mask_0, x = v_179_cast_fp16)[name = tensor<string, []>("op_40346_cast_fp16")];
            tensor<int32, [4]> var_40350_begin_0 = const()[name = tensor<string, []>("op_40350_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_40350_end_0 = const()[name = tensor<string, []>("op_40350_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_40350_end_mask_0 = const()[name = tensor<string, []>("op_40350_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_40350_cast_fp16 = slice_by_index(begin = var_40350_begin_0, end = var_40350_end_0, end_mask = var_40350_end_mask_0, x = v_179_cast_fp16)[name = tensor<string, []>("op_40350_cast_fp16")];
            tensor<int32, [4]> var_40354_begin_0 = const()[name = tensor<string, []>("op_40354_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_40354_end_0 = const()[name = tensor<string, []>("op_40354_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_40354_end_mask_0 = const()[name = tensor<string, []>("op_40354_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_40354_cast_fp16 = slice_by_index(begin = var_40354_begin_0, end = var_40354_end_0, end_mask = var_40354_end_mask_0, x = v_179_cast_fp16)[name = tensor<string, []>("op_40354_cast_fp16")];
            tensor<int32, [4]> var_40358_begin_0 = const()[name = tensor<string, []>("op_40358_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_40358_end_0 = const()[name = tensor<string, []>("op_40358_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_40358_end_mask_0 = const()[name = tensor<string, []>("op_40358_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_40358_cast_fp16 = slice_by_index(begin = var_40358_begin_0, end = var_40358_end_0, end_mask = var_40358_end_mask_0, x = v_179_cast_fp16)[name = tensor<string, []>("op_40358_cast_fp16")];
            tensor<int32, [4]> var_40362_begin_0 = const()[name = tensor<string, []>("op_40362_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_40362_end_0 = const()[name = tensor<string, []>("op_40362_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_40362_end_mask_0 = const()[name = tensor<string, []>("op_40362_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_40362_cast_fp16 = slice_by_index(begin = var_40362_begin_0, end = var_40362_end_0, end_mask = var_40362_end_mask_0, x = v_179_cast_fp16)[name = tensor<string, []>("op_40362_cast_fp16")];
            tensor<int32, [4]> var_40366_begin_0 = const()[name = tensor<string, []>("op_40366_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_40366_end_0 = const()[name = tensor<string, []>("op_40366_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_40366_end_mask_0 = const()[name = tensor<string, []>("op_40366_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_40366_cast_fp16 = slice_by_index(begin = var_40366_begin_0, end = var_40366_end_0, end_mask = var_40366_end_mask_0, x = v_179_cast_fp16)[name = tensor<string, []>("op_40366_cast_fp16")];
            tensor<int32, [4]> var_40370_begin_0 = const()[name = tensor<string, []>("op_40370_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_40370_end_0 = const()[name = tensor<string, []>("op_40370_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_40370_end_mask_0 = const()[name = tensor<string, []>("op_40370_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_40370_cast_fp16 = slice_by_index(begin = var_40370_begin_0, end = var_40370_end_0, end_mask = var_40370_end_mask_0, x = v_179_cast_fp16)[name = tensor<string, []>("op_40370_cast_fp16")];
            tensor<int32, [4]> var_40374_begin_0 = const()[name = tensor<string, []>("op_40374_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_40374_end_0 = const()[name = tensor<string, []>("op_40374_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_40374_end_mask_0 = const()[name = tensor<string, []>("op_40374_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_40374_cast_fp16 = slice_by_index(begin = var_40374_begin_0, end = var_40374_end_0, end_mask = var_40374_end_mask_0, x = v_179_cast_fp16)[name = tensor<string, []>("op_40374_cast_fp16")];
            tensor<int32, [4]> var_40378_begin_0 = const()[name = tensor<string, []>("op_40378_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_40378_end_0 = const()[name = tensor<string, []>("op_40378_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_40378_end_mask_0 = const()[name = tensor<string, []>("op_40378_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_40378_cast_fp16 = slice_by_index(begin = var_40378_begin_0, end = var_40378_end_0, end_mask = var_40378_end_mask_0, x = v_179_cast_fp16)[name = tensor<string, []>("op_40378_cast_fp16")];
            tensor<int32, [4]> var_40382_begin_0 = const()[name = tensor<string, []>("op_40382_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_40382_end_0 = const()[name = tensor<string, []>("op_40382_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_40382_end_mask_0 = const()[name = tensor<string, []>("op_40382_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_40382_cast_fp16 = slice_by_index(begin = var_40382_begin_0, end = var_40382_end_0, end_mask = var_40382_end_mask_0, x = v_179_cast_fp16)[name = tensor<string, []>("op_40382_cast_fp16")];
            tensor<int32, [4]> var_40386_begin_0 = const()[name = tensor<string, []>("op_40386_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_40386_end_0 = const()[name = tensor<string, []>("op_40386_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_40386_end_mask_0 = const()[name = tensor<string, []>("op_40386_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_40386_cast_fp16 = slice_by_index(begin = var_40386_begin_0, end = var_40386_end_0, end_mask = var_40386_end_mask_0, x = v_179_cast_fp16)[name = tensor<string, []>("op_40386_cast_fp16")];
            tensor<string, []> var_40390_equation_0 = const()[name = tensor<string, []>("op_40390_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_40390_cast_fp16 = einsum(equation = var_40390_equation_0, values = (var_40232_cast_fp16, var_40149_cast_fp16))[name = tensor<string, []>("op_40390_cast_fp16")];
            tensor<fp16, []> var_40391_to_fp16 = const()[name = tensor<string, []>("op_40391_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3401_cast_fp16 = mul(x = var_40390_cast_fp16, y = var_40391_to_fp16)[name = tensor<string, []>("aw_3401_cast_fp16")];
            tensor<string, []> var_40394_equation_0 = const()[name = tensor<string, []>("op_40394_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_40394_cast_fp16 = einsum(equation = var_40394_equation_0, values = (var_40236_cast_fp16, var_40153_cast_fp16))[name = tensor<string, []>("op_40394_cast_fp16")];
            tensor<fp16, []> var_40395_to_fp16 = const()[name = tensor<string, []>("op_40395_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3403_cast_fp16 = mul(x = var_40394_cast_fp16, y = var_40395_to_fp16)[name = tensor<string, []>("aw_3403_cast_fp16")];
            tensor<string, []> var_40398_equation_0 = const()[name = tensor<string, []>("op_40398_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_40398_cast_fp16 = einsum(equation = var_40398_equation_0, values = (var_40240_cast_fp16, var_40157_cast_fp16))[name = tensor<string, []>("op_40398_cast_fp16")];
            tensor<fp16, []> var_40399_to_fp16 = const()[name = tensor<string, []>("op_40399_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3405_cast_fp16 = mul(x = var_40398_cast_fp16, y = var_40399_to_fp16)[name = tensor<string, []>("aw_3405_cast_fp16")];
            tensor<string, []> var_40402_equation_0 = const()[name = tensor<string, []>("op_40402_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_40402_cast_fp16 = einsum(equation = var_40402_equation_0, values = (var_40244_cast_fp16, var_40161_cast_fp16))[name = tensor<string, []>("op_40402_cast_fp16")];
            tensor<fp16, []> var_40403_to_fp16 = const()[name = tensor<string, []>("op_40403_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3407_cast_fp16 = mul(x = var_40402_cast_fp16, y = var_40403_to_fp16)[name = tensor<string, []>("aw_3407_cast_fp16")];
            tensor<string, []> var_40406_equation_0 = const()[name = tensor<string, []>("op_40406_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_40406_cast_fp16 = einsum(equation = var_40406_equation_0, values = (var_40248_cast_fp16, var_40165_cast_fp16))[name = tensor<string, []>("op_40406_cast_fp16")];
            tensor<fp16, []> var_40407_to_fp16 = const()[name = tensor<string, []>("op_40407_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3409_cast_fp16 = mul(x = var_40406_cast_fp16, y = var_40407_to_fp16)[name = tensor<string, []>("aw_3409_cast_fp16")];
            tensor<string, []> var_40410_equation_0 = const()[name = tensor<string, []>("op_40410_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_40410_cast_fp16 = einsum(equation = var_40410_equation_0, values = (var_40252_cast_fp16, var_40169_cast_fp16))[name = tensor<string, []>("op_40410_cast_fp16")];
            tensor<fp16, []> var_40411_to_fp16 = const()[name = tensor<string, []>("op_40411_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3411_cast_fp16 = mul(x = var_40410_cast_fp16, y = var_40411_to_fp16)[name = tensor<string, []>("aw_3411_cast_fp16")];
            tensor<string, []> var_40414_equation_0 = const()[name = tensor<string, []>("op_40414_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_40414_cast_fp16 = einsum(equation = var_40414_equation_0, values = (var_40256_cast_fp16, var_40173_cast_fp16))[name = tensor<string, []>("op_40414_cast_fp16")];
            tensor<fp16, []> var_40415_to_fp16 = const()[name = tensor<string, []>("op_40415_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3413_cast_fp16 = mul(x = var_40414_cast_fp16, y = var_40415_to_fp16)[name = tensor<string, []>("aw_3413_cast_fp16")];
            tensor<string, []> var_40418_equation_0 = const()[name = tensor<string, []>("op_40418_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_40418_cast_fp16 = einsum(equation = var_40418_equation_0, values = (var_40260_cast_fp16, var_40177_cast_fp16))[name = tensor<string, []>("op_40418_cast_fp16")];
            tensor<fp16, []> var_40419_to_fp16 = const()[name = tensor<string, []>("op_40419_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3415_cast_fp16 = mul(x = var_40418_cast_fp16, y = var_40419_to_fp16)[name = tensor<string, []>("aw_3415_cast_fp16")];
            tensor<string, []> var_40422_equation_0 = const()[name = tensor<string, []>("op_40422_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_40422_cast_fp16 = einsum(equation = var_40422_equation_0, values = (var_40264_cast_fp16, var_40181_cast_fp16))[name = tensor<string, []>("op_40422_cast_fp16")];
            tensor<fp16, []> var_40423_to_fp16 = const()[name = tensor<string, []>("op_40423_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3417_cast_fp16 = mul(x = var_40422_cast_fp16, y = var_40423_to_fp16)[name = tensor<string, []>("aw_3417_cast_fp16")];
            tensor<string, []> var_40426_equation_0 = const()[name = tensor<string, []>("op_40426_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_40426_cast_fp16 = einsum(equation = var_40426_equation_0, values = (var_40268_cast_fp16, var_40185_cast_fp16))[name = tensor<string, []>("op_40426_cast_fp16")];
            tensor<fp16, []> var_40427_to_fp16 = const()[name = tensor<string, []>("op_40427_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3419_cast_fp16 = mul(x = var_40426_cast_fp16, y = var_40427_to_fp16)[name = tensor<string, []>("aw_3419_cast_fp16")];
            tensor<string, []> var_40430_equation_0 = const()[name = tensor<string, []>("op_40430_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_40430_cast_fp16 = einsum(equation = var_40430_equation_0, values = (var_40272_cast_fp16, var_40189_cast_fp16))[name = tensor<string, []>("op_40430_cast_fp16")];
            tensor<fp16, []> var_40431_to_fp16 = const()[name = tensor<string, []>("op_40431_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3421_cast_fp16 = mul(x = var_40430_cast_fp16, y = var_40431_to_fp16)[name = tensor<string, []>("aw_3421_cast_fp16")];
            tensor<string, []> var_40434_equation_0 = const()[name = tensor<string, []>("op_40434_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_40434_cast_fp16 = einsum(equation = var_40434_equation_0, values = (var_40276_cast_fp16, var_40193_cast_fp16))[name = tensor<string, []>("op_40434_cast_fp16")];
            tensor<fp16, []> var_40435_to_fp16 = const()[name = tensor<string, []>("op_40435_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3423_cast_fp16 = mul(x = var_40434_cast_fp16, y = var_40435_to_fp16)[name = tensor<string, []>("aw_3423_cast_fp16")];
            tensor<string, []> var_40438_equation_0 = const()[name = tensor<string, []>("op_40438_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_40438_cast_fp16 = einsum(equation = var_40438_equation_0, values = (var_40280_cast_fp16, var_40197_cast_fp16))[name = tensor<string, []>("op_40438_cast_fp16")];
            tensor<fp16, []> var_40439_to_fp16 = const()[name = tensor<string, []>("op_40439_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3425_cast_fp16 = mul(x = var_40438_cast_fp16, y = var_40439_to_fp16)[name = tensor<string, []>("aw_3425_cast_fp16")];
            tensor<string, []> var_40442_equation_0 = const()[name = tensor<string, []>("op_40442_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_40442_cast_fp16 = einsum(equation = var_40442_equation_0, values = (var_40284_cast_fp16, var_40201_cast_fp16))[name = tensor<string, []>("op_40442_cast_fp16")];
            tensor<fp16, []> var_40443_to_fp16 = const()[name = tensor<string, []>("op_40443_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3427_cast_fp16 = mul(x = var_40442_cast_fp16, y = var_40443_to_fp16)[name = tensor<string, []>("aw_3427_cast_fp16")];
            tensor<string, []> var_40446_equation_0 = const()[name = tensor<string, []>("op_40446_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_40446_cast_fp16 = einsum(equation = var_40446_equation_0, values = (var_40288_cast_fp16, var_40205_cast_fp16))[name = tensor<string, []>("op_40446_cast_fp16")];
            tensor<fp16, []> var_40447_to_fp16 = const()[name = tensor<string, []>("op_40447_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3429_cast_fp16 = mul(x = var_40446_cast_fp16, y = var_40447_to_fp16)[name = tensor<string, []>("aw_3429_cast_fp16")];
            tensor<string, []> var_40450_equation_0 = const()[name = tensor<string, []>("op_40450_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_40450_cast_fp16 = einsum(equation = var_40450_equation_0, values = (var_40292_cast_fp16, var_40209_cast_fp16))[name = tensor<string, []>("op_40450_cast_fp16")];
            tensor<fp16, []> var_40451_to_fp16 = const()[name = tensor<string, []>("op_40451_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3431_cast_fp16 = mul(x = var_40450_cast_fp16, y = var_40451_to_fp16)[name = tensor<string, []>("aw_3431_cast_fp16")];
            tensor<string, []> var_40454_equation_0 = const()[name = tensor<string, []>("op_40454_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_40454_cast_fp16 = einsum(equation = var_40454_equation_0, values = (var_40296_cast_fp16, var_40213_cast_fp16))[name = tensor<string, []>("op_40454_cast_fp16")];
            tensor<fp16, []> var_40455_to_fp16 = const()[name = tensor<string, []>("op_40455_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3433_cast_fp16 = mul(x = var_40454_cast_fp16, y = var_40455_to_fp16)[name = tensor<string, []>("aw_3433_cast_fp16")];
            tensor<string, []> var_40458_equation_0 = const()[name = tensor<string, []>("op_40458_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_40458_cast_fp16 = einsum(equation = var_40458_equation_0, values = (var_40300_cast_fp16, var_40217_cast_fp16))[name = tensor<string, []>("op_40458_cast_fp16")];
            tensor<fp16, []> var_40459_to_fp16 = const()[name = tensor<string, []>("op_40459_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3435_cast_fp16 = mul(x = var_40458_cast_fp16, y = var_40459_to_fp16)[name = tensor<string, []>("aw_3435_cast_fp16")];
            tensor<string, []> var_40462_equation_0 = const()[name = tensor<string, []>("op_40462_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_40462_cast_fp16 = einsum(equation = var_40462_equation_0, values = (var_40304_cast_fp16, var_40221_cast_fp16))[name = tensor<string, []>("op_40462_cast_fp16")];
            tensor<fp16, []> var_40463_to_fp16 = const()[name = tensor<string, []>("op_40463_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3437_cast_fp16 = mul(x = var_40462_cast_fp16, y = var_40463_to_fp16)[name = tensor<string, []>("aw_3437_cast_fp16")];
            tensor<string, []> var_40466_equation_0 = const()[name = tensor<string, []>("op_40466_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_40466_cast_fp16 = einsum(equation = var_40466_equation_0, values = (var_40308_cast_fp16, var_40225_cast_fp16))[name = tensor<string, []>("op_40466_cast_fp16")];
            tensor<fp16, []> var_40467_to_fp16 = const()[name = tensor<string, []>("op_40467_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3439_cast_fp16 = mul(x = var_40466_cast_fp16, y = var_40467_to_fp16)[name = tensor<string, []>("aw_3439_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_40469_cast_fp16 = softmax(axis = var_30355, x = aw_3401_cast_fp16)[name = tensor<string, []>("op_40469_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_40470_cast_fp16 = softmax(axis = var_30355, x = aw_3403_cast_fp16)[name = tensor<string, []>("op_40470_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_40471_cast_fp16 = softmax(axis = var_30355, x = aw_3405_cast_fp16)[name = tensor<string, []>("op_40471_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_40472_cast_fp16 = softmax(axis = var_30355, x = aw_3407_cast_fp16)[name = tensor<string, []>("op_40472_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_40473_cast_fp16 = softmax(axis = var_30355, x = aw_3409_cast_fp16)[name = tensor<string, []>("op_40473_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_40474_cast_fp16 = softmax(axis = var_30355, x = aw_3411_cast_fp16)[name = tensor<string, []>("op_40474_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_40475_cast_fp16 = softmax(axis = var_30355, x = aw_3413_cast_fp16)[name = tensor<string, []>("op_40475_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_40476_cast_fp16 = softmax(axis = var_30355, x = aw_3415_cast_fp16)[name = tensor<string, []>("op_40476_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_40477_cast_fp16 = softmax(axis = var_30355, x = aw_3417_cast_fp16)[name = tensor<string, []>("op_40477_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_40478_cast_fp16 = softmax(axis = var_30355, x = aw_3419_cast_fp16)[name = tensor<string, []>("op_40478_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_40479_cast_fp16 = softmax(axis = var_30355, x = aw_3421_cast_fp16)[name = tensor<string, []>("op_40479_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_40480_cast_fp16 = softmax(axis = var_30355, x = aw_3423_cast_fp16)[name = tensor<string, []>("op_40480_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_40481_cast_fp16 = softmax(axis = var_30355, x = aw_3425_cast_fp16)[name = tensor<string, []>("op_40481_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_40482_cast_fp16 = softmax(axis = var_30355, x = aw_3427_cast_fp16)[name = tensor<string, []>("op_40482_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_40483_cast_fp16 = softmax(axis = var_30355, x = aw_3429_cast_fp16)[name = tensor<string, []>("op_40483_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_40484_cast_fp16 = softmax(axis = var_30355, x = aw_3431_cast_fp16)[name = tensor<string, []>("op_40484_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_40485_cast_fp16 = softmax(axis = var_30355, x = aw_3433_cast_fp16)[name = tensor<string, []>("op_40485_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_40486_cast_fp16 = softmax(axis = var_30355, x = aw_3435_cast_fp16)[name = tensor<string, []>("op_40486_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_40487_cast_fp16 = softmax(axis = var_30355, x = aw_3437_cast_fp16)[name = tensor<string, []>("op_40487_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_40488_cast_fp16 = softmax(axis = var_30355, x = aw_3439_cast_fp16)[name = tensor<string, []>("op_40488_cast_fp16")];
            tensor<string, []> var_40490_equation_0 = const()[name = tensor<string, []>("op_40490_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40490_cast_fp16 = einsum(equation = var_40490_equation_0, values = (var_40310_cast_fp16, var_40469_cast_fp16))[name = tensor<string, []>("op_40490_cast_fp16")];
            tensor<string, []> var_40492_equation_0 = const()[name = tensor<string, []>("op_40492_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40492_cast_fp16 = einsum(equation = var_40492_equation_0, values = (var_40314_cast_fp16, var_40470_cast_fp16))[name = tensor<string, []>("op_40492_cast_fp16")];
            tensor<string, []> var_40494_equation_0 = const()[name = tensor<string, []>("op_40494_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40494_cast_fp16 = einsum(equation = var_40494_equation_0, values = (var_40318_cast_fp16, var_40471_cast_fp16))[name = tensor<string, []>("op_40494_cast_fp16")];
            tensor<string, []> var_40496_equation_0 = const()[name = tensor<string, []>("op_40496_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40496_cast_fp16 = einsum(equation = var_40496_equation_0, values = (var_40322_cast_fp16, var_40472_cast_fp16))[name = tensor<string, []>("op_40496_cast_fp16")];
            tensor<string, []> var_40498_equation_0 = const()[name = tensor<string, []>("op_40498_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40498_cast_fp16 = einsum(equation = var_40498_equation_0, values = (var_40326_cast_fp16, var_40473_cast_fp16))[name = tensor<string, []>("op_40498_cast_fp16")];
            tensor<string, []> var_40500_equation_0 = const()[name = tensor<string, []>("op_40500_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40500_cast_fp16 = einsum(equation = var_40500_equation_0, values = (var_40330_cast_fp16, var_40474_cast_fp16))[name = tensor<string, []>("op_40500_cast_fp16")];
            tensor<string, []> var_40502_equation_0 = const()[name = tensor<string, []>("op_40502_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40502_cast_fp16 = einsum(equation = var_40502_equation_0, values = (var_40334_cast_fp16, var_40475_cast_fp16))[name = tensor<string, []>("op_40502_cast_fp16")];
            tensor<string, []> var_40504_equation_0 = const()[name = tensor<string, []>("op_40504_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40504_cast_fp16 = einsum(equation = var_40504_equation_0, values = (var_40338_cast_fp16, var_40476_cast_fp16))[name = tensor<string, []>("op_40504_cast_fp16")];
            tensor<string, []> var_40506_equation_0 = const()[name = tensor<string, []>("op_40506_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40506_cast_fp16 = einsum(equation = var_40506_equation_0, values = (var_40342_cast_fp16, var_40477_cast_fp16))[name = tensor<string, []>("op_40506_cast_fp16")];
            tensor<string, []> var_40508_equation_0 = const()[name = tensor<string, []>("op_40508_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40508_cast_fp16 = einsum(equation = var_40508_equation_0, values = (var_40346_cast_fp16, var_40478_cast_fp16))[name = tensor<string, []>("op_40508_cast_fp16")];
            tensor<string, []> var_40510_equation_0 = const()[name = tensor<string, []>("op_40510_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40510_cast_fp16 = einsum(equation = var_40510_equation_0, values = (var_40350_cast_fp16, var_40479_cast_fp16))[name = tensor<string, []>("op_40510_cast_fp16")];
            tensor<string, []> var_40512_equation_0 = const()[name = tensor<string, []>("op_40512_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40512_cast_fp16 = einsum(equation = var_40512_equation_0, values = (var_40354_cast_fp16, var_40480_cast_fp16))[name = tensor<string, []>("op_40512_cast_fp16")];
            tensor<string, []> var_40514_equation_0 = const()[name = tensor<string, []>("op_40514_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40514_cast_fp16 = einsum(equation = var_40514_equation_0, values = (var_40358_cast_fp16, var_40481_cast_fp16))[name = tensor<string, []>("op_40514_cast_fp16")];
            tensor<string, []> var_40516_equation_0 = const()[name = tensor<string, []>("op_40516_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40516_cast_fp16 = einsum(equation = var_40516_equation_0, values = (var_40362_cast_fp16, var_40482_cast_fp16))[name = tensor<string, []>("op_40516_cast_fp16")];
            tensor<string, []> var_40518_equation_0 = const()[name = tensor<string, []>("op_40518_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40518_cast_fp16 = einsum(equation = var_40518_equation_0, values = (var_40366_cast_fp16, var_40483_cast_fp16))[name = tensor<string, []>("op_40518_cast_fp16")];
            tensor<string, []> var_40520_equation_0 = const()[name = tensor<string, []>("op_40520_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40520_cast_fp16 = einsum(equation = var_40520_equation_0, values = (var_40370_cast_fp16, var_40484_cast_fp16))[name = tensor<string, []>("op_40520_cast_fp16")];
            tensor<string, []> var_40522_equation_0 = const()[name = tensor<string, []>("op_40522_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40522_cast_fp16 = einsum(equation = var_40522_equation_0, values = (var_40374_cast_fp16, var_40485_cast_fp16))[name = tensor<string, []>("op_40522_cast_fp16")];
            tensor<string, []> var_40524_equation_0 = const()[name = tensor<string, []>("op_40524_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40524_cast_fp16 = einsum(equation = var_40524_equation_0, values = (var_40378_cast_fp16, var_40486_cast_fp16))[name = tensor<string, []>("op_40524_cast_fp16")];
            tensor<string, []> var_40526_equation_0 = const()[name = tensor<string, []>("op_40526_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40526_cast_fp16 = einsum(equation = var_40526_equation_0, values = (var_40382_cast_fp16, var_40487_cast_fp16))[name = tensor<string, []>("op_40526_cast_fp16")];
            tensor<string, []> var_40528_equation_0 = const()[name = tensor<string, []>("op_40528_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40528_cast_fp16 = einsum(equation = var_40528_equation_0, values = (var_40386_cast_fp16, var_40488_cast_fp16))[name = tensor<string, []>("op_40528_cast_fp16")];
            tensor<bool, []> input_539_interleave_0 = const()[name = tensor<string, []>("input_539_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_539_cast_fp16 = concat(axis = var_30355, interleave = input_539_interleave_0, values = (var_40490_cast_fp16, var_40492_cast_fp16, var_40494_cast_fp16, var_40496_cast_fp16, var_40498_cast_fp16, var_40500_cast_fp16, var_40502_cast_fp16, var_40504_cast_fp16, var_40506_cast_fp16, var_40508_cast_fp16, var_40510_cast_fp16, var_40512_cast_fp16, var_40514_cast_fp16, var_40516_cast_fp16, var_40518_cast_fp16, var_40520_cast_fp16, var_40522_cast_fp16, var_40524_cast_fp16, var_40526_cast_fp16, var_40528_cast_fp16))[name = tensor<string, []>("input_539_cast_fp16")];
            tensor<string, []> var_40538_pad_type_0 = const()[name = tensor<string, []>("op_40538_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_40538_strides_0 = const()[name = tensor<string, []>("op_40538_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_40538_pad_0 = const()[name = tensor<string, []>("op_40538_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_40538_dilations_0 = const()[name = tensor<string, []>("op_40538_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_40538_groups_0 = const()[name = tensor<string, []>("op_40538_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(314140800))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(315369664))), name = tensor<string, []>("up_blocks_0_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_0_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_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, []>(315369856)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_40538_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_40538_dilations_0, groups = var_40538_groups_0, pad = var_40538_pad_0, pad_type = var_40538_pad_type_0, strides = var_40538_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized, x = input_539_cast_fp16)[name = tensor<string, []>("op_40538_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_269_cast_fp16 = add(x = var_40538_cast_fp16, y = inputs_267_cast_fp16)[name = tensor<string, []>("inputs_269_cast_fp16")];
            tensor<int32, [1]> input_541_axes_0 = const()[name = tensor<string, []>("input_541_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> input_541_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_541_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(315372480)))];
            tensor<fp16, [1280]> input_541_beta_0_to_fp16 = const()[name = tensor<string, []>("input_541_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(315375104)))];
            tensor<fp16, []> var_40548_to_fp16 = const()[name = tensor<string, []>("op_40548_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> input_541_cast_fp16 = layer_norm(axes = input_541_axes_0, beta = input_541_beta_0_to_fp16, epsilon = var_40548_to_fp16, gamma = input_541_gamma_0_to_fp16, x = inputs_269_cast_fp16)[name = tensor<string, []>("input_541_cast_fp16")];
            tensor<string, []> var_40568_pad_type_0 = const()[name = tensor<string, []>("op_40568_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_40568_strides_0 = const()[name = tensor<string, []>("op_40568_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_40568_pad_0 = const()[name = tensor<string, []>("op_40568_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_40568_dilations_0 = const()[name = tensor<string, []>("op_40568_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_40568_groups_0 = const()[name = tensor<string, []>("op_40568_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [9830400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(315377728))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(325208192))), name = tensor<string, []>("up_blocks_0_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_0_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(325208384)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_40568_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_40568_dilations_0, groups = var_40568_groups_0, pad = var_40568_pad_0, pad_type = var_40568_pad_type_0, strides = var_40568_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized, x = input_541_cast_fp16)[name = tensor<string, []>("op_40568_cast_fp16")];
            tensor<int32, [2]> var_40569_split_sizes_0 = const()[name = tensor<string, []>("op_40569_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_40569_axis_0 = const()[name = tensor<string, []>("op_40569_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_40569_cast_fp16_0, tensor<fp16, [2, 5120, 1, 1024]> var_40569_cast_fp16_1 = split(axis = var_40569_axis_0, split_sizes = var_40569_split_sizes_0, x = var_40568_cast_fp16)[name = tensor<string, []>("op_40569_cast_fp16")];
            tensor<string, []> var_40571_mode_0 = const()[name = tensor<string, []>("op_40571_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_40571_cast_fp16 = gelu(mode = var_40571_mode_0, x = var_40569_cast_fp16_1)[name = tensor<string, []>("op_40571_cast_fp16")];
            tensor<fp16, [2, 5120, 1, 1024]> input_543_cast_fp16 = mul(x = var_40569_cast_fp16_0, y = var_40571_cast_fp16)[name = tensor<string, []>("input_543_cast_fp16")];
            tensor<string, []> var_40579_pad_type_0 = const()[name = tensor<string, []>("op_40579_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_40579_strides_0 = const()[name = tensor<string, []>("op_40579_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_40579_pad_0 = const()[name = tensor<string, []>("op_40579_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_40579_dilations_0 = const()[name = tensor<string, []>("op_40579_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_40579_groups_0 = const()[name = tensor<string, []>("op_40579_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [4915200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(325228928))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(330144192))), name = tensor<string, []>("up_blocks_0_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_0_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_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, []>(330144384)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_40579_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_40579_dilations_0, groups = var_40579_groups_0, pad = var_40579_pad_0, pad_type = var_40579_pad_type_0, strides = var_40579_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_543_cast_fp16)[name = tensor<string, []>("op_40579_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_271_cast_fp16 = add(x = var_40579_cast_fp16, y = inputs_269_cast_fp16)[name = tensor<string, []>("inputs_271_cast_fp16")];
            tensor<int32, [1]> hidden_states_367_axes_0 = const()[name = tensor<string, []>("hidden_states_367_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_367_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_367_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(330147008)))];
            tensor<fp16, [1280]> hidden_states_367_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_367_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(330149632)))];
            tensor<fp16, []> var_40595_to_fp16 = const()[name = tensor<string, []>("op_40595_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_367_cast_fp16 = layer_norm(axes = hidden_states_367_axes_0, beta = hidden_states_367_beta_0_to_fp16, epsilon = var_40595_to_fp16, gamma = hidden_states_367_gamma_0_to_fp16, x = inputs_271_cast_fp16)[name = tensor<string, []>("hidden_states_367_cast_fp16")];
            tensor<string, []> q_181_pad_type_0 = const()[name = tensor<string, []>("q_181_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_181_strides_0 = const()[name = tensor<string, []>("q_181_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_181_pad_0 = const()[name = tensor<string, []>("q_181_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_181_dilations_0 = const()[name = tensor<string, []>("q_181_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_181_groups_0 = const()[name = tensor<string, []>("q_181_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_1_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(330152256))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(331381120))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_1_attn1_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_181_cast_fp16 = conv(dilations = q_181_dilations_0, groups = q_181_groups_0, pad = q_181_pad_0, pad_type = q_181_pad_type_0, strides = q_181_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_1_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_367_cast_fp16)[name = tensor<string, []>("q_181_cast_fp16")];
            tensor<string, []> k_361_pad_type_0 = const()[name = tensor<string, []>("k_361_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_361_strides_0 = const()[name = tensor<string, []>("k_361_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_361_pad_0 = const()[name = tensor<string, []>("k_361_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_361_dilations_0 = const()[name = tensor<string, []>("k_361_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_361_groups_0 = const()[name = tensor<string, []>("k_361_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_1_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(331381312))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(332610176))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_1_attn1_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> k_361_cast_fp16 = conv(dilations = k_361_dilations_0, groups = k_361_groups_0, pad = k_361_pad_0, pad_type = k_361_pad_type_0, strides = k_361_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_1_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_367_cast_fp16)[name = tensor<string, []>("k_361_cast_fp16")];
            tensor<string, []> v_181_pad_type_0 = const()[name = tensor<string, []>("v_181_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_181_strides_0 = const()[name = tensor<string, []>("v_181_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_181_pad_0 = const()[name = tensor<string, []>("v_181_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_181_dilations_0 = const()[name = tensor<string, []>("v_181_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_181_groups_0 = const()[name = tensor<string, []>("v_181_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_1_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(332610368))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(333839232))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_1_attn1_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> v_181_cast_fp16 = conv(dilations = v_181_dilations_0, groups = v_181_groups_0, pad = v_181_pad_0, pad_type = v_181_pad_type_0, strides = v_181_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_1_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_367_cast_fp16)[name = tensor<string, []>("v_181_cast_fp16")];
            tensor<int32, [4]> var_40628_begin_0 = const()[name = tensor<string, []>("op_40628_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_40628_end_0 = const()[name = tensor<string, []>("op_40628_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_40628_end_mask_0 = const()[name = tensor<string, []>("op_40628_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40628_cast_fp16 = slice_by_index(begin = var_40628_begin_0, end = var_40628_end_0, end_mask = var_40628_end_mask_0, x = q_181_cast_fp16)[name = tensor<string, []>("op_40628_cast_fp16")];
            tensor<int32, [4]> var_40632_begin_0 = const()[name = tensor<string, []>("op_40632_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_40632_end_0 = const()[name = tensor<string, []>("op_40632_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_40632_end_mask_0 = const()[name = tensor<string, []>("op_40632_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40632_cast_fp16 = slice_by_index(begin = var_40632_begin_0, end = var_40632_end_0, end_mask = var_40632_end_mask_0, x = q_181_cast_fp16)[name = tensor<string, []>("op_40632_cast_fp16")];
            tensor<int32, [4]> var_40636_begin_0 = const()[name = tensor<string, []>("op_40636_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_40636_end_0 = const()[name = tensor<string, []>("op_40636_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_40636_end_mask_0 = const()[name = tensor<string, []>("op_40636_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40636_cast_fp16 = slice_by_index(begin = var_40636_begin_0, end = var_40636_end_0, end_mask = var_40636_end_mask_0, x = q_181_cast_fp16)[name = tensor<string, []>("op_40636_cast_fp16")];
            tensor<int32, [4]> var_40640_begin_0 = const()[name = tensor<string, []>("op_40640_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_40640_end_0 = const()[name = tensor<string, []>("op_40640_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_40640_end_mask_0 = const()[name = tensor<string, []>("op_40640_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40640_cast_fp16 = slice_by_index(begin = var_40640_begin_0, end = var_40640_end_0, end_mask = var_40640_end_mask_0, x = q_181_cast_fp16)[name = tensor<string, []>("op_40640_cast_fp16")];
            tensor<int32, [4]> var_40644_begin_0 = const()[name = tensor<string, []>("op_40644_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_40644_end_0 = const()[name = tensor<string, []>("op_40644_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_40644_end_mask_0 = const()[name = tensor<string, []>("op_40644_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40644_cast_fp16 = slice_by_index(begin = var_40644_begin_0, end = var_40644_end_0, end_mask = var_40644_end_mask_0, x = q_181_cast_fp16)[name = tensor<string, []>("op_40644_cast_fp16")];
            tensor<int32, [4]> var_40648_begin_0 = const()[name = tensor<string, []>("op_40648_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_40648_end_0 = const()[name = tensor<string, []>("op_40648_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_40648_end_mask_0 = const()[name = tensor<string, []>("op_40648_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40648_cast_fp16 = slice_by_index(begin = var_40648_begin_0, end = var_40648_end_0, end_mask = var_40648_end_mask_0, x = q_181_cast_fp16)[name = tensor<string, []>("op_40648_cast_fp16")];
            tensor<int32, [4]> var_40652_begin_0 = const()[name = tensor<string, []>("op_40652_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_40652_end_0 = const()[name = tensor<string, []>("op_40652_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_40652_end_mask_0 = const()[name = tensor<string, []>("op_40652_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40652_cast_fp16 = slice_by_index(begin = var_40652_begin_0, end = var_40652_end_0, end_mask = var_40652_end_mask_0, x = q_181_cast_fp16)[name = tensor<string, []>("op_40652_cast_fp16")];
            tensor<int32, [4]> var_40656_begin_0 = const()[name = tensor<string, []>("op_40656_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_40656_end_0 = const()[name = tensor<string, []>("op_40656_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_40656_end_mask_0 = const()[name = tensor<string, []>("op_40656_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40656_cast_fp16 = slice_by_index(begin = var_40656_begin_0, end = var_40656_end_0, end_mask = var_40656_end_mask_0, x = q_181_cast_fp16)[name = tensor<string, []>("op_40656_cast_fp16")];
            tensor<int32, [4]> var_40660_begin_0 = const()[name = tensor<string, []>("op_40660_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_40660_end_0 = const()[name = tensor<string, []>("op_40660_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_40660_end_mask_0 = const()[name = tensor<string, []>("op_40660_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40660_cast_fp16 = slice_by_index(begin = var_40660_begin_0, end = var_40660_end_0, end_mask = var_40660_end_mask_0, x = q_181_cast_fp16)[name = tensor<string, []>("op_40660_cast_fp16")];
            tensor<int32, [4]> var_40664_begin_0 = const()[name = tensor<string, []>("op_40664_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_40664_end_0 = const()[name = tensor<string, []>("op_40664_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_40664_end_mask_0 = const()[name = tensor<string, []>("op_40664_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40664_cast_fp16 = slice_by_index(begin = var_40664_begin_0, end = var_40664_end_0, end_mask = var_40664_end_mask_0, x = q_181_cast_fp16)[name = tensor<string, []>("op_40664_cast_fp16")];
            tensor<int32, [4]> var_40668_begin_0 = const()[name = tensor<string, []>("op_40668_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_40668_end_0 = const()[name = tensor<string, []>("op_40668_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_40668_end_mask_0 = const()[name = tensor<string, []>("op_40668_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40668_cast_fp16 = slice_by_index(begin = var_40668_begin_0, end = var_40668_end_0, end_mask = var_40668_end_mask_0, x = q_181_cast_fp16)[name = tensor<string, []>("op_40668_cast_fp16")];
            tensor<int32, [4]> var_40672_begin_0 = const()[name = tensor<string, []>("op_40672_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_40672_end_0 = const()[name = tensor<string, []>("op_40672_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_40672_end_mask_0 = const()[name = tensor<string, []>("op_40672_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40672_cast_fp16 = slice_by_index(begin = var_40672_begin_0, end = var_40672_end_0, end_mask = var_40672_end_mask_0, x = q_181_cast_fp16)[name = tensor<string, []>("op_40672_cast_fp16")];
            tensor<int32, [4]> var_40676_begin_0 = const()[name = tensor<string, []>("op_40676_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_40676_end_0 = const()[name = tensor<string, []>("op_40676_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_40676_end_mask_0 = const()[name = tensor<string, []>("op_40676_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40676_cast_fp16 = slice_by_index(begin = var_40676_begin_0, end = var_40676_end_0, end_mask = var_40676_end_mask_0, x = q_181_cast_fp16)[name = tensor<string, []>("op_40676_cast_fp16")];
            tensor<int32, [4]> var_40680_begin_0 = const()[name = tensor<string, []>("op_40680_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_40680_end_0 = const()[name = tensor<string, []>("op_40680_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_40680_end_mask_0 = const()[name = tensor<string, []>("op_40680_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40680_cast_fp16 = slice_by_index(begin = var_40680_begin_0, end = var_40680_end_0, end_mask = var_40680_end_mask_0, x = q_181_cast_fp16)[name = tensor<string, []>("op_40680_cast_fp16")];
            tensor<int32, [4]> var_40684_begin_0 = const()[name = tensor<string, []>("op_40684_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_40684_end_0 = const()[name = tensor<string, []>("op_40684_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_40684_end_mask_0 = const()[name = tensor<string, []>("op_40684_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40684_cast_fp16 = slice_by_index(begin = var_40684_begin_0, end = var_40684_end_0, end_mask = var_40684_end_mask_0, x = q_181_cast_fp16)[name = tensor<string, []>("op_40684_cast_fp16")];
            tensor<int32, [4]> var_40688_begin_0 = const()[name = tensor<string, []>("op_40688_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_40688_end_0 = const()[name = tensor<string, []>("op_40688_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_40688_end_mask_0 = const()[name = tensor<string, []>("op_40688_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40688_cast_fp16 = slice_by_index(begin = var_40688_begin_0, end = var_40688_end_0, end_mask = var_40688_end_mask_0, x = q_181_cast_fp16)[name = tensor<string, []>("op_40688_cast_fp16")];
            tensor<int32, [4]> var_40692_begin_0 = const()[name = tensor<string, []>("op_40692_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_40692_end_0 = const()[name = tensor<string, []>("op_40692_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_40692_end_mask_0 = const()[name = tensor<string, []>("op_40692_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40692_cast_fp16 = slice_by_index(begin = var_40692_begin_0, end = var_40692_end_0, end_mask = var_40692_end_mask_0, x = q_181_cast_fp16)[name = tensor<string, []>("op_40692_cast_fp16")];
            tensor<int32, [4]> var_40696_begin_0 = const()[name = tensor<string, []>("op_40696_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_40696_end_0 = const()[name = tensor<string, []>("op_40696_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_40696_end_mask_0 = const()[name = tensor<string, []>("op_40696_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40696_cast_fp16 = slice_by_index(begin = var_40696_begin_0, end = var_40696_end_0, end_mask = var_40696_end_mask_0, x = q_181_cast_fp16)[name = tensor<string, []>("op_40696_cast_fp16")];
            tensor<int32, [4]> var_40700_begin_0 = const()[name = tensor<string, []>("op_40700_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_40700_end_0 = const()[name = tensor<string, []>("op_40700_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_40700_end_mask_0 = const()[name = tensor<string, []>("op_40700_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40700_cast_fp16 = slice_by_index(begin = var_40700_begin_0, end = var_40700_end_0, end_mask = var_40700_end_mask_0, x = q_181_cast_fp16)[name = tensor<string, []>("op_40700_cast_fp16")];
            tensor<int32, [4]> var_40704_begin_0 = const()[name = tensor<string, []>("op_40704_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_40704_end_0 = const()[name = tensor<string, []>("op_40704_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_40704_end_mask_0 = const()[name = tensor<string, []>("op_40704_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40704_cast_fp16 = slice_by_index(begin = var_40704_begin_0, end = var_40704_end_0, end_mask = var_40704_end_mask_0, x = q_181_cast_fp16)[name = tensor<string, []>("op_40704_cast_fp16")];
            tensor<int32, [4]> k_363_perm_0 = const()[name = tensor<string, []>("k_363_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_40711_begin_0 = const()[name = tensor<string, []>("op_40711_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_40711_end_0 = const()[name = tensor<string, []>("op_40711_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_40711_end_mask_0 = const()[name = tensor<string, []>("op_40711_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> k_363_cast_fp16 = transpose(perm = k_363_perm_0, x = k_361_cast_fp16)[name = tensor<string, []>("transpose_49")];
            tensor<fp16, [2, 1024, 1, 64]> var_40711_cast_fp16 = slice_by_index(begin = var_40711_begin_0, end = var_40711_end_0, end_mask = var_40711_end_mask_0, x = k_363_cast_fp16)[name = tensor<string, []>("op_40711_cast_fp16")];
            tensor<int32, [4]> var_40715_begin_0 = const()[name = tensor<string, []>("op_40715_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_40715_end_0 = const()[name = tensor<string, []>("op_40715_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_40715_end_mask_0 = const()[name = tensor<string, []>("op_40715_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_40715_cast_fp16 = slice_by_index(begin = var_40715_begin_0, end = var_40715_end_0, end_mask = var_40715_end_mask_0, x = k_363_cast_fp16)[name = tensor<string, []>("op_40715_cast_fp16")];
            tensor<int32, [4]> var_40719_begin_0 = const()[name = tensor<string, []>("op_40719_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_40719_end_0 = const()[name = tensor<string, []>("op_40719_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_40719_end_mask_0 = const()[name = tensor<string, []>("op_40719_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_40719_cast_fp16 = slice_by_index(begin = var_40719_begin_0, end = var_40719_end_0, end_mask = var_40719_end_mask_0, x = k_363_cast_fp16)[name = tensor<string, []>("op_40719_cast_fp16")];
            tensor<int32, [4]> var_40723_begin_0 = const()[name = tensor<string, []>("op_40723_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_40723_end_0 = const()[name = tensor<string, []>("op_40723_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_40723_end_mask_0 = const()[name = tensor<string, []>("op_40723_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_40723_cast_fp16 = slice_by_index(begin = var_40723_begin_0, end = var_40723_end_0, end_mask = var_40723_end_mask_0, x = k_363_cast_fp16)[name = tensor<string, []>("op_40723_cast_fp16")];
            tensor<int32, [4]> var_40727_begin_0 = const()[name = tensor<string, []>("op_40727_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_40727_end_0 = const()[name = tensor<string, []>("op_40727_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_40727_end_mask_0 = const()[name = tensor<string, []>("op_40727_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_40727_cast_fp16 = slice_by_index(begin = var_40727_begin_0, end = var_40727_end_0, end_mask = var_40727_end_mask_0, x = k_363_cast_fp16)[name = tensor<string, []>("op_40727_cast_fp16")];
            tensor<int32, [4]> var_40731_begin_0 = const()[name = tensor<string, []>("op_40731_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_40731_end_0 = const()[name = tensor<string, []>("op_40731_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_40731_end_mask_0 = const()[name = tensor<string, []>("op_40731_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_40731_cast_fp16 = slice_by_index(begin = var_40731_begin_0, end = var_40731_end_0, end_mask = var_40731_end_mask_0, x = k_363_cast_fp16)[name = tensor<string, []>("op_40731_cast_fp16")];
            tensor<int32, [4]> var_40735_begin_0 = const()[name = tensor<string, []>("op_40735_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_40735_end_0 = const()[name = tensor<string, []>("op_40735_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_40735_end_mask_0 = const()[name = tensor<string, []>("op_40735_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_40735_cast_fp16 = slice_by_index(begin = var_40735_begin_0, end = var_40735_end_0, end_mask = var_40735_end_mask_0, x = k_363_cast_fp16)[name = tensor<string, []>("op_40735_cast_fp16")];
            tensor<int32, [4]> var_40739_begin_0 = const()[name = tensor<string, []>("op_40739_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_40739_end_0 = const()[name = tensor<string, []>("op_40739_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_40739_end_mask_0 = const()[name = tensor<string, []>("op_40739_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_40739_cast_fp16 = slice_by_index(begin = var_40739_begin_0, end = var_40739_end_0, end_mask = var_40739_end_mask_0, x = k_363_cast_fp16)[name = tensor<string, []>("op_40739_cast_fp16")];
            tensor<int32, [4]> var_40743_begin_0 = const()[name = tensor<string, []>("op_40743_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_40743_end_0 = const()[name = tensor<string, []>("op_40743_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_40743_end_mask_0 = const()[name = tensor<string, []>("op_40743_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_40743_cast_fp16 = slice_by_index(begin = var_40743_begin_0, end = var_40743_end_0, end_mask = var_40743_end_mask_0, x = k_363_cast_fp16)[name = tensor<string, []>("op_40743_cast_fp16")];
            tensor<int32, [4]> var_40747_begin_0 = const()[name = tensor<string, []>("op_40747_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_40747_end_0 = const()[name = tensor<string, []>("op_40747_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_40747_end_mask_0 = const()[name = tensor<string, []>("op_40747_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_40747_cast_fp16 = slice_by_index(begin = var_40747_begin_0, end = var_40747_end_0, end_mask = var_40747_end_mask_0, x = k_363_cast_fp16)[name = tensor<string, []>("op_40747_cast_fp16")];
            tensor<int32, [4]> var_40751_begin_0 = const()[name = tensor<string, []>("op_40751_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_40751_end_0 = const()[name = tensor<string, []>("op_40751_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_40751_end_mask_0 = const()[name = tensor<string, []>("op_40751_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_40751_cast_fp16 = slice_by_index(begin = var_40751_begin_0, end = var_40751_end_0, end_mask = var_40751_end_mask_0, x = k_363_cast_fp16)[name = tensor<string, []>("op_40751_cast_fp16")];
            tensor<int32, [4]> var_40755_begin_0 = const()[name = tensor<string, []>("op_40755_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_40755_end_0 = const()[name = tensor<string, []>("op_40755_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_40755_end_mask_0 = const()[name = tensor<string, []>("op_40755_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_40755_cast_fp16 = slice_by_index(begin = var_40755_begin_0, end = var_40755_end_0, end_mask = var_40755_end_mask_0, x = k_363_cast_fp16)[name = tensor<string, []>("op_40755_cast_fp16")];
            tensor<int32, [4]> var_40759_begin_0 = const()[name = tensor<string, []>("op_40759_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_40759_end_0 = const()[name = tensor<string, []>("op_40759_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_40759_end_mask_0 = const()[name = tensor<string, []>("op_40759_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_40759_cast_fp16 = slice_by_index(begin = var_40759_begin_0, end = var_40759_end_0, end_mask = var_40759_end_mask_0, x = k_363_cast_fp16)[name = tensor<string, []>("op_40759_cast_fp16")];
            tensor<int32, [4]> var_40763_begin_0 = const()[name = tensor<string, []>("op_40763_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_40763_end_0 = const()[name = tensor<string, []>("op_40763_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_40763_end_mask_0 = const()[name = tensor<string, []>("op_40763_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_40763_cast_fp16 = slice_by_index(begin = var_40763_begin_0, end = var_40763_end_0, end_mask = var_40763_end_mask_0, x = k_363_cast_fp16)[name = tensor<string, []>("op_40763_cast_fp16")];
            tensor<int32, [4]> var_40767_begin_0 = const()[name = tensor<string, []>("op_40767_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_40767_end_0 = const()[name = tensor<string, []>("op_40767_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_40767_end_mask_0 = const()[name = tensor<string, []>("op_40767_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_40767_cast_fp16 = slice_by_index(begin = var_40767_begin_0, end = var_40767_end_0, end_mask = var_40767_end_mask_0, x = k_363_cast_fp16)[name = tensor<string, []>("op_40767_cast_fp16")];
            tensor<int32, [4]> var_40771_begin_0 = const()[name = tensor<string, []>("op_40771_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_40771_end_0 = const()[name = tensor<string, []>("op_40771_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_40771_end_mask_0 = const()[name = tensor<string, []>("op_40771_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_40771_cast_fp16 = slice_by_index(begin = var_40771_begin_0, end = var_40771_end_0, end_mask = var_40771_end_mask_0, x = k_363_cast_fp16)[name = tensor<string, []>("op_40771_cast_fp16")];
            tensor<int32, [4]> var_40775_begin_0 = const()[name = tensor<string, []>("op_40775_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_40775_end_0 = const()[name = tensor<string, []>("op_40775_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_40775_end_mask_0 = const()[name = tensor<string, []>("op_40775_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_40775_cast_fp16 = slice_by_index(begin = var_40775_begin_0, end = var_40775_end_0, end_mask = var_40775_end_mask_0, x = k_363_cast_fp16)[name = tensor<string, []>("op_40775_cast_fp16")];
            tensor<int32, [4]> var_40779_begin_0 = const()[name = tensor<string, []>("op_40779_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_40779_end_0 = const()[name = tensor<string, []>("op_40779_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_40779_end_mask_0 = const()[name = tensor<string, []>("op_40779_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_40779_cast_fp16 = slice_by_index(begin = var_40779_begin_0, end = var_40779_end_0, end_mask = var_40779_end_mask_0, x = k_363_cast_fp16)[name = tensor<string, []>("op_40779_cast_fp16")];
            tensor<int32, [4]> var_40783_begin_0 = const()[name = tensor<string, []>("op_40783_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_40783_end_0 = const()[name = tensor<string, []>("op_40783_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_40783_end_mask_0 = const()[name = tensor<string, []>("op_40783_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_40783_cast_fp16 = slice_by_index(begin = var_40783_begin_0, end = var_40783_end_0, end_mask = var_40783_end_mask_0, x = k_363_cast_fp16)[name = tensor<string, []>("op_40783_cast_fp16")];
            tensor<int32, [4]> var_40787_begin_0 = const()[name = tensor<string, []>("op_40787_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_40787_end_0 = const()[name = tensor<string, []>("op_40787_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_40787_end_mask_0 = const()[name = tensor<string, []>("op_40787_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_40787_cast_fp16 = slice_by_index(begin = var_40787_begin_0, end = var_40787_end_0, end_mask = var_40787_end_mask_0, x = k_363_cast_fp16)[name = tensor<string, []>("op_40787_cast_fp16")];
            tensor<int32, [4]> var_40789_begin_0 = const()[name = tensor<string, []>("op_40789_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_40789_end_0 = const()[name = tensor<string, []>("op_40789_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_40789_end_mask_0 = const()[name = tensor<string, []>("op_40789_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40789_cast_fp16 = slice_by_index(begin = var_40789_begin_0, end = var_40789_end_0, end_mask = var_40789_end_mask_0, x = v_181_cast_fp16)[name = tensor<string, []>("op_40789_cast_fp16")];
            tensor<int32, [4]> var_40793_begin_0 = const()[name = tensor<string, []>("op_40793_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_40793_end_0 = const()[name = tensor<string, []>("op_40793_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_40793_end_mask_0 = const()[name = tensor<string, []>("op_40793_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40793_cast_fp16 = slice_by_index(begin = var_40793_begin_0, end = var_40793_end_0, end_mask = var_40793_end_mask_0, x = v_181_cast_fp16)[name = tensor<string, []>("op_40793_cast_fp16")];
            tensor<int32, [4]> var_40797_begin_0 = const()[name = tensor<string, []>("op_40797_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_40797_end_0 = const()[name = tensor<string, []>("op_40797_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_40797_end_mask_0 = const()[name = tensor<string, []>("op_40797_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40797_cast_fp16 = slice_by_index(begin = var_40797_begin_0, end = var_40797_end_0, end_mask = var_40797_end_mask_0, x = v_181_cast_fp16)[name = tensor<string, []>("op_40797_cast_fp16")];
            tensor<int32, [4]> var_40801_begin_0 = const()[name = tensor<string, []>("op_40801_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_40801_end_0 = const()[name = tensor<string, []>("op_40801_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_40801_end_mask_0 = const()[name = tensor<string, []>("op_40801_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40801_cast_fp16 = slice_by_index(begin = var_40801_begin_0, end = var_40801_end_0, end_mask = var_40801_end_mask_0, x = v_181_cast_fp16)[name = tensor<string, []>("op_40801_cast_fp16")];
            tensor<int32, [4]> var_40805_begin_0 = const()[name = tensor<string, []>("op_40805_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_40805_end_0 = const()[name = tensor<string, []>("op_40805_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_40805_end_mask_0 = const()[name = tensor<string, []>("op_40805_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40805_cast_fp16 = slice_by_index(begin = var_40805_begin_0, end = var_40805_end_0, end_mask = var_40805_end_mask_0, x = v_181_cast_fp16)[name = tensor<string, []>("op_40805_cast_fp16")];
            tensor<int32, [4]> var_40809_begin_0 = const()[name = tensor<string, []>("op_40809_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_40809_end_0 = const()[name = tensor<string, []>("op_40809_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_40809_end_mask_0 = const()[name = tensor<string, []>("op_40809_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40809_cast_fp16 = slice_by_index(begin = var_40809_begin_0, end = var_40809_end_0, end_mask = var_40809_end_mask_0, x = v_181_cast_fp16)[name = tensor<string, []>("op_40809_cast_fp16")];
            tensor<int32, [4]> var_40813_begin_0 = const()[name = tensor<string, []>("op_40813_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_40813_end_0 = const()[name = tensor<string, []>("op_40813_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_40813_end_mask_0 = const()[name = tensor<string, []>("op_40813_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40813_cast_fp16 = slice_by_index(begin = var_40813_begin_0, end = var_40813_end_0, end_mask = var_40813_end_mask_0, x = v_181_cast_fp16)[name = tensor<string, []>("op_40813_cast_fp16")];
            tensor<int32, [4]> var_40817_begin_0 = const()[name = tensor<string, []>("op_40817_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_40817_end_0 = const()[name = tensor<string, []>("op_40817_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_40817_end_mask_0 = const()[name = tensor<string, []>("op_40817_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40817_cast_fp16 = slice_by_index(begin = var_40817_begin_0, end = var_40817_end_0, end_mask = var_40817_end_mask_0, x = v_181_cast_fp16)[name = tensor<string, []>("op_40817_cast_fp16")];
            tensor<int32, [4]> var_40821_begin_0 = const()[name = tensor<string, []>("op_40821_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_40821_end_0 = const()[name = tensor<string, []>("op_40821_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_40821_end_mask_0 = const()[name = tensor<string, []>("op_40821_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40821_cast_fp16 = slice_by_index(begin = var_40821_begin_0, end = var_40821_end_0, end_mask = var_40821_end_mask_0, x = v_181_cast_fp16)[name = tensor<string, []>("op_40821_cast_fp16")];
            tensor<int32, [4]> var_40825_begin_0 = const()[name = tensor<string, []>("op_40825_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_40825_end_0 = const()[name = tensor<string, []>("op_40825_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_40825_end_mask_0 = const()[name = tensor<string, []>("op_40825_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40825_cast_fp16 = slice_by_index(begin = var_40825_begin_0, end = var_40825_end_0, end_mask = var_40825_end_mask_0, x = v_181_cast_fp16)[name = tensor<string, []>("op_40825_cast_fp16")];
            tensor<int32, [4]> var_40829_begin_0 = const()[name = tensor<string, []>("op_40829_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_40829_end_0 = const()[name = tensor<string, []>("op_40829_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_40829_end_mask_0 = const()[name = tensor<string, []>("op_40829_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40829_cast_fp16 = slice_by_index(begin = var_40829_begin_0, end = var_40829_end_0, end_mask = var_40829_end_mask_0, x = v_181_cast_fp16)[name = tensor<string, []>("op_40829_cast_fp16")];
            tensor<int32, [4]> var_40833_begin_0 = const()[name = tensor<string, []>("op_40833_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_40833_end_0 = const()[name = tensor<string, []>("op_40833_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_40833_end_mask_0 = const()[name = tensor<string, []>("op_40833_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40833_cast_fp16 = slice_by_index(begin = var_40833_begin_0, end = var_40833_end_0, end_mask = var_40833_end_mask_0, x = v_181_cast_fp16)[name = tensor<string, []>("op_40833_cast_fp16")];
            tensor<int32, [4]> var_40837_begin_0 = const()[name = tensor<string, []>("op_40837_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_40837_end_0 = const()[name = tensor<string, []>("op_40837_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_40837_end_mask_0 = const()[name = tensor<string, []>("op_40837_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40837_cast_fp16 = slice_by_index(begin = var_40837_begin_0, end = var_40837_end_0, end_mask = var_40837_end_mask_0, x = v_181_cast_fp16)[name = tensor<string, []>("op_40837_cast_fp16")];
            tensor<int32, [4]> var_40841_begin_0 = const()[name = tensor<string, []>("op_40841_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_40841_end_0 = const()[name = tensor<string, []>("op_40841_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_40841_end_mask_0 = const()[name = tensor<string, []>("op_40841_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40841_cast_fp16 = slice_by_index(begin = var_40841_begin_0, end = var_40841_end_0, end_mask = var_40841_end_mask_0, x = v_181_cast_fp16)[name = tensor<string, []>("op_40841_cast_fp16")];
            tensor<int32, [4]> var_40845_begin_0 = const()[name = tensor<string, []>("op_40845_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_40845_end_0 = const()[name = tensor<string, []>("op_40845_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_40845_end_mask_0 = const()[name = tensor<string, []>("op_40845_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40845_cast_fp16 = slice_by_index(begin = var_40845_begin_0, end = var_40845_end_0, end_mask = var_40845_end_mask_0, x = v_181_cast_fp16)[name = tensor<string, []>("op_40845_cast_fp16")];
            tensor<int32, [4]> var_40849_begin_0 = const()[name = tensor<string, []>("op_40849_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_40849_end_0 = const()[name = tensor<string, []>("op_40849_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_40849_end_mask_0 = const()[name = tensor<string, []>("op_40849_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40849_cast_fp16 = slice_by_index(begin = var_40849_begin_0, end = var_40849_end_0, end_mask = var_40849_end_mask_0, x = v_181_cast_fp16)[name = tensor<string, []>("op_40849_cast_fp16")];
            tensor<int32, [4]> var_40853_begin_0 = const()[name = tensor<string, []>("op_40853_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_40853_end_0 = const()[name = tensor<string, []>("op_40853_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_40853_end_mask_0 = const()[name = tensor<string, []>("op_40853_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40853_cast_fp16 = slice_by_index(begin = var_40853_begin_0, end = var_40853_end_0, end_mask = var_40853_end_mask_0, x = v_181_cast_fp16)[name = tensor<string, []>("op_40853_cast_fp16")];
            tensor<int32, [4]> var_40857_begin_0 = const()[name = tensor<string, []>("op_40857_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_40857_end_0 = const()[name = tensor<string, []>("op_40857_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_40857_end_mask_0 = const()[name = tensor<string, []>("op_40857_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40857_cast_fp16 = slice_by_index(begin = var_40857_begin_0, end = var_40857_end_0, end_mask = var_40857_end_mask_0, x = v_181_cast_fp16)[name = tensor<string, []>("op_40857_cast_fp16")];
            tensor<int32, [4]> var_40861_begin_0 = const()[name = tensor<string, []>("op_40861_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_40861_end_0 = const()[name = tensor<string, []>("op_40861_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_40861_end_mask_0 = const()[name = tensor<string, []>("op_40861_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40861_cast_fp16 = slice_by_index(begin = var_40861_begin_0, end = var_40861_end_0, end_mask = var_40861_end_mask_0, x = v_181_cast_fp16)[name = tensor<string, []>("op_40861_cast_fp16")];
            tensor<int32, [4]> var_40865_begin_0 = const()[name = tensor<string, []>("op_40865_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_40865_end_0 = const()[name = tensor<string, []>("op_40865_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_40865_end_mask_0 = const()[name = tensor<string, []>("op_40865_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40865_cast_fp16 = slice_by_index(begin = var_40865_begin_0, end = var_40865_end_0, end_mask = var_40865_end_mask_0, x = v_181_cast_fp16)[name = tensor<string, []>("op_40865_cast_fp16")];
            tensor<string, []> var_40869_equation_0 = const()[name = tensor<string, []>("op_40869_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40869_cast_fp16 = einsum(equation = var_40869_equation_0, values = (var_40711_cast_fp16, var_40628_cast_fp16))[name = tensor<string, []>("op_40869_cast_fp16")];
            tensor<fp16, []> var_40870_to_fp16 = const()[name = tensor<string, []>("op_40870_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3441_cast_fp16 = mul(x = var_40869_cast_fp16, y = var_40870_to_fp16)[name = tensor<string, []>("aw_3441_cast_fp16")];
            tensor<string, []> var_40873_equation_0 = const()[name = tensor<string, []>("op_40873_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40873_cast_fp16 = einsum(equation = var_40873_equation_0, values = (var_40715_cast_fp16, var_40632_cast_fp16))[name = tensor<string, []>("op_40873_cast_fp16")];
            tensor<fp16, []> var_40874_to_fp16 = const()[name = tensor<string, []>("op_40874_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3443_cast_fp16 = mul(x = var_40873_cast_fp16, y = var_40874_to_fp16)[name = tensor<string, []>("aw_3443_cast_fp16")];
            tensor<string, []> var_40877_equation_0 = const()[name = tensor<string, []>("op_40877_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40877_cast_fp16 = einsum(equation = var_40877_equation_0, values = (var_40719_cast_fp16, var_40636_cast_fp16))[name = tensor<string, []>("op_40877_cast_fp16")];
            tensor<fp16, []> var_40878_to_fp16 = const()[name = tensor<string, []>("op_40878_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3445_cast_fp16 = mul(x = var_40877_cast_fp16, y = var_40878_to_fp16)[name = tensor<string, []>("aw_3445_cast_fp16")];
            tensor<string, []> var_40881_equation_0 = const()[name = tensor<string, []>("op_40881_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40881_cast_fp16 = einsum(equation = var_40881_equation_0, values = (var_40723_cast_fp16, var_40640_cast_fp16))[name = tensor<string, []>("op_40881_cast_fp16")];
            tensor<fp16, []> var_40882_to_fp16 = const()[name = tensor<string, []>("op_40882_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3447_cast_fp16 = mul(x = var_40881_cast_fp16, y = var_40882_to_fp16)[name = tensor<string, []>("aw_3447_cast_fp16")];
            tensor<string, []> var_40885_equation_0 = const()[name = tensor<string, []>("op_40885_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40885_cast_fp16 = einsum(equation = var_40885_equation_0, values = (var_40727_cast_fp16, var_40644_cast_fp16))[name = tensor<string, []>("op_40885_cast_fp16")];
            tensor<fp16, []> var_40886_to_fp16 = const()[name = tensor<string, []>("op_40886_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3449_cast_fp16 = mul(x = var_40885_cast_fp16, y = var_40886_to_fp16)[name = tensor<string, []>("aw_3449_cast_fp16")];
            tensor<string, []> var_40889_equation_0 = const()[name = tensor<string, []>("op_40889_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40889_cast_fp16 = einsum(equation = var_40889_equation_0, values = (var_40731_cast_fp16, var_40648_cast_fp16))[name = tensor<string, []>("op_40889_cast_fp16")];
            tensor<fp16, []> var_40890_to_fp16 = const()[name = tensor<string, []>("op_40890_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3451_cast_fp16 = mul(x = var_40889_cast_fp16, y = var_40890_to_fp16)[name = tensor<string, []>("aw_3451_cast_fp16")];
            tensor<string, []> var_40893_equation_0 = const()[name = tensor<string, []>("op_40893_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40893_cast_fp16 = einsum(equation = var_40893_equation_0, values = (var_40735_cast_fp16, var_40652_cast_fp16))[name = tensor<string, []>("op_40893_cast_fp16")];
            tensor<fp16, []> var_40894_to_fp16 = const()[name = tensor<string, []>("op_40894_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3453_cast_fp16 = mul(x = var_40893_cast_fp16, y = var_40894_to_fp16)[name = tensor<string, []>("aw_3453_cast_fp16")];
            tensor<string, []> var_40897_equation_0 = const()[name = tensor<string, []>("op_40897_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40897_cast_fp16 = einsum(equation = var_40897_equation_0, values = (var_40739_cast_fp16, var_40656_cast_fp16))[name = tensor<string, []>("op_40897_cast_fp16")];
            tensor<fp16, []> var_40898_to_fp16 = const()[name = tensor<string, []>("op_40898_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3455_cast_fp16 = mul(x = var_40897_cast_fp16, y = var_40898_to_fp16)[name = tensor<string, []>("aw_3455_cast_fp16")];
            tensor<string, []> var_40901_equation_0 = const()[name = tensor<string, []>("op_40901_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40901_cast_fp16 = einsum(equation = var_40901_equation_0, values = (var_40743_cast_fp16, var_40660_cast_fp16))[name = tensor<string, []>("op_40901_cast_fp16")];
            tensor<fp16, []> var_40902_to_fp16 = const()[name = tensor<string, []>("op_40902_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3457_cast_fp16 = mul(x = var_40901_cast_fp16, y = var_40902_to_fp16)[name = tensor<string, []>("aw_3457_cast_fp16")];
            tensor<string, []> var_40905_equation_0 = const()[name = tensor<string, []>("op_40905_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40905_cast_fp16 = einsum(equation = var_40905_equation_0, values = (var_40747_cast_fp16, var_40664_cast_fp16))[name = tensor<string, []>("op_40905_cast_fp16")];
            tensor<fp16, []> var_40906_to_fp16 = const()[name = tensor<string, []>("op_40906_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3459_cast_fp16 = mul(x = var_40905_cast_fp16, y = var_40906_to_fp16)[name = tensor<string, []>("aw_3459_cast_fp16")];
            tensor<string, []> var_40909_equation_0 = const()[name = tensor<string, []>("op_40909_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40909_cast_fp16 = einsum(equation = var_40909_equation_0, values = (var_40751_cast_fp16, var_40668_cast_fp16))[name = tensor<string, []>("op_40909_cast_fp16")];
            tensor<fp16, []> var_40910_to_fp16 = const()[name = tensor<string, []>("op_40910_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3461_cast_fp16 = mul(x = var_40909_cast_fp16, y = var_40910_to_fp16)[name = tensor<string, []>("aw_3461_cast_fp16")];
            tensor<string, []> var_40913_equation_0 = const()[name = tensor<string, []>("op_40913_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40913_cast_fp16 = einsum(equation = var_40913_equation_0, values = (var_40755_cast_fp16, var_40672_cast_fp16))[name = tensor<string, []>("op_40913_cast_fp16")];
            tensor<fp16, []> var_40914_to_fp16 = const()[name = tensor<string, []>("op_40914_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3463_cast_fp16 = mul(x = var_40913_cast_fp16, y = var_40914_to_fp16)[name = tensor<string, []>("aw_3463_cast_fp16")];
            tensor<string, []> var_40917_equation_0 = const()[name = tensor<string, []>("op_40917_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40917_cast_fp16 = einsum(equation = var_40917_equation_0, values = (var_40759_cast_fp16, var_40676_cast_fp16))[name = tensor<string, []>("op_40917_cast_fp16")];
            tensor<fp16, []> var_40918_to_fp16 = const()[name = tensor<string, []>("op_40918_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3465_cast_fp16 = mul(x = var_40917_cast_fp16, y = var_40918_to_fp16)[name = tensor<string, []>("aw_3465_cast_fp16")];
            tensor<string, []> var_40921_equation_0 = const()[name = tensor<string, []>("op_40921_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40921_cast_fp16 = einsum(equation = var_40921_equation_0, values = (var_40763_cast_fp16, var_40680_cast_fp16))[name = tensor<string, []>("op_40921_cast_fp16")];
            tensor<fp16, []> var_40922_to_fp16 = const()[name = tensor<string, []>("op_40922_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3467_cast_fp16 = mul(x = var_40921_cast_fp16, y = var_40922_to_fp16)[name = tensor<string, []>("aw_3467_cast_fp16")];
            tensor<string, []> var_40925_equation_0 = const()[name = tensor<string, []>("op_40925_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40925_cast_fp16 = einsum(equation = var_40925_equation_0, values = (var_40767_cast_fp16, var_40684_cast_fp16))[name = tensor<string, []>("op_40925_cast_fp16")];
            tensor<fp16, []> var_40926_to_fp16 = const()[name = tensor<string, []>("op_40926_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3469_cast_fp16 = mul(x = var_40925_cast_fp16, y = var_40926_to_fp16)[name = tensor<string, []>("aw_3469_cast_fp16")];
            tensor<string, []> var_40929_equation_0 = const()[name = tensor<string, []>("op_40929_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40929_cast_fp16 = einsum(equation = var_40929_equation_0, values = (var_40771_cast_fp16, var_40688_cast_fp16))[name = tensor<string, []>("op_40929_cast_fp16")];
            tensor<fp16, []> var_40930_to_fp16 = const()[name = tensor<string, []>("op_40930_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3471_cast_fp16 = mul(x = var_40929_cast_fp16, y = var_40930_to_fp16)[name = tensor<string, []>("aw_3471_cast_fp16")];
            tensor<string, []> var_40933_equation_0 = const()[name = tensor<string, []>("op_40933_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40933_cast_fp16 = einsum(equation = var_40933_equation_0, values = (var_40775_cast_fp16, var_40692_cast_fp16))[name = tensor<string, []>("op_40933_cast_fp16")];
            tensor<fp16, []> var_40934_to_fp16 = const()[name = tensor<string, []>("op_40934_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3473_cast_fp16 = mul(x = var_40933_cast_fp16, y = var_40934_to_fp16)[name = tensor<string, []>("aw_3473_cast_fp16")];
            tensor<string, []> var_40937_equation_0 = const()[name = tensor<string, []>("op_40937_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40937_cast_fp16 = einsum(equation = var_40937_equation_0, values = (var_40779_cast_fp16, var_40696_cast_fp16))[name = tensor<string, []>("op_40937_cast_fp16")];
            tensor<fp16, []> var_40938_to_fp16 = const()[name = tensor<string, []>("op_40938_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3475_cast_fp16 = mul(x = var_40937_cast_fp16, y = var_40938_to_fp16)[name = tensor<string, []>("aw_3475_cast_fp16")];
            tensor<string, []> var_40941_equation_0 = const()[name = tensor<string, []>("op_40941_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40941_cast_fp16 = einsum(equation = var_40941_equation_0, values = (var_40783_cast_fp16, var_40700_cast_fp16))[name = tensor<string, []>("op_40941_cast_fp16")];
            tensor<fp16, []> var_40942_to_fp16 = const()[name = tensor<string, []>("op_40942_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3477_cast_fp16 = mul(x = var_40941_cast_fp16, y = var_40942_to_fp16)[name = tensor<string, []>("aw_3477_cast_fp16")];
            tensor<string, []> var_40945_equation_0 = const()[name = tensor<string, []>("op_40945_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40945_cast_fp16 = einsum(equation = var_40945_equation_0, values = (var_40787_cast_fp16, var_40704_cast_fp16))[name = tensor<string, []>("op_40945_cast_fp16")];
            tensor<fp16, []> var_40946_to_fp16 = const()[name = tensor<string, []>("op_40946_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3479_cast_fp16 = mul(x = var_40945_cast_fp16, y = var_40946_to_fp16)[name = tensor<string, []>("aw_3479_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40948_cast_fp16 = softmax(axis = var_30355, x = aw_3441_cast_fp16)[name = tensor<string, []>("op_40948_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40949_cast_fp16 = softmax(axis = var_30355, x = aw_3443_cast_fp16)[name = tensor<string, []>("op_40949_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40950_cast_fp16 = softmax(axis = var_30355, x = aw_3445_cast_fp16)[name = tensor<string, []>("op_40950_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40951_cast_fp16 = softmax(axis = var_30355, x = aw_3447_cast_fp16)[name = tensor<string, []>("op_40951_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40952_cast_fp16 = softmax(axis = var_30355, x = aw_3449_cast_fp16)[name = tensor<string, []>("op_40952_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40953_cast_fp16 = softmax(axis = var_30355, x = aw_3451_cast_fp16)[name = tensor<string, []>("op_40953_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40954_cast_fp16 = softmax(axis = var_30355, x = aw_3453_cast_fp16)[name = tensor<string, []>("op_40954_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40955_cast_fp16 = softmax(axis = var_30355, x = aw_3455_cast_fp16)[name = tensor<string, []>("op_40955_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40956_cast_fp16 = softmax(axis = var_30355, x = aw_3457_cast_fp16)[name = tensor<string, []>("op_40956_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40957_cast_fp16 = softmax(axis = var_30355, x = aw_3459_cast_fp16)[name = tensor<string, []>("op_40957_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40958_cast_fp16 = softmax(axis = var_30355, x = aw_3461_cast_fp16)[name = tensor<string, []>("op_40958_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40959_cast_fp16 = softmax(axis = var_30355, x = aw_3463_cast_fp16)[name = tensor<string, []>("op_40959_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40960_cast_fp16 = softmax(axis = var_30355, x = aw_3465_cast_fp16)[name = tensor<string, []>("op_40960_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40961_cast_fp16 = softmax(axis = var_30355, x = aw_3467_cast_fp16)[name = tensor<string, []>("op_40961_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40962_cast_fp16 = softmax(axis = var_30355, x = aw_3469_cast_fp16)[name = tensor<string, []>("op_40962_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40963_cast_fp16 = softmax(axis = var_30355, x = aw_3471_cast_fp16)[name = tensor<string, []>("op_40963_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40964_cast_fp16 = softmax(axis = var_30355, x = aw_3473_cast_fp16)[name = tensor<string, []>("op_40964_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40965_cast_fp16 = softmax(axis = var_30355, x = aw_3475_cast_fp16)[name = tensor<string, []>("op_40965_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40966_cast_fp16 = softmax(axis = var_30355, x = aw_3477_cast_fp16)[name = tensor<string, []>("op_40966_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40967_cast_fp16 = softmax(axis = var_30355, x = aw_3479_cast_fp16)[name = tensor<string, []>("op_40967_cast_fp16")];
            tensor<string, []> var_40969_equation_0 = const()[name = tensor<string, []>("op_40969_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40969_cast_fp16 = einsum(equation = var_40969_equation_0, values = (var_40789_cast_fp16, var_40948_cast_fp16))[name = tensor<string, []>("op_40969_cast_fp16")];
            tensor<string, []> var_40971_equation_0 = const()[name = tensor<string, []>("op_40971_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40971_cast_fp16 = einsum(equation = var_40971_equation_0, values = (var_40793_cast_fp16, var_40949_cast_fp16))[name = tensor<string, []>("op_40971_cast_fp16")];
            tensor<string, []> var_40973_equation_0 = const()[name = tensor<string, []>("op_40973_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40973_cast_fp16 = einsum(equation = var_40973_equation_0, values = (var_40797_cast_fp16, var_40950_cast_fp16))[name = tensor<string, []>("op_40973_cast_fp16")];
            tensor<string, []> var_40975_equation_0 = const()[name = tensor<string, []>("op_40975_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40975_cast_fp16 = einsum(equation = var_40975_equation_0, values = (var_40801_cast_fp16, var_40951_cast_fp16))[name = tensor<string, []>("op_40975_cast_fp16")];
            tensor<string, []> var_40977_equation_0 = const()[name = tensor<string, []>("op_40977_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40977_cast_fp16 = einsum(equation = var_40977_equation_0, values = (var_40805_cast_fp16, var_40952_cast_fp16))[name = tensor<string, []>("op_40977_cast_fp16")];
            tensor<string, []> var_40979_equation_0 = const()[name = tensor<string, []>("op_40979_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40979_cast_fp16 = einsum(equation = var_40979_equation_0, values = (var_40809_cast_fp16, var_40953_cast_fp16))[name = tensor<string, []>("op_40979_cast_fp16")];
            tensor<string, []> var_40981_equation_0 = const()[name = tensor<string, []>("op_40981_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40981_cast_fp16 = einsum(equation = var_40981_equation_0, values = (var_40813_cast_fp16, var_40954_cast_fp16))[name = tensor<string, []>("op_40981_cast_fp16")];
            tensor<string, []> var_40983_equation_0 = const()[name = tensor<string, []>("op_40983_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40983_cast_fp16 = einsum(equation = var_40983_equation_0, values = (var_40817_cast_fp16, var_40955_cast_fp16))[name = tensor<string, []>("op_40983_cast_fp16")];
            tensor<string, []> var_40985_equation_0 = const()[name = tensor<string, []>("op_40985_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40985_cast_fp16 = einsum(equation = var_40985_equation_0, values = (var_40821_cast_fp16, var_40956_cast_fp16))[name = tensor<string, []>("op_40985_cast_fp16")];
            tensor<string, []> var_40987_equation_0 = const()[name = tensor<string, []>("op_40987_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40987_cast_fp16 = einsum(equation = var_40987_equation_0, values = (var_40825_cast_fp16, var_40957_cast_fp16))[name = tensor<string, []>("op_40987_cast_fp16")];
            tensor<string, []> var_40989_equation_0 = const()[name = tensor<string, []>("op_40989_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40989_cast_fp16 = einsum(equation = var_40989_equation_0, values = (var_40829_cast_fp16, var_40958_cast_fp16))[name = tensor<string, []>("op_40989_cast_fp16")];
            tensor<string, []> var_40991_equation_0 = const()[name = tensor<string, []>("op_40991_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40991_cast_fp16 = einsum(equation = var_40991_equation_0, values = (var_40833_cast_fp16, var_40959_cast_fp16))[name = tensor<string, []>("op_40991_cast_fp16")];
            tensor<string, []> var_40993_equation_0 = const()[name = tensor<string, []>("op_40993_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40993_cast_fp16 = einsum(equation = var_40993_equation_0, values = (var_40837_cast_fp16, var_40960_cast_fp16))[name = tensor<string, []>("op_40993_cast_fp16")];
            tensor<string, []> var_40995_equation_0 = const()[name = tensor<string, []>("op_40995_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40995_cast_fp16 = einsum(equation = var_40995_equation_0, values = (var_40841_cast_fp16, var_40961_cast_fp16))[name = tensor<string, []>("op_40995_cast_fp16")];
            tensor<string, []> var_40997_equation_0 = const()[name = tensor<string, []>("op_40997_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40997_cast_fp16 = einsum(equation = var_40997_equation_0, values = (var_40845_cast_fp16, var_40962_cast_fp16))[name = tensor<string, []>("op_40997_cast_fp16")];
            tensor<string, []> var_40999_equation_0 = const()[name = tensor<string, []>("op_40999_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40999_cast_fp16 = einsum(equation = var_40999_equation_0, values = (var_40849_cast_fp16, var_40963_cast_fp16))[name = tensor<string, []>("op_40999_cast_fp16")];
            tensor<string, []> var_41001_equation_0 = const()[name = tensor<string, []>("op_41001_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41001_cast_fp16 = einsum(equation = var_41001_equation_0, values = (var_40853_cast_fp16, var_40964_cast_fp16))[name = tensor<string, []>("op_41001_cast_fp16")];
            tensor<string, []> var_41003_equation_0 = const()[name = tensor<string, []>("op_41003_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41003_cast_fp16 = einsum(equation = var_41003_equation_0, values = (var_40857_cast_fp16, var_40965_cast_fp16))[name = tensor<string, []>("op_41003_cast_fp16")];
            tensor<string, []> var_41005_equation_0 = const()[name = tensor<string, []>("op_41005_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41005_cast_fp16 = einsum(equation = var_41005_equation_0, values = (var_40861_cast_fp16, var_40966_cast_fp16))[name = tensor<string, []>("op_41005_cast_fp16")];
            tensor<string, []> var_41007_equation_0 = const()[name = tensor<string, []>("op_41007_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41007_cast_fp16 = einsum(equation = var_41007_equation_0, values = (var_40865_cast_fp16, var_40967_cast_fp16))[name = tensor<string, []>("op_41007_cast_fp16")];
            tensor<bool, []> input_545_interleave_0 = const()[name = tensor<string, []>("input_545_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_545_cast_fp16 = concat(axis = var_30355, interleave = input_545_interleave_0, values = (var_40969_cast_fp16, var_40971_cast_fp16, var_40973_cast_fp16, var_40975_cast_fp16, var_40977_cast_fp16, var_40979_cast_fp16, var_40981_cast_fp16, var_40983_cast_fp16, var_40985_cast_fp16, var_40987_cast_fp16, var_40989_cast_fp16, var_40991_cast_fp16, var_40993_cast_fp16, var_40995_cast_fp16, var_40997_cast_fp16, var_40999_cast_fp16, var_41001_cast_fp16, var_41003_cast_fp16, var_41005_cast_fp16, var_41007_cast_fp16))[name = tensor<string, []>("input_545_cast_fp16")];
            tensor<string, []> var_41017_pad_type_0 = const()[name = tensor<string, []>("op_41017_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_41017_strides_0 = const()[name = tensor<string, []>("op_41017_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_41017_pad_0 = const()[name = tensor<string, []>("op_41017_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_41017_dilations_0 = const()[name = tensor<string, []>("op_41017_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_41017_groups_0 = const()[name = tensor<string, []>("op_41017_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_1_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(333839424))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(335068288))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_1_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_1_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_1_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(335068480)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_41017_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_1_attn1_to_out_0_bias_to_fp16, dilations = var_41017_dilations_0, groups = var_41017_groups_0, pad = var_41017_pad_0, pad_type = var_41017_pad_type_0, strides = var_41017_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_1_attn1_to_out_0_weight_to_fp16_palettized, x = input_545_cast_fp16)[name = tensor<string, []>("op_41017_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_273_cast_fp16 = add(x = var_41017_cast_fp16, y = inputs_271_cast_fp16)[name = tensor<string, []>("inputs_273_cast_fp16")];
            tensor<int32, [1]> hidden_states_369_axes_0 = const()[name = tensor<string, []>("hidden_states_369_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_369_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_369_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(335071104)))];
            tensor<fp16, [1280]> hidden_states_369_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_369_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(335073728)))];
            tensor<fp16, []> var_41027_to_fp16 = const()[name = tensor<string, []>("op_41027_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_369_cast_fp16 = layer_norm(axes = hidden_states_369_axes_0, beta = hidden_states_369_beta_0_to_fp16, epsilon = var_41027_to_fp16, gamma = hidden_states_369_gamma_0_to_fp16, x = inputs_273_cast_fp16)[name = tensor<string, []>("hidden_states_369_cast_fp16")];
            tensor<string, []> q_183_pad_type_0 = const()[name = tensor<string, []>("q_183_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_183_strides_0 = const()[name = tensor<string, []>("q_183_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_183_pad_0 = const()[name = tensor<string, []>("q_183_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_183_dilations_0 = const()[name = tensor<string, []>("q_183_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_183_groups_0 = const()[name = tensor<string, []>("q_183_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_1_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(335076352))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(336305216))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_1_attn2_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_183_cast_fp16 = conv(dilations = q_183_dilations_0, groups = q_183_groups_0, pad = q_183_pad_0, pad_type = q_183_pad_type_0, strides = q_183_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_1_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_369_cast_fp16)[name = tensor<string, []>("q_183_cast_fp16")];
            tensor<string, []> k_365_pad_type_0 = const()[name = tensor<string, []>("k_365_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_365_strides_0 = const()[name = tensor<string, []>("k_365_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_365_pad_0 = const()[name = tensor<string, []>("k_365_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_365_dilations_0 = const()[name = tensor<string, []>("k_365_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_365_groups_0 = const()[name = tensor<string, []>("k_365_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_1_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(336305408))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(338271552))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_1_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> k_365_cast_fp16 = conv(dilations = k_365_dilations_0, groups = k_365_groups_0, pad = k_365_pad_0, pad_type = k_365_pad_type_0, strides = k_365_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_1_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("k_365_cast_fp16")];
            tensor<string, []> v_183_pad_type_0 = const()[name = tensor<string, []>("v_183_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_183_strides_0 = const()[name = tensor<string, []>("v_183_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_183_pad_0 = const()[name = tensor<string, []>("v_183_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_183_dilations_0 = const()[name = tensor<string, []>("v_183_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_183_groups_0 = const()[name = tensor<string, []>("v_183_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_1_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(338271744))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(340237888))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_1_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> v_183_cast_fp16 = conv(dilations = v_183_dilations_0, groups = v_183_groups_0, pad = v_183_pad_0, pad_type = v_183_pad_type_0, strides = v_183_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_1_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("v_183_cast_fp16")];
            tensor<int32, [4]> var_41060_begin_0 = const()[name = tensor<string, []>("op_41060_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_41060_end_0 = const()[name = tensor<string, []>("op_41060_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_41060_end_mask_0 = const()[name = tensor<string, []>("op_41060_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41060_cast_fp16 = slice_by_index(begin = var_41060_begin_0, end = var_41060_end_0, end_mask = var_41060_end_mask_0, x = q_183_cast_fp16)[name = tensor<string, []>("op_41060_cast_fp16")];
            tensor<int32, [4]> var_41064_begin_0 = const()[name = tensor<string, []>("op_41064_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_41064_end_0 = const()[name = tensor<string, []>("op_41064_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_41064_end_mask_0 = const()[name = tensor<string, []>("op_41064_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41064_cast_fp16 = slice_by_index(begin = var_41064_begin_0, end = var_41064_end_0, end_mask = var_41064_end_mask_0, x = q_183_cast_fp16)[name = tensor<string, []>("op_41064_cast_fp16")];
            tensor<int32, [4]> var_41068_begin_0 = const()[name = tensor<string, []>("op_41068_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_41068_end_0 = const()[name = tensor<string, []>("op_41068_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_41068_end_mask_0 = const()[name = tensor<string, []>("op_41068_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41068_cast_fp16 = slice_by_index(begin = var_41068_begin_0, end = var_41068_end_0, end_mask = var_41068_end_mask_0, x = q_183_cast_fp16)[name = tensor<string, []>("op_41068_cast_fp16")];
            tensor<int32, [4]> var_41072_begin_0 = const()[name = tensor<string, []>("op_41072_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_41072_end_0 = const()[name = tensor<string, []>("op_41072_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_41072_end_mask_0 = const()[name = tensor<string, []>("op_41072_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41072_cast_fp16 = slice_by_index(begin = var_41072_begin_0, end = var_41072_end_0, end_mask = var_41072_end_mask_0, x = q_183_cast_fp16)[name = tensor<string, []>("op_41072_cast_fp16")];
            tensor<int32, [4]> var_41076_begin_0 = const()[name = tensor<string, []>("op_41076_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_41076_end_0 = const()[name = tensor<string, []>("op_41076_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_41076_end_mask_0 = const()[name = tensor<string, []>("op_41076_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41076_cast_fp16 = slice_by_index(begin = var_41076_begin_0, end = var_41076_end_0, end_mask = var_41076_end_mask_0, x = q_183_cast_fp16)[name = tensor<string, []>("op_41076_cast_fp16")];
            tensor<int32, [4]> var_41080_begin_0 = const()[name = tensor<string, []>("op_41080_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_41080_end_0 = const()[name = tensor<string, []>("op_41080_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_41080_end_mask_0 = const()[name = tensor<string, []>("op_41080_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41080_cast_fp16 = slice_by_index(begin = var_41080_begin_0, end = var_41080_end_0, end_mask = var_41080_end_mask_0, x = q_183_cast_fp16)[name = tensor<string, []>("op_41080_cast_fp16")];
            tensor<int32, [4]> var_41084_begin_0 = const()[name = tensor<string, []>("op_41084_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_41084_end_0 = const()[name = tensor<string, []>("op_41084_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_41084_end_mask_0 = const()[name = tensor<string, []>("op_41084_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41084_cast_fp16 = slice_by_index(begin = var_41084_begin_0, end = var_41084_end_0, end_mask = var_41084_end_mask_0, x = q_183_cast_fp16)[name = tensor<string, []>("op_41084_cast_fp16")];
            tensor<int32, [4]> var_41088_begin_0 = const()[name = tensor<string, []>("op_41088_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_41088_end_0 = const()[name = tensor<string, []>("op_41088_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_41088_end_mask_0 = const()[name = tensor<string, []>("op_41088_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41088_cast_fp16 = slice_by_index(begin = var_41088_begin_0, end = var_41088_end_0, end_mask = var_41088_end_mask_0, x = q_183_cast_fp16)[name = tensor<string, []>("op_41088_cast_fp16")];
            tensor<int32, [4]> var_41092_begin_0 = const()[name = tensor<string, []>("op_41092_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_41092_end_0 = const()[name = tensor<string, []>("op_41092_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_41092_end_mask_0 = const()[name = tensor<string, []>("op_41092_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41092_cast_fp16 = slice_by_index(begin = var_41092_begin_0, end = var_41092_end_0, end_mask = var_41092_end_mask_0, x = q_183_cast_fp16)[name = tensor<string, []>("op_41092_cast_fp16")];
            tensor<int32, [4]> var_41096_begin_0 = const()[name = tensor<string, []>("op_41096_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_41096_end_0 = const()[name = tensor<string, []>("op_41096_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_41096_end_mask_0 = const()[name = tensor<string, []>("op_41096_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41096_cast_fp16 = slice_by_index(begin = var_41096_begin_0, end = var_41096_end_0, end_mask = var_41096_end_mask_0, x = q_183_cast_fp16)[name = tensor<string, []>("op_41096_cast_fp16")];
            tensor<int32, [4]> var_41100_begin_0 = const()[name = tensor<string, []>("op_41100_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_41100_end_0 = const()[name = tensor<string, []>("op_41100_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_41100_end_mask_0 = const()[name = tensor<string, []>("op_41100_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41100_cast_fp16 = slice_by_index(begin = var_41100_begin_0, end = var_41100_end_0, end_mask = var_41100_end_mask_0, x = q_183_cast_fp16)[name = tensor<string, []>("op_41100_cast_fp16")];
            tensor<int32, [4]> var_41104_begin_0 = const()[name = tensor<string, []>("op_41104_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_41104_end_0 = const()[name = tensor<string, []>("op_41104_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_41104_end_mask_0 = const()[name = tensor<string, []>("op_41104_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41104_cast_fp16 = slice_by_index(begin = var_41104_begin_0, end = var_41104_end_0, end_mask = var_41104_end_mask_0, x = q_183_cast_fp16)[name = tensor<string, []>("op_41104_cast_fp16")];
            tensor<int32, [4]> var_41108_begin_0 = const()[name = tensor<string, []>("op_41108_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_41108_end_0 = const()[name = tensor<string, []>("op_41108_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_41108_end_mask_0 = const()[name = tensor<string, []>("op_41108_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41108_cast_fp16 = slice_by_index(begin = var_41108_begin_0, end = var_41108_end_0, end_mask = var_41108_end_mask_0, x = q_183_cast_fp16)[name = tensor<string, []>("op_41108_cast_fp16")];
            tensor<int32, [4]> var_41112_begin_0 = const()[name = tensor<string, []>("op_41112_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_41112_end_0 = const()[name = tensor<string, []>("op_41112_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_41112_end_mask_0 = const()[name = tensor<string, []>("op_41112_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41112_cast_fp16 = slice_by_index(begin = var_41112_begin_0, end = var_41112_end_0, end_mask = var_41112_end_mask_0, x = q_183_cast_fp16)[name = tensor<string, []>("op_41112_cast_fp16")];
            tensor<int32, [4]> var_41116_begin_0 = const()[name = tensor<string, []>("op_41116_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_41116_end_0 = const()[name = tensor<string, []>("op_41116_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_41116_end_mask_0 = const()[name = tensor<string, []>("op_41116_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41116_cast_fp16 = slice_by_index(begin = var_41116_begin_0, end = var_41116_end_0, end_mask = var_41116_end_mask_0, x = q_183_cast_fp16)[name = tensor<string, []>("op_41116_cast_fp16")];
            tensor<int32, [4]> var_41120_begin_0 = const()[name = tensor<string, []>("op_41120_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_41120_end_0 = const()[name = tensor<string, []>("op_41120_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_41120_end_mask_0 = const()[name = tensor<string, []>("op_41120_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41120_cast_fp16 = slice_by_index(begin = var_41120_begin_0, end = var_41120_end_0, end_mask = var_41120_end_mask_0, x = q_183_cast_fp16)[name = tensor<string, []>("op_41120_cast_fp16")];
            tensor<int32, [4]> var_41124_begin_0 = const()[name = tensor<string, []>("op_41124_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_41124_end_0 = const()[name = tensor<string, []>("op_41124_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_41124_end_mask_0 = const()[name = tensor<string, []>("op_41124_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41124_cast_fp16 = slice_by_index(begin = var_41124_begin_0, end = var_41124_end_0, end_mask = var_41124_end_mask_0, x = q_183_cast_fp16)[name = tensor<string, []>("op_41124_cast_fp16")];
            tensor<int32, [4]> var_41128_begin_0 = const()[name = tensor<string, []>("op_41128_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_41128_end_0 = const()[name = tensor<string, []>("op_41128_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_41128_end_mask_0 = const()[name = tensor<string, []>("op_41128_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41128_cast_fp16 = slice_by_index(begin = var_41128_begin_0, end = var_41128_end_0, end_mask = var_41128_end_mask_0, x = q_183_cast_fp16)[name = tensor<string, []>("op_41128_cast_fp16")];
            tensor<int32, [4]> var_41132_begin_0 = const()[name = tensor<string, []>("op_41132_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_41132_end_0 = const()[name = tensor<string, []>("op_41132_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_41132_end_mask_0 = const()[name = tensor<string, []>("op_41132_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41132_cast_fp16 = slice_by_index(begin = var_41132_begin_0, end = var_41132_end_0, end_mask = var_41132_end_mask_0, x = q_183_cast_fp16)[name = tensor<string, []>("op_41132_cast_fp16")];
            tensor<int32, [4]> var_41136_begin_0 = const()[name = tensor<string, []>("op_41136_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_41136_end_0 = const()[name = tensor<string, []>("op_41136_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_41136_end_mask_0 = const()[name = tensor<string, []>("op_41136_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41136_cast_fp16 = slice_by_index(begin = var_41136_begin_0, end = var_41136_end_0, end_mask = var_41136_end_mask_0, x = q_183_cast_fp16)[name = tensor<string, []>("op_41136_cast_fp16")];
            tensor<int32, [4]> k_367_perm_0 = const()[name = tensor<string, []>("k_367_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_41143_begin_0 = const()[name = tensor<string, []>("op_41143_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_41143_end_0 = const()[name = tensor<string, []>("op_41143_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_41143_end_mask_0 = const()[name = tensor<string, []>("op_41143_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> k_367_cast_fp16 = transpose(perm = k_367_perm_0, x = k_365_cast_fp16)[name = tensor<string, []>("transpose_48")];
            tensor<fp16, [2, 77, 1, 64]> var_41143_cast_fp16 = slice_by_index(begin = var_41143_begin_0, end = var_41143_end_0, end_mask = var_41143_end_mask_0, x = k_367_cast_fp16)[name = tensor<string, []>("op_41143_cast_fp16")];
            tensor<int32, [4]> var_41147_begin_0 = const()[name = tensor<string, []>("op_41147_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_41147_end_0 = const()[name = tensor<string, []>("op_41147_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_41147_end_mask_0 = const()[name = tensor<string, []>("op_41147_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_41147_cast_fp16 = slice_by_index(begin = var_41147_begin_0, end = var_41147_end_0, end_mask = var_41147_end_mask_0, x = k_367_cast_fp16)[name = tensor<string, []>("op_41147_cast_fp16")];
            tensor<int32, [4]> var_41151_begin_0 = const()[name = tensor<string, []>("op_41151_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_41151_end_0 = const()[name = tensor<string, []>("op_41151_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_41151_end_mask_0 = const()[name = tensor<string, []>("op_41151_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_41151_cast_fp16 = slice_by_index(begin = var_41151_begin_0, end = var_41151_end_0, end_mask = var_41151_end_mask_0, x = k_367_cast_fp16)[name = tensor<string, []>("op_41151_cast_fp16")];
            tensor<int32, [4]> var_41155_begin_0 = const()[name = tensor<string, []>("op_41155_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_41155_end_0 = const()[name = tensor<string, []>("op_41155_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_41155_end_mask_0 = const()[name = tensor<string, []>("op_41155_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_41155_cast_fp16 = slice_by_index(begin = var_41155_begin_0, end = var_41155_end_0, end_mask = var_41155_end_mask_0, x = k_367_cast_fp16)[name = tensor<string, []>("op_41155_cast_fp16")];
            tensor<int32, [4]> var_41159_begin_0 = const()[name = tensor<string, []>("op_41159_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_41159_end_0 = const()[name = tensor<string, []>("op_41159_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_41159_end_mask_0 = const()[name = tensor<string, []>("op_41159_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_41159_cast_fp16 = slice_by_index(begin = var_41159_begin_0, end = var_41159_end_0, end_mask = var_41159_end_mask_0, x = k_367_cast_fp16)[name = tensor<string, []>("op_41159_cast_fp16")];
            tensor<int32, [4]> var_41163_begin_0 = const()[name = tensor<string, []>("op_41163_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_41163_end_0 = const()[name = tensor<string, []>("op_41163_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_41163_end_mask_0 = const()[name = tensor<string, []>("op_41163_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_41163_cast_fp16 = slice_by_index(begin = var_41163_begin_0, end = var_41163_end_0, end_mask = var_41163_end_mask_0, x = k_367_cast_fp16)[name = tensor<string, []>("op_41163_cast_fp16")];
            tensor<int32, [4]> var_41167_begin_0 = const()[name = tensor<string, []>("op_41167_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_41167_end_0 = const()[name = tensor<string, []>("op_41167_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_41167_end_mask_0 = const()[name = tensor<string, []>("op_41167_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_41167_cast_fp16 = slice_by_index(begin = var_41167_begin_0, end = var_41167_end_0, end_mask = var_41167_end_mask_0, x = k_367_cast_fp16)[name = tensor<string, []>("op_41167_cast_fp16")];
            tensor<int32, [4]> var_41171_begin_0 = const()[name = tensor<string, []>("op_41171_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_41171_end_0 = const()[name = tensor<string, []>("op_41171_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_41171_end_mask_0 = const()[name = tensor<string, []>("op_41171_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_41171_cast_fp16 = slice_by_index(begin = var_41171_begin_0, end = var_41171_end_0, end_mask = var_41171_end_mask_0, x = k_367_cast_fp16)[name = tensor<string, []>("op_41171_cast_fp16")];
            tensor<int32, [4]> var_41175_begin_0 = const()[name = tensor<string, []>("op_41175_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_41175_end_0 = const()[name = tensor<string, []>("op_41175_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_41175_end_mask_0 = const()[name = tensor<string, []>("op_41175_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_41175_cast_fp16 = slice_by_index(begin = var_41175_begin_0, end = var_41175_end_0, end_mask = var_41175_end_mask_0, x = k_367_cast_fp16)[name = tensor<string, []>("op_41175_cast_fp16")];
            tensor<int32, [4]> var_41179_begin_0 = const()[name = tensor<string, []>("op_41179_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_41179_end_0 = const()[name = tensor<string, []>("op_41179_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_41179_end_mask_0 = const()[name = tensor<string, []>("op_41179_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_41179_cast_fp16 = slice_by_index(begin = var_41179_begin_0, end = var_41179_end_0, end_mask = var_41179_end_mask_0, x = k_367_cast_fp16)[name = tensor<string, []>("op_41179_cast_fp16")];
            tensor<int32, [4]> var_41183_begin_0 = const()[name = tensor<string, []>("op_41183_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_41183_end_0 = const()[name = tensor<string, []>("op_41183_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_41183_end_mask_0 = const()[name = tensor<string, []>("op_41183_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_41183_cast_fp16 = slice_by_index(begin = var_41183_begin_0, end = var_41183_end_0, end_mask = var_41183_end_mask_0, x = k_367_cast_fp16)[name = tensor<string, []>("op_41183_cast_fp16")];
            tensor<int32, [4]> var_41187_begin_0 = const()[name = tensor<string, []>("op_41187_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_41187_end_0 = const()[name = tensor<string, []>("op_41187_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_41187_end_mask_0 = const()[name = tensor<string, []>("op_41187_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_41187_cast_fp16 = slice_by_index(begin = var_41187_begin_0, end = var_41187_end_0, end_mask = var_41187_end_mask_0, x = k_367_cast_fp16)[name = tensor<string, []>("op_41187_cast_fp16")];
            tensor<int32, [4]> var_41191_begin_0 = const()[name = tensor<string, []>("op_41191_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_41191_end_0 = const()[name = tensor<string, []>("op_41191_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_41191_end_mask_0 = const()[name = tensor<string, []>("op_41191_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_41191_cast_fp16 = slice_by_index(begin = var_41191_begin_0, end = var_41191_end_0, end_mask = var_41191_end_mask_0, x = k_367_cast_fp16)[name = tensor<string, []>("op_41191_cast_fp16")];
            tensor<int32, [4]> var_41195_begin_0 = const()[name = tensor<string, []>("op_41195_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_41195_end_0 = const()[name = tensor<string, []>("op_41195_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_41195_end_mask_0 = const()[name = tensor<string, []>("op_41195_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_41195_cast_fp16 = slice_by_index(begin = var_41195_begin_0, end = var_41195_end_0, end_mask = var_41195_end_mask_0, x = k_367_cast_fp16)[name = tensor<string, []>("op_41195_cast_fp16")];
            tensor<int32, [4]> var_41199_begin_0 = const()[name = tensor<string, []>("op_41199_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_41199_end_0 = const()[name = tensor<string, []>("op_41199_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_41199_end_mask_0 = const()[name = tensor<string, []>("op_41199_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_41199_cast_fp16 = slice_by_index(begin = var_41199_begin_0, end = var_41199_end_0, end_mask = var_41199_end_mask_0, x = k_367_cast_fp16)[name = tensor<string, []>("op_41199_cast_fp16")];
            tensor<int32, [4]> var_41203_begin_0 = const()[name = tensor<string, []>("op_41203_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_41203_end_0 = const()[name = tensor<string, []>("op_41203_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_41203_end_mask_0 = const()[name = tensor<string, []>("op_41203_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_41203_cast_fp16 = slice_by_index(begin = var_41203_begin_0, end = var_41203_end_0, end_mask = var_41203_end_mask_0, x = k_367_cast_fp16)[name = tensor<string, []>("op_41203_cast_fp16")];
            tensor<int32, [4]> var_41207_begin_0 = const()[name = tensor<string, []>("op_41207_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_41207_end_0 = const()[name = tensor<string, []>("op_41207_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_41207_end_mask_0 = const()[name = tensor<string, []>("op_41207_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_41207_cast_fp16 = slice_by_index(begin = var_41207_begin_0, end = var_41207_end_0, end_mask = var_41207_end_mask_0, x = k_367_cast_fp16)[name = tensor<string, []>("op_41207_cast_fp16")];
            tensor<int32, [4]> var_41211_begin_0 = const()[name = tensor<string, []>("op_41211_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_41211_end_0 = const()[name = tensor<string, []>("op_41211_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_41211_end_mask_0 = const()[name = tensor<string, []>("op_41211_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_41211_cast_fp16 = slice_by_index(begin = var_41211_begin_0, end = var_41211_end_0, end_mask = var_41211_end_mask_0, x = k_367_cast_fp16)[name = tensor<string, []>("op_41211_cast_fp16")];
            tensor<int32, [4]> var_41215_begin_0 = const()[name = tensor<string, []>("op_41215_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_41215_end_0 = const()[name = tensor<string, []>("op_41215_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_41215_end_mask_0 = const()[name = tensor<string, []>("op_41215_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_41215_cast_fp16 = slice_by_index(begin = var_41215_begin_0, end = var_41215_end_0, end_mask = var_41215_end_mask_0, x = k_367_cast_fp16)[name = tensor<string, []>("op_41215_cast_fp16")];
            tensor<int32, [4]> var_41219_begin_0 = const()[name = tensor<string, []>("op_41219_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_41219_end_0 = const()[name = tensor<string, []>("op_41219_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_41219_end_mask_0 = const()[name = tensor<string, []>("op_41219_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_41219_cast_fp16 = slice_by_index(begin = var_41219_begin_0, end = var_41219_end_0, end_mask = var_41219_end_mask_0, x = k_367_cast_fp16)[name = tensor<string, []>("op_41219_cast_fp16")];
            tensor<int32, [4]> var_41221_begin_0 = const()[name = tensor<string, []>("op_41221_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_41221_end_0 = const()[name = tensor<string, []>("op_41221_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_41221_end_mask_0 = const()[name = tensor<string, []>("op_41221_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_41221_cast_fp16 = slice_by_index(begin = var_41221_begin_0, end = var_41221_end_0, end_mask = var_41221_end_mask_0, x = v_183_cast_fp16)[name = tensor<string, []>("op_41221_cast_fp16")];
            tensor<int32, [4]> var_41225_begin_0 = const()[name = tensor<string, []>("op_41225_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_41225_end_0 = const()[name = tensor<string, []>("op_41225_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_41225_end_mask_0 = const()[name = tensor<string, []>("op_41225_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_41225_cast_fp16 = slice_by_index(begin = var_41225_begin_0, end = var_41225_end_0, end_mask = var_41225_end_mask_0, x = v_183_cast_fp16)[name = tensor<string, []>("op_41225_cast_fp16")];
            tensor<int32, [4]> var_41229_begin_0 = const()[name = tensor<string, []>("op_41229_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_41229_end_0 = const()[name = tensor<string, []>("op_41229_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_41229_end_mask_0 = const()[name = tensor<string, []>("op_41229_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_41229_cast_fp16 = slice_by_index(begin = var_41229_begin_0, end = var_41229_end_0, end_mask = var_41229_end_mask_0, x = v_183_cast_fp16)[name = tensor<string, []>("op_41229_cast_fp16")];
            tensor<int32, [4]> var_41233_begin_0 = const()[name = tensor<string, []>("op_41233_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_41233_end_0 = const()[name = tensor<string, []>("op_41233_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_41233_end_mask_0 = const()[name = tensor<string, []>("op_41233_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_41233_cast_fp16 = slice_by_index(begin = var_41233_begin_0, end = var_41233_end_0, end_mask = var_41233_end_mask_0, x = v_183_cast_fp16)[name = tensor<string, []>("op_41233_cast_fp16")];
            tensor<int32, [4]> var_41237_begin_0 = const()[name = tensor<string, []>("op_41237_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_41237_end_0 = const()[name = tensor<string, []>("op_41237_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_41237_end_mask_0 = const()[name = tensor<string, []>("op_41237_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_41237_cast_fp16 = slice_by_index(begin = var_41237_begin_0, end = var_41237_end_0, end_mask = var_41237_end_mask_0, x = v_183_cast_fp16)[name = tensor<string, []>("op_41237_cast_fp16")];
            tensor<int32, [4]> var_41241_begin_0 = const()[name = tensor<string, []>("op_41241_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_41241_end_0 = const()[name = tensor<string, []>("op_41241_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_41241_end_mask_0 = const()[name = tensor<string, []>("op_41241_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_41241_cast_fp16 = slice_by_index(begin = var_41241_begin_0, end = var_41241_end_0, end_mask = var_41241_end_mask_0, x = v_183_cast_fp16)[name = tensor<string, []>("op_41241_cast_fp16")];
            tensor<int32, [4]> var_41245_begin_0 = const()[name = tensor<string, []>("op_41245_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_41245_end_0 = const()[name = tensor<string, []>("op_41245_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_41245_end_mask_0 = const()[name = tensor<string, []>("op_41245_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_41245_cast_fp16 = slice_by_index(begin = var_41245_begin_0, end = var_41245_end_0, end_mask = var_41245_end_mask_0, x = v_183_cast_fp16)[name = tensor<string, []>("op_41245_cast_fp16")];
            tensor<int32, [4]> var_41249_begin_0 = const()[name = tensor<string, []>("op_41249_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_41249_end_0 = const()[name = tensor<string, []>("op_41249_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_41249_end_mask_0 = const()[name = tensor<string, []>("op_41249_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_41249_cast_fp16 = slice_by_index(begin = var_41249_begin_0, end = var_41249_end_0, end_mask = var_41249_end_mask_0, x = v_183_cast_fp16)[name = tensor<string, []>("op_41249_cast_fp16")];
            tensor<int32, [4]> var_41253_begin_0 = const()[name = tensor<string, []>("op_41253_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_41253_end_0 = const()[name = tensor<string, []>("op_41253_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_41253_end_mask_0 = const()[name = tensor<string, []>("op_41253_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_41253_cast_fp16 = slice_by_index(begin = var_41253_begin_0, end = var_41253_end_0, end_mask = var_41253_end_mask_0, x = v_183_cast_fp16)[name = tensor<string, []>("op_41253_cast_fp16")];
            tensor<int32, [4]> var_41257_begin_0 = const()[name = tensor<string, []>("op_41257_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_41257_end_0 = const()[name = tensor<string, []>("op_41257_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_41257_end_mask_0 = const()[name = tensor<string, []>("op_41257_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_41257_cast_fp16 = slice_by_index(begin = var_41257_begin_0, end = var_41257_end_0, end_mask = var_41257_end_mask_0, x = v_183_cast_fp16)[name = tensor<string, []>("op_41257_cast_fp16")];
            tensor<int32, [4]> var_41261_begin_0 = const()[name = tensor<string, []>("op_41261_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_41261_end_0 = const()[name = tensor<string, []>("op_41261_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_41261_end_mask_0 = const()[name = tensor<string, []>("op_41261_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_41261_cast_fp16 = slice_by_index(begin = var_41261_begin_0, end = var_41261_end_0, end_mask = var_41261_end_mask_0, x = v_183_cast_fp16)[name = tensor<string, []>("op_41261_cast_fp16")];
            tensor<int32, [4]> var_41265_begin_0 = const()[name = tensor<string, []>("op_41265_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_41265_end_0 = const()[name = tensor<string, []>("op_41265_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_41265_end_mask_0 = const()[name = tensor<string, []>("op_41265_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_41265_cast_fp16 = slice_by_index(begin = var_41265_begin_0, end = var_41265_end_0, end_mask = var_41265_end_mask_0, x = v_183_cast_fp16)[name = tensor<string, []>("op_41265_cast_fp16")];
            tensor<int32, [4]> var_41269_begin_0 = const()[name = tensor<string, []>("op_41269_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_41269_end_0 = const()[name = tensor<string, []>("op_41269_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_41269_end_mask_0 = const()[name = tensor<string, []>("op_41269_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_41269_cast_fp16 = slice_by_index(begin = var_41269_begin_0, end = var_41269_end_0, end_mask = var_41269_end_mask_0, x = v_183_cast_fp16)[name = tensor<string, []>("op_41269_cast_fp16")];
            tensor<int32, [4]> var_41273_begin_0 = const()[name = tensor<string, []>("op_41273_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_41273_end_0 = const()[name = tensor<string, []>("op_41273_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_41273_end_mask_0 = const()[name = tensor<string, []>("op_41273_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_41273_cast_fp16 = slice_by_index(begin = var_41273_begin_0, end = var_41273_end_0, end_mask = var_41273_end_mask_0, x = v_183_cast_fp16)[name = tensor<string, []>("op_41273_cast_fp16")];
            tensor<int32, [4]> var_41277_begin_0 = const()[name = tensor<string, []>("op_41277_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_41277_end_0 = const()[name = tensor<string, []>("op_41277_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_41277_end_mask_0 = const()[name = tensor<string, []>("op_41277_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_41277_cast_fp16 = slice_by_index(begin = var_41277_begin_0, end = var_41277_end_0, end_mask = var_41277_end_mask_0, x = v_183_cast_fp16)[name = tensor<string, []>("op_41277_cast_fp16")];
            tensor<int32, [4]> var_41281_begin_0 = const()[name = tensor<string, []>("op_41281_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_41281_end_0 = const()[name = tensor<string, []>("op_41281_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_41281_end_mask_0 = const()[name = tensor<string, []>("op_41281_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_41281_cast_fp16 = slice_by_index(begin = var_41281_begin_0, end = var_41281_end_0, end_mask = var_41281_end_mask_0, x = v_183_cast_fp16)[name = tensor<string, []>("op_41281_cast_fp16")];
            tensor<int32, [4]> var_41285_begin_0 = const()[name = tensor<string, []>("op_41285_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_41285_end_0 = const()[name = tensor<string, []>("op_41285_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_41285_end_mask_0 = const()[name = tensor<string, []>("op_41285_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_41285_cast_fp16 = slice_by_index(begin = var_41285_begin_0, end = var_41285_end_0, end_mask = var_41285_end_mask_0, x = v_183_cast_fp16)[name = tensor<string, []>("op_41285_cast_fp16")];
            tensor<int32, [4]> var_41289_begin_0 = const()[name = tensor<string, []>("op_41289_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_41289_end_0 = const()[name = tensor<string, []>("op_41289_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_41289_end_mask_0 = const()[name = tensor<string, []>("op_41289_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_41289_cast_fp16 = slice_by_index(begin = var_41289_begin_0, end = var_41289_end_0, end_mask = var_41289_end_mask_0, x = v_183_cast_fp16)[name = tensor<string, []>("op_41289_cast_fp16")];
            tensor<int32, [4]> var_41293_begin_0 = const()[name = tensor<string, []>("op_41293_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_41293_end_0 = const()[name = tensor<string, []>("op_41293_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_41293_end_mask_0 = const()[name = tensor<string, []>("op_41293_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_41293_cast_fp16 = slice_by_index(begin = var_41293_begin_0, end = var_41293_end_0, end_mask = var_41293_end_mask_0, x = v_183_cast_fp16)[name = tensor<string, []>("op_41293_cast_fp16")];
            tensor<int32, [4]> var_41297_begin_0 = const()[name = tensor<string, []>("op_41297_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_41297_end_0 = const()[name = tensor<string, []>("op_41297_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_41297_end_mask_0 = const()[name = tensor<string, []>("op_41297_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_41297_cast_fp16 = slice_by_index(begin = var_41297_begin_0, end = var_41297_end_0, end_mask = var_41297_end_mask_0, x = v_183_cast_fp16)[name = tensor<string, []>("op_41297_cast_fp16")];
            tensor<string, []> var_41301_equation_0 = const()[name = tensor<string, []>("op_41301_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_41301_cast_fp16 = einsum(equation = var_41301_equation_0, values = (var_41143_cast_fp16, var_41060_cast_fp16))[name = tensor<string, []>("op_41301_cast_fp16")];
            tensor<fp16, []> var_41302_to_fp16 = const()[name = tensor<string, []>("op_41302_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3481_cast_fp16 = mul(x = var_41301_cast_fp16, y = var_41302_to_fp16)[name = tensor<string, []>("aw_3481_cast_fp16")];
            tensor<string, []> var_41305_equation_0 = const()[name = tensor<string, []>("op_41305_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_41305_cast_fp16 = einsum(equation = var_41305_equation_0, values = (var_41147_cast_fp16, var_41064_cast_fp16))[name = tensor<string, []>("op_41305_cast_fp16")];
            tensor<fp16, []> var_41306_to_fp16 = const()[name = tensor<string, []>("op_41306_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3483_cast_fp16 = mul(x = var_41305_cast_fp16, y = var_41306_to_fp16)[name = tensor<string, []>("aw_3483_cast_fp16")];
            tensor<string, []> var_41309_equation_0 = const()[name = tensor<string, []>("op_41309_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_41309_cast_fp16 = einsum(equation = var_41309_equation_0, values = (var_41151_cast_fp16, var_41068_cast_fp16))[name = tensor<string, []>("op_41309_cast_fp16")];
            tensor<fp16, []> var_41310_to_fp16 = const()[name = tensor<string, []>("op_41310_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3485_cast_fp16 = mul(x = var_41309_cast_fp16, y = var_41310_to_fp16)[name = tensor<string, []>("aw_3485_cast_fp16")];
            tensor<string, []> var_41313_equation_0 = const()[name = tensor<string, []>("op_41313_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_41313_cast_fp16 = einsum(equation = var_41313_equation_0, values = (var_41155_cast_fp16, var_41072_cast_fp16))[name = tensor<string, []>("op_41313_cast_fp16")];
            tensor<fp16, []> var_41314_to_fp16 = const()[name = tensor<string, []>("op_41314_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3487_cast_fp16 = mul(x = var_41313_cast_fp16, y = var_41314_to_fp16)[name = tensor<string, []>("aw_3487_cast_fp16")];
            tensor<string, []> var_41317_equation_0 = const()[name = tensor<string, []>("op_41317_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_41317_cast_fp16 = einsum(equation = var_41317_equation_0, values = (var_41159_cast_fp16, var_41076_cast_fp16))[name = tensor<string, []>("op_41317_cast_fp16")];
            tensor<fp16, []> var_41318_to_fp16 = const()[name = tensor<string, []>("op_41318_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3489_cast_fp16 = mul(x = var_41317_cast_fp16, y = var_41318_to_fp16)[name = tensor<string, []>("aw_3489_cast_fp16")];
            tensor<string, []> var_41321_equation_0 = const()[name = tensor<string, []>("op_41321_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_41321_cast_fp16 = einsum(equation = var_41321_equation_0, values = (var_41163_cast_fp16, var_41080_cast_fp16))[name = tensor<string, []>("op_41321_cast_fp16")];
            tensor<fp16, []> var_41322_to_fp16 = const()[name = tensor<string, []>("op_41322_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3491_cast_fp16 = mul(x = var_41321_cast_fp16, y = var_41322_to_fp16)[name = tensor<string, []>("aw_3491_cast_fp16")];
            tensor<string, []> var_41325_equation_0 = const()[name = tensor<string, []>("op_41325_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_41325_cast_fp16 = einsum(equation = var_41325_equation_0, values = (var_41167_cast_fp16, var_41084_cast_fp16))[name = tensor<string, []>("op_41325_cast_fp16")];
            tensor<fp16, []> var_41326_to_fp16 = const()[name = tensor<string, []>("op_41326_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3493_cast_fp16 = mul(x = var_41325_cast_fp16, y = var_41326_to_fp16)[name = tensor<string, []>("aw_3493_cast_fp16")];
            tensor<string, []> var_41329_equation_0 = const()[name = tensor<string, []>("op_41329_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_41329_cast_fp16 = einsum(equation = var_41329_equation_0, values = (var_41171_cast_fp16, var_41088_cast_fp16))[name = tensor<string, []>("op_41329_cast_fp16")];
            tensor<fp16, []> var_41330_to_fp16 = const()[name = tensor<string, []>("op_41330_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3495_cast_fp16 = mul(x = var_41329_cast_fp16, y = var_41330_to_fp16)[name = tensor<string, []>("aw_3495_cast_fp16")];
            tensor<string, []> var_41333_equation_0 = const()[name = tensor<string, []>("op_41333_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_41333_cast_fp16 = einsum(equation = var_41333_equation_0, values = (var_41175_cast_fp16, var_41092_cast_fp16))[name = tensor<string, []>("op_41333_cast_fp16")];
            tensor<fp16, []> var_41334_to_fp16 = const()[name = tensor<string, []>("op_41334_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3497_cast_fp16 = mul(x = var_41333_cast_fp16, y = var_41334_to_fp16)[name = tensor<string, []>("aw_3497_cast_fp16")];
            tensor<string, []> var_41337_equation_0 = const()[name = tensor<string, []>("op_41337_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_41337_cast_fp16 = einsum(equation = var_41337_equation_0, values = (var_41179_cast_fp16, var_41096_cast_fp16))[name = tensor<string, []>("op_41337_cast_fp16")];
            tensor<fp16, []> var_41338_to_fp16 = const()[name = tensor<string, []>("op_41338_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3499_cast_fp16 = mul(x = var_41337_cast_fp16, y = var_41338_to_fp16)[name = tensor<string, []>("aw_3499_cast_fp16")];
            tensor<string, []> var_41341_equation_0 = const()[name = tensor<string, []>("op_41341_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_41341_cast_fp16 = einsum(equation = var_41341_equation_0, values = (var_41183_cast_fp16, var_41100_cast_fp16))[name = tensor<string, []>("op_41341_cast_fp16")];
            tensor<fp16, []> var_41342_to_fp16 = const()[name = tensor<string, []>("op_41342_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3501_cast_fp16 = mul(x = var_41341_cast_fp16, y = var_41342_to_fp16)[name = tensor<string, []>("aw_3501_cast_fp16")];
            tensor<string, []> var_41345_equation_0 = const()[name = tensor<string, []>("op_41345_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_41345_cast_fp16 = einsum(equation = var_41345_equation_0, values = (var_41187_cast_fp16, var_41104_cast_fp16))[name = tensor<string, []>("op_41345_cast_fp16")];
            tensor<fp16, []> var_41346_to_fp16 = const()[name = tensor<string, []>("op_41346_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3503_cast_fp16 = mul(x = var_41345_cast_fp16, y = var_41346_to_fp16)[name = tensor<string, []>("aw_3503_cast_fp16")];
            tensor<string, []> var_41349_equation_0 = const()[name = tensor<string, []>("op_41349_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_41349_cast_fp16 = einsum(equation = var_41349_equation_0, values = (var_41191_cast_fp16, var_41108_cast_fp16))[name = tensor<string, []>("op_41349_cast_fp16")];
            tensor<fp16, []> var_41350_to_fp16 = const()[name = tensor<string, []>("op_41350_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3505_cast_fp16 = mul(x = var_41349_cast_fp16, y = var_41350_to_fp16)[name = tensor<string, []>("aw_3505_cast_fp16")];
            tensor<string, []> var_41353_equation_0 = const()[name = tensor<string, []>("op_41353_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_41353_cast_fp16 = einsum(equation = var_41353_equation_0, values = (var_41195_cast_fp16, var_41112_cast_fp16))[name = tensor<string, []>("op_41353_cast_fp16")];
            tensor<fp16, []> var_41354_to_fp16 = const()[name = tensor<string, []>("op_41354_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3507_cast_fp16 = mul(x = var_41353_cast_fp16, y = var_41354_to_fp16)[name = tensor<string, []>("aw_3507_cast_fp16")];
            tensor<string, []> var_41357_equation_0 = const()[name = tensor<string, []>("op_41357_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_41357_cast_fp16 = einsum(equation = var_41357_equation_0, values = (var_41199_cast_fp16, var_41116_cast_fp16))[name = tensor<string, []>("op_41357_cast_fp16")];
            tensor<fp16, []> var_41358_to_fp16 = const()[name = tensor<string, []>("op_41358_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3509_cast_fp16 = mul(x = var_41357_cast_fp16, y = var_41358_to_fp16)[name = tensor<string, []>("aw_3509_cast_fp16")];
            tensor<string, []> var_41361_equation_0 = const()[name = tensor<string, []>("op_41361_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_41361_cast_fp16 = einsum(equation = var_41361_equation_0, values = (var_41203_cast_fp16, var_41120_cast_fp16))[name = tensor<string, []>("op_41361_cast_fp16")];
            tensor<fp16, []> var_41362_to_fp16 = const()[name = tensor<string, []>("op_41362_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3511_cast_fp16 = mul(x = var_41361_cast_fp16, y = var_41362_to_fp16)[name = tensor<string, []>("aw_3511_cast_fp16")];
            tensor<string, []> var_41365_equation_0 = const()[name = tensor<string, []>("op_41365_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_41365_cast_fp16 = einsum(equation = var_41365_equation_0, values = (var_41207_cast_fp16, var_41124_cast_fp16))[name = tensor<string, []>("op_41365_cast_fp16")];
            tensor<fp16, []> var_41366_to_fp16 = const()[name = tensor<string, []>("op_41366_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3513_cast_fp16 = mul(x = var_41365_cast_fp16, y = var_41366_to_fp16)[name = tensor<string, []>("aw_3513_cast_fp16")];
            tensor<string, []> var_41369_equation_0 = const()[name = tensor<string, []>("op_41369_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_41369_cast_fp16 = einsum(equation = var_41369_equation_0, values = (var_41211_cast_fp16, var_41128_cast_fp16))[name = tensor<string, []>("op_41369_cast_fp16")];
            tensor<fp16, []> var_41370_to_fp16 = const()[name = tensor<string, []>("op_41370_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3515_cast_fp16 = mul(x = var_41369_cast_fp16, y = var_41370_to_fp16)[name = tensor<string, []>("aw_3515_cast_fp16")];
            tensor<string, []> var_41373_equation_0 = const()[name = tensor<string, []>("op_41373_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_41373_cast_fp16 = einsum(equation = var_41373_equation_0, values = (var_41215_cast_fp16, var_41132_cast_fp16))[name = tensor<string, []>("op_41373_cast_fp16")];
            tensor<fp16, []> var_41374_to_fp16 = const()[name = tensor<string, []>("op_41374_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3517_cast_fp16 = mul(x = var_41373_cast_fp16, y = var_41374_to_fp16)[name = tensor<string, []>("aw_3517_cast_fp16")];
            tensor<string, []> var_41377_equation_0 = const()[name = tensor<string, []>("op_41377_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_41377_cast_fp16 = einsum(equation = var_41377_equation_0, values = (var_41219_cast_fp16, var_41136_cast_fp16))[name = tensor<string, []>("op_41377_cast_fp16")];
            tensor<fp16, []> var_41378_to_fp16 = const()[name = tensor<string, []>("op_41378_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3519_cast_fp16 = mul(x = var_41377_cast_fp16, y = var_41378_to_fp16)[name = tensor<string, []>("aw_3519_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_41380_cast_fp16 = softmax(axis = var_30355, x = aw_3481_cast_fp16)[name = tensor<string, []>("op_41380_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_41381_cast_fp16 = softmax(axis = var_30355, x = aw_3483_cast_fp16)[name = tensor<string, []>("op_41381_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_41382_cast_fp16 = softmax(axis = var_30355, x = aw_3485_cast_fp16)[name = tensor<string, []>("op_41382_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_41383_cast_fp16 = softmax(axis = var_30355, x = aw_3487_cast_fp16)[name = tensor<string, []>("op_41383_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_41384_cast_fp16 = softmax(axis = var_30355, x = aw_3489_cast_fp16)[name = tensor<string, []>("op_41384_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_41385_cast_fp16 = softmax(axis = var_30355, x = aw_3491_cast_fp16)[name = tensor<string, []>("op_41385_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_41386_cast_fp16 = softmax(axis = var_30355, x = aw_3493_cast_fp16)[name = tensor<string, []>("op_41386_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_41387_cast_fp16 = softmax(axis = var_30355, x = aw_3495_cast_fp16)[name = tensor<string, []>("op_41387_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_41388_cast_fp16 = softmax(axis = var_30355, x = aw_3497_cast_fp16)[name = tensor<string, []>("op_41388_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_41389_cast_fp16 = softmax(axis = var_30355, x = aw_3499_cast_fp16)[name = tensor<string, []>("op_41389_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_41390_cast_fp16 = softmax(axis = var_30355, x = aw_3501_cast_fp16)[name = tensor<string, []>("op_41390_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_41391_cast_fp16 = softmax(axis = var_30355, x = aw_3503_cast_fp16)[name = tensor<string, []>("op_41391_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_41392_cast_fp16 = softmax(axis = var_30355, x = aw_3505_cast_fp16)[name = tensor<string, []>("op_41392_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_41393_cast_fp16 = softmax(axis = var_30355, x = aw_3507_cast_fp16)[name = tensor<string, []>("op_41393_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_41394_cast_fp16 = softmax(axis = var_30355, x = aw_3509_cast_fp16)[name = tensor<string, []>("op_41394_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_41395_cast_fp16 = softmax(axis = var_30355, x = aw_3511_cast_fp16)[name = tensor<string, []>("op_41395_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_41396_cast_fp16 = softmax(axis = var_30355, x = aw_3513_cast_fp16)[name = tensor<string, []>("op_41396_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_41397_cast_fp16 = softmax(axis = var_30355, x = aw_3515_cast_fp16)[name = tensor<string, []>("op_41397_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_41398_cast_fp16 = softmax(axis = var_30355, x = aw_3517_cast_fp16)[name = tensor<string, []>("op_41398_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_41399_cast_fp16 = softmax(axis = var_30355, x = aw_3519_cast_fp16)[name = tensor<string, []>("op_41399_cast_fp16")];
            tensor<string, []> var_41401_equation_0 = const()[name = tensor<string, []>("op_41401_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41401_cast_fp16 = einsum(equation = var_41401_equation_0, values = (var_41221_cast_fp16, var_41380_cast_fp16))[name = tensor<string, []>("op_41401_cast_fp16")];
            tensor<string, []> var_41403_equation_0 = const()[name = tensor<string, []>("op_41403_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41403_cast_fp16 = einsum(equation = var_41403_equation_0, values = (var_41225_cast_fp16, var_41381_cast_fp16))[name = tensor<string, []>("op_41403_cast_fp16")];
            tensor<string, []> var_41405_equation_0 = const()[name = tensor<string, []>("op_41405_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41405_cast_fp16 = einsum(equation = var_41405_equation_0, values = (var_41229_cast_fp16, var_41382_cast_fp16))[name = tensor<string, []>("op_41405_cast_fp16")];
            tensor<string, []> var_41407_equation_0 = const()[name = tensor<string, []>("op_41407_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41407_cast_fp16 = einsum(equation = var_41407_equation_0, values = (var_41233_cast_fp16, var_41383_cast_fp16))[name = tensor<string, []>("op_41407_cast_fp16")];
            tensor<string, []> var_41409_equation_0 = const()[name = tensor<string, []>("op_41409_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41409_cast_fp16 = einsum(equation = var_41409_equation_0, values = (var_41237_cast_fp16, var_41384_cast_fp16))[name = tensor<string, []>("op_41409_cast_fp16")];
            tensor<string, []> var_41411_equation_0 = const()[name = tensor<string, []>("op_41411_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41411_cast_fp16 = einsum(equation = var_41411_equation_0, values = (var_41241_cast_fp16, var_41385_cast_fp16))[name = tensor<string, []>("op_41411_cast_fp16")];
            tensor<string, []> var_41413_equation_0 = const()[name = tensor<string, []>("op_41413_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41413_cast_fp16 = einsum(equation = var_41413_equation_0, values = (var_41245_cast_fp16, var_41386_cast_fp16))[name = tensor<string, []>("op_41413_cast_fp16")];
            tensor<string, []> var_41415_equation_0 = const()[name = tensor<string, []>("op_41415_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41415_cast_fp16 = einsum(equation = var_41415_equation_0, values = (var_41249_cast_fp16, var_41387_cast_fp16))[name = tensor<string, []>("op_41415_cast_fp16")];
            tensor<string, []> var_41417_equation_0 = const()[name = tensor<string, []>("op_41417_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41417_cast_fp16 = einsum(equation = var_41417_equation_0, values = (var_41253_cast_fp16, var_41388_cast_fp16))[name = tensor<string, []>("op_41417_cast_fp16")];
            tensor<string, []> var_41419_equation_0 = const()[name = tensor<string, []>("op_41419_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41419_cast_fp16 = einsum(equation = var_41419_equation_0, values = (var_41257_cast_fp16, var_41389_cast_fp16))[name = tensor<string, []>("op_41419_cast_fp16")];
            tensor<string, []> var_41421_equation_0 = const()[name = tensor<string, []>("op_41421_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41421_cast_fp16 = einsum(equation = var_41421_equation_0, values = (var_41261_cast_fp16, var_41390_cast_fp16))[name = tensor<string, []>("op_41421_cast_fp16")];
            tensor<string, []> var_41423_equation_0 = const()[name = tensor<string, []>("op_41423_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41423_cast_fp16 = einsum(equation = var_41423_equation_0, values = (var_41265_cast_fp16, var_41391_cast_fp16))[name = tensor<string, []>("op_41423_cast_fp16")];
            tensor<string, []> var_41425_equation_0 = const()[name = tensor<string, []>("op_41425_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41425_cast_fp16 = einsum(equation = var_41425_equation_0, values = (var_41269_cast_fp16, var_41392_cast_fp16))[name = tensor<string, []>("op_41425_cast_fp16")];
            tensor<string, []> var_41427_equation_0 = const()[name = tensor<string, []>("op_41427_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41427_cast_fp16 = einsum(equation = var_41427_equation_0, values = (var_41273_cast_fp16, var_41393_cast_fp16))[name = tensor<string, []>("op_41427_cast_fp16")];
            tensor<string, []> var_41429_equation_0 = const()[name = tensor<string, []>("op_41429_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41429_cast_fp16 = einsum(equation = var_41429_equation_0, values = (var_41277_cast_fp16, var_41394_cast_fp16))[name = tensor<string, []>("op_41429_cast_fp16")];
            tensor<string, []> var_41431_equation_0 = const()[name = tensor<string, []>("op_41431_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41431_cast_fp16 = einsum(equation = var_41431_equation_0, values = (var_41281_cast_fp16, var_41395_cast_fp16))[name = tensor<string, []>("op_41431_cast_fp16")];
            tensor<string, []> var_41433_equation_0 = const()[name = tensor<string, []>("op_41433_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41433_cast_fp16 = einsum(equation = var_41433_equation_0, values = (var_41285_cast_fp16, var_41396_cast_fp16))[name = tensor<string, []>("op_41433_cast_fp16")];
            tensor<string, []> var_41435_equation_0 = const()[name = tensor<string, []>("op_41435_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41435_cast_fp16 = einsum(equation = var_41435_equation_0, values = (var_41289_cast_fp16, var_41397_cast_fp16))[name = tensor<string, []>("op_41435_cast_fp16")];
            tensor<string, []> var_41437_equation_0 = const()[name = tensor<string, []>("op_41437_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41437_cast_fp16 = einsum(equation = var_41437_equation_0, values = (var_41293_cast_fp16, var_41398_cast_fp16))[name = tensor<string, []>("op_41437_cast_fp16")];
            tensor<string, []> var_41439_equation_0 = const()[name = tensor<string, []>("op_41439_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41439_cast_fp16 = einsum(equation = var_41439_equation_0, values = (var_41297_cast_fp16, var_41399_cast_fp16))[name = tensor<string, []>("op_41439_cast_fp16")];
            tensor<bool, []> input_547_interleave_0 = const()[name = tensor<string, []>("input_547_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_547_cast_fp16 = concat(axis = var_30355, interleave = input_547_interleave_0, values = (var_41401_cast_fp16, var_41403_cast_fp16, var_41405_cast_fp16, var_41407_cast_fp16, var_41409_cast_fp16, var_41411_cast_fp16, var_41413_cast_fp16, var_41415_cast_fp16, var_41417_cast_fp16, var_41419_cast_fp16, var_41421_cast_fp16, var_41423_cast_fp16, var_41425_cast_fp16, var_41427_cast_fp16, var_41429_cast_fp16, var_41431_cast_fp16, var_41433_cast_fp16, var_41435_cast_fp16, var_41437_cast_fp16, var_41439_cast_fp16))[name = tensor<string, []>("input_547_cast_fp16")];
            tensor<string, []> var_41449_pad_type_0 = const()[name = tensor<string, []>("op_41449_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_41449_strides_0 = const()[name = tensor<string, []>("op_41449_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_41449_pad_0 = const()[name = tensor<string, []>("op_41449_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_41449_dilations_0 = const()[name = tensor<string, []>("op_41449_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_41449_groups_0 = const()[name = tensor<string, []>("op_41449_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_1_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(340238080))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(341466944))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_1_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_1_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_1_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(341467136)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_41449_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_1_attn2_to_out_0_bias_to_fp16, dilations = var_41449_dilations_0, groups = var_41449_groups_0, pad = var_41449_pad_0, pad_type = var_41449_pad_type_0, strides = var_41449_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_1_attn2_to_out_0_weight_to_fp16_palettized, x = input_547_cast_fp16)[name = tensor<string, []>("op_41449_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_275_cast_fp16 = add(x = var_41449_cast_fp16, y = inputs_273_cast_fp16)[name = tensor<string, []>("inputs_275_cast_fp16")];
            tensor<int32, [1]> input_549_axes_0 = const()[name = tensor<string, []>("input_549_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> input_549_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_549_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(341469760)))];
            tensor<fp16, [1280]> input_549_beta_0_to_fp16 = const()[name = tensor<string, []>("input_549_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(341472384)))];
            tensor<fp16, []> var_41459_to_fp16 = const()[name = tensor<string, []>("op_41459_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> input_549_cast_fp16 = layer_norm(axes = input_549_axes_0, beta = input_549_beta_0_to_fp16, epsilon = var_41459_to_fp16, gamma = input_549_gamma_0_to_fp16, x = inputs_275_cast_fp16)[name = tensor<string, []>("input_549_cast_fp16")];
            tensor<string, []> var_41479_pad_type_0 = const()[name = tensor<string, []>("op_41479_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_41479_strides_0 = const()[name = tensor<string, []>("op_41479_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_41479_pad_0 = const()[name = tensor<string, []>("op_41479_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_41479_dilations_0 = const()[name = tensor<string, []>("op_41479_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_41479_groups_0 = const()[name = tensor<string, []>("op_41479_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_1_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [9830400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(341475008))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(351305472))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_1_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([10240, 1280, 1, 1])];
            tensor<fp16, [10240]> up_blocks_0_attentions_1_transformer_blocks_1_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_1_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(351305664)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_41479_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_1_ff_net_0_proj_bias_to_fp16, dilations = var_41479_dilations_0, groups = var_41479_groups_0, pad = var_41479_pad_0, pad_type = var_41479_pad_type_0, strides = var_41479_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_1_ff_net_0_proj_weight_to_fp16_palettized, x = input_549_cast_fp16)[name = tensor<string, []>("op_41479_cast_fp16")];
            tensor<int32, [2]> var_41480_split_sizes_0 = const()[name = tensor<string, []>("op_41480_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_41480_axis_0 = const()[name = tensor<string, []>("op_41480_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_41480_cast_fp16_0, tensor<fp16, [2, 5120, 1, 1024]> var_41480_cast_fp16_1 = split(axis = var_41480_axis_0, split_sizes = var_41480_split_sizes_0, x = var_41479_cast_fp16)[name = tensor<string, []>("op_41480_cast_fp16")];
            tensor<string, []> var_41482_mode_0 = const()[name = tensor<string, []>("op_41482_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_41482_cast_fp16 = gelu(mode = var_41482_mode_0, x = var_41480_cast_fp16_1)[name = tensor<string, []>("op_41482_cast_fp16")];
            tensor<fp16, [2, 5120, 1, 1024]> input_551_cast_fp16 = mul(x = var_41480_cast_fp16_0, y = var_41482_cast_fp16)[name = tensor<string, []>("input_551_cast_fp16")];
            tensor<string, []> var_41490_pad_type_0 = const()[name = tensor<string, []>("op_41490_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_41490_strides_0 = const()[name = tensor<string, []>("op_41490_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_41490_pad_0 = const()[name = tensor<string, []>("op_41490_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_41490_dilations_0 = const()[name = tensor<string, []>("op_41490_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_41490_groups_0 = const()[name = tensor<string, []>("op_41490_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_1_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [4915200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(351326208))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(356241472))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_1_ff_net_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 5120, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_1_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_1_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(356241664)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_41490_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_1_ff_net_2_bias_to_fp16, dilations = var_41490_dilations_0, groups = var_41490_groups_0, pad = var_41490_pad_0, pad_type = var_41490_pad_type_0, strides = var_41490_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_1_ff_net_2_weight_to_fp16_palettized, x = input_551_cast_fp16)[name = tensor<string, []>("op_41490_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_277_cast_fp16 = add(x = var_41490_cast_fp16, y = inputs_275_cast_fp16)[name = tensor<string, []>("inputs_277_cast_fp16")];
            tensor<int32, [1]> hidden_states_373_axes_0 = const()[name = tensor<string, []>("hidden_states_373_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_373_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_373_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(356244288)))];
            tensor<fp16, [1280]> hidden_states_373_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_373_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(356246912)))];
            tensor<fp16, []> var_41506_to_fp16 = const()[name = tensor<string, []>("op_41506_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_373_cast_fp16 = layer_norm(axes = hidden_states_373_axes_0, beta = hidden_states_373_beta_0_to_fp16, epsilon = var_41506_to_fp16, gamma = hidden_states_373_gamma_0_to_fp16, x = inputs_277_cast_fp16)[name = tensor<string, []>("hidden_states_373_cast_fp16")];
            tensor<string, []> q_185_pad_type_0 = const()[name = tensor<string, []>("q_185_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_185_strides_0 = const()[name = tensor<string, []>("q_185_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_185_pad_0 = const()[name = tensor<string, []>("q_185_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_185_dilations_0 = const()[name = tensor<string, []>("q_185_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_185_groups_0 = const()[name = tensor<string, []>("q_185_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_2_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(356249536))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(357478400))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_2_attn1_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_185_cast_fp16 = conv(dilations = q_185_dilations_0, groups = q_185_groups_0, pad = q_185_pad_0, pad_type = q_185_pad_type_0, strides = q_185_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_2_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_373_cast_fp16)[name = tensor<string, []>("q_185_cast_fp16")];
            tensor<string, []> k_369_pad_type_0 = const()[name = tensor<string, []>("k_369_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_369_strides_0 = const()[name = tensor<string, []>("k_369_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_369_pad_0 = const()[name = tensor<string, []>("k_369_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_369_dilations_0 = const()[name = tensor<string, []>("k_369_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_369_groups_0 = const()[name = tensor<string, []>("k_369_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_2_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(357478592))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(358707456))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_2_attn1_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> k_369_cast_fp16 = conv(dilations = k_369_dilations_0, groups = k_369_groups_0, pad = k_369_pad_0, pad_type = k_369_pad_type_0, strides = k_369_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_2_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_373_cast_fp16)[name = tensor<string, []>("k_369_cast_fp16")];
            tensor<string, []> v_185_pad_type_0 = const()[name = tensor<string, []>("v_185_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_185_strides_0 = const()[name = tensor<string, []>("v_185_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_185_pad_0 = const()[name = tensor<string, []>("v_185_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_185_dilations_0 = const()[name = tensor<string, []>("v_185_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_185_groups_0 = const()[name = tensor<string, []>("v_185_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_2_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(358707648))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(359936512))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_2_attn1_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> v_185_cast_fp16 = conv(dilations = v_185_dilations_0, groups = v_185_groups_0, pad = v_185_pad_0, pad_type = v_185_pad_type_0, strides = v_185_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_2_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_373_cast_fp16)[name = tensor<string, []>("v_185_cast_fp16")];
            tensor<int32, [4]> var_41539_begin_0 = const()[name = tensor<string, []>("op_41539_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_41539_end_0 = const()[name = tensor<string, []>("op_41539_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_41539_end_mask_0 = const()[name = tensor<string, []>("op_41539_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41539_cast_fp16 = slice_by_index(begin = var_41539_begin_0, end = var_41539_end_0, end_mask = var_41539_end_mask_0, x = q_185_cast_fp16)[name = tensor<string, []>("op_41539_cast_fp16")];
            tensor<int32, [4]> var_41543_begin_0 = const()[name = tensor<string, []>("op_41543_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_41543_end_0 = const()[name = tensor<string, []>("op_41543_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_41543_end_mask_0 = const()[name = tensor<string, []>("op_41543_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41543_cast_fp16 = slice_by_index(begin = var_41543_begin_0, end = var_41543_end_0, end_mask = var_41543_end_mask_0, x = q_185_cast_fp16)[name = tensor<string, []>("op_41543_cast_fp16")];
            tensor<int32, [4]> var_41547_begin_0 = const()[name = tensor<string, []>("op_41547_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_41547_end_0 = const()[name = tensor<string, []>("op_41547_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_41547_end_mask_0 = const()[name = tensor<string, []>("op_41547_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41547_cast_fp16 = slice_by_index(begin = var_41547_begin_0, end = var_41547_end_0, end_mask = var_41547_end_mask_0, x = q_185_cast_fp16)[name = tensor<string, []>("op_41547_cast_fp16")];
            tensor<int32, [4]> var_41551_begin_0 = const()[name = tensor<string, []>("op_41551_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_41551_end_0 = const()[name = tensor<string, []>("op_41551_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_41551_end_mask_0 = const()[name = tensor<string, []>("op_41551_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41551_cast_fp16 = slice_by_index(begin = var_41551_begin_0, end = var_41551_end_0, end_mask = var_41551_end_mask_0, x = q_185_cast_fp16)[name = tensor<string, []>("op_41551_cast_fp16")];
            tensor<int32, [4]> var_41555_begin_0 = const()[name = tensor<string, []>("op_41555_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_41555_end_0 = const()[name = tensor<string, []>("op_41555_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_41555_end_mask_0 = const()[name = tensor<string, []>("op_41555_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41555_cast_fp16 = slice_by_index(begin = var_41555_begin_0, end = var_41555_end_0, end_mask = var_41555_end_mask_0, x = q_185_cast_fp16)[name = tensor<string, []>("op_41555_cast_fp16")];
            tensor<int32, [4]> var_41559_begin_0 = const()[name = tensor<string, []>("op_41559_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_41559_end_0 = const()[name = tensor<string, []>("op_41559_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_41559_end_mask_0 = const()[name = tensor<string, []>("op_41559_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41559_cast_fp16 = slice_by_index(begin = var_41559_begin_0, end = var_41559_end_0, end_mask = var_41559_end_mask_0, x = q_185_cast_fp16)[name = tensor<string, []>("op_41559_cast_fp16")];
            tensor<int32, [4]> var_41563_begin_0 = const()[name = tensor<string, []>("op_41563_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_41563_end_0 = const()[name = tensor<string, []>("op_41563_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_41563_end_mask_0 = const()[name = tensor<string, []>("op_41563_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41563_cast_fp16 = slice_by_index(begin = var_41563_begin_0, end = var_41563_end_0, end_mask = var_41563_end_mask_0, x = q_185_cast_fp16)[name = tensor<string, []>("op_41563_cast_fp16")];
            tensor<int32, [4]> var_41567_begin_0 = const()[name = tensor<string, []>("op_41567_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_41567_end_0 = const()[name = tensor<string, []>("op_41567_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_41567_end_mask_0 = const()[name = tensor<string, []>("op_41567_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41567_cast_fp16 = slice_by_index(begin = var_41567_begin_0, end = var_41567_end_0, end_mask = var_41567_end_mask_0, x = q_185_cast_fp16)[name = tensor<string, []>("op_41567_cast_fp16")];
            tensor<int32, [4]> var_41571_begin_0 = const()[name = tensor<string, []>("op_41571_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_41571_end_0 = const()[name = tensor<string, []>("op_41571_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_41571_end_mask_0 = const()[name = tensor<string, []>("op_41571_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41571_cast_fp16 = slice_by_index(begin = var_41571_begin_0, end = var_41571_end_0, end_mask = var_41571_end_mask_0, x = q_185_cast_fp16)[name = tensor<string, []>("op_41571_cast_fp16")];
            tensor<int32, [4]> var_41575_begin_0 = const()[name = tensor<string, []>("op_41575_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_41575_end_0 = const()[name = tensor<string, []>("op_41575_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_41575_end_mask_0 = const()[name = tensor<string, []>("op_41575_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41575_cast_fp16 = slice_by_index(begin = var_41575_begin_0, end = var_41575_end_0, end_mask = var_41575_end_mask_0, x = q_185_cast_fp16)[name = tensor<string, []>("op_41575_cast_fp16")];
            tensor<int32, [4]> var_41579_begin_0 = const()[name = tensor<string, []>("op_41579_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_41579_end_0 = const()[name = tensor<string, []>("op_41579_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_41579_end_mask_0 = const()[name = tensor<string, []>("op_41579_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41579_cast_fp16 = slice_by_index(begin = var_41579_begin_0, end = var_41579_end_0, end_mask = var_41579_end_mask_0, x = q_185_cast_fp16)[name = tensor<string, []>("op_41579_cast_fp16")];
            tensor<int32, [4]> var_41583_begin_0 = const()[name = tensor<string, []>("op_41583_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_41583_end_0 = const()[name = tensor<string, []>("op_41583_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_41583_end_mask_0 = const()[name = tensor<string, []>("op_41583_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41583_cast_fp16 = slice_by_index(begin = var_41583_begin_0, end = var_41583_end_0, end_mask = var_41583_end_mask_0, x = q_185_cast_fp16)[name = tensor<string, []>("op_41583_cast_fp16")];
            tensor<int32, [4]> var_41587_begin_0 = const()[name = tensor<string, []>("op_41587_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_41587_end_0 = const()[name = tensor<string, []>("op_41587_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_41587_end_mask_0 = const()[name = tensor<string, []>("op_41587_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41587_cast_fp16 = slice_by_index(begin = var_41587_begin_0, end = var_41587_end_0, end_mask = var_41587_end_mask_0, x = q_185_cast_fp16)[name = tensor<string, []>("op_41587_cast_fp16")];
            tensor<int32, [4]> var_41591_begin_0 = const()[name = tensor<string, []>("op_41591_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_41591_end_0 = const()[name = tensor<string, []>("op_41591_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_41591_end_mask_0 = const()[name = tensor<string, []>("op_41591_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41591_cast_fp16 = slice_by_index(begin = var_41591_begin_0, end = var_41591_end_0, end_mask = var_41591_end_mask_0, x = q_185_cast_fp16)[name = tensor<string, []>("op_41591_cast_fp16")];
            tensor<int32, [4]> var_41595_begin_0 = const()[name = tensor<string, []>("op_41595_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_41595_end_0 = const()[name = tensor<string, []>("op_41595_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_41595_end_mask_0 = const()[name = tensor<string, []>("op_41595_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41595_cast_fp16 = slice_by_index(begin = var_41595_begin_0, end = var_41595_end_0, end_mask = var_41595_end_mask_0, x = q_185_cast_fp16)[name = tensor<string, []>("op_41595_cast_fp16")];
            tensor<int32, [4]> var_41599_begin_0 = const()[name = tensor<string, []>("op_41599_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_41599_end_0 = const()[name = tensor<string, []>("op_41599_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_41599_end_mask_0 = const()[name = tensor<string, []>("op_41599_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41599_cast_fp16 = slice_by_index(begin = var_41599_begin_0, end = var_41599_end_0, end_mask = var_41599_end_mask_0, x = q_185_cast_fp16)[name = tensor<string, []>("op_41599_cast_fp16")];
            tensor<int32, [4]> var_41603_begin_0 = const()[name = tensor<string, []>("op_41603_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_41603_end_0 = const()[name = tensor<string, []>("op_41603_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_41603_end_mask_0 = const()[name = tensor<string, []>("op_41603_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41603_cast_fp16 = slice_by_index(begin = var_41603_begin_0, end = var_41603_end_0, end_mask = var_41603_end_mask_0, x = q_185_cast_fp16)[name = tensor<string, []>("op_41603_cast_fp16")];
            tensor<int32, [4]> var_41607_begin_0 = const()[name = tensor<string, []>("op_41607_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_41607_end_0 = const()[name = tensor<string, []>("op_41607_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_41607_end_mask_0 = const()[name = tensor<string, []>("op_41607_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41607_cast_fp16 = slice_by_index(begin = var_41607_begin_0, end = var_41607_end_0, end_mask = var_41607_end_mask_0, x = q_185_cast_fp16)[name = tensor<string, []>("op_41607_cast_fp16")];
            tensor<int32, [4]> var_41611_begin_0 = const()[name = tensor<string, []>("op_41611_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_41611_end_0 = const()[name = tensor<string, []>("op_41611_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_41611_end_mask_0 = const()[name = tensor<string, []>("op_41611_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41611_cast_fp16 = slice_by_index(begin = var_41611_begin_0, end = var_41611_end_0, end_mask = var_41611_end_mask_0, x = q_185_cast_fp16)[name = tensor<string, []>("op_41611_cast_fp16")];
            tensor<int32, [4]> var_41615_begin_0 = const()[name = tensor<string, []>("op_41615_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_41615_end_0 = const()[name = tensor<string, []>("op_41615_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_41615_end_mask_0 = const()[name = tensor<string, []>("op_41615_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41615_cast_fp16 = slice_by_index(begin = var_41615_begin_0, end = var_41615_end_0, end_mask = var_41615_end_mask_0, x = q_185_cast_fp16)[name = tensor<string, []>("op_41615_cast_fp16")];
            tensor<int32, [4]> k_371_perm_0 = const()[name = tensor<string, []>("k_371_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_41622_begin_0 = const()[name = tensor<string, []>("op_41622_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_41622_end_0 = const()[name = tensor<string, []>("op_41622_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_41622_end_mask_0 = const()[name = tensor<string, []>("op_41622_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> k_371_cast_fp16 = transpose(perm = k_371_perm_0, x = k_369_cast_fp16)[name = tensor<string, []>("transpose_47")];
            tensor<fp16, [2, 1024, 1, 64]> var_41622_cast_fp16 = slice_by_index(begin = var_41622_begin_0, end = var_41622_end_0, end_mask = var_41622_end_mask_0, x = k_371_cast_fp16)[name = tensor<string, []>("op_41622_cast_fp16")];
            tensor<int32, [4]> var_41626_begin_0 = const()[name = tensor<string, []>("op_41626_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_41626_end_0 = const()[name = tensor<string, []>("op_41626_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_41626_end_mask_0 = const()[name = tensor<string, []>("op_41626_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_41626_cast_fp16 = slice_by_index(begin = var_41626_begin_0, end = var_41626_end_0, end_mask = var_41626_end_mask_0, x = k_371_cast_fp16)[name = tensor<string, []>("op_41626_cast_fp16")];
            tensor<int32, [4]> var_41630_begin_0 = const()[name = tensor<string, []>("op_41630_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_41630_end_0 = const()[name = tensor<string, []>("op_41630_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_41630_end_mask_0 = const()[name = tensor<string, []>("op_41630_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_41630_cast_fp16 = slice_by_index(begin = var_41630_begin_0, end = var_41630_end_0, end_mask = var_41630_end_mask_0, x = k_371_cast_fp16)[name = tensor<string, []>("op_41630_cast_fp16")];
            tensor<int32, [4]> var_41634_begin_0 = const()[name = tensor<string, []>("op_41634_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_41634_end_0 = const()[name = tensor<string, []>("op_41634_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_41634_end_mask_0 = const()[name = tensor<string, []>("op_41634_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_41634_cast_fp16 = slice_by_index(begin = var_41634_begin_0, end = var_41634_end_0, end_mask = var_41634_end_mask_0, x = k_371_cast_fp16)[name = tensor<string, []>("op_41634_cast_fp16")];
            tensor<int32, [4]> var_41638_begin_0 = const()[name = tensor<string, []>("op_41638_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_41638_end_0 = const()[name = tensor<string, []>("op_41638_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_41638_end_mask_0 = const()[name = tensor<string, []>("op_41638_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_41638_cast_fp16 = slice_by_index(begin = var_41638_begin_0, end = var_41638_end_0, end_mask = var_41638_end_mask_0, x = k_371_cast_fp16)[name = tensor<string, []>("op_41638_cast_fp16")];
            tensor<int32, [4]> var_41642_begin_0 = const()[name = tensor<string, []>("op_41642_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_41642_end_0 = const()[name = tensor<string, []>("op_41642_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_41642_end_mask_0 = const()[name = tensor<string, []>("op_41642_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_41642_cast_fp16 = slice_by_index(begin = var_41642_begin_0, end = var_41642_end_0, end_mask = var_41642_end_mask_0, x = k_371_cast_fp16)[name = tensor<string, []>("op_41642_cast_fp16")];
            tensor<int32, [4]> var_41646_begin_0 = const()[name = tensor<string, []>("op_41646_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_41646_end_0 = const()[name = tensor<string, []>("op_41646_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_41646_end_mask_0 = const()[name = tensor<string, []>("op_41646_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_41646_cast_fp16 = slice_by_index(begin = var_41646_begin_0, end = var_41646_end_0, end_mask = var_41646_end_mask_0, x = k_371_cast_fp16)[name = tensor<string, []>("op_41646_cast_fp16")];
            tensor<int32, [4]> var_41650_begin_0 = const()[name = tensor<string, []>("op_41650_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_41650_end_0 = const()[name = tensor<string, []>("op_41650_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_41650_end_mask_0 = const()[name = tensor<string, []>("op_41650_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_41650_cast_fp16 = slice_by_index(begin = var_41650_begin_0, end = var_41650_end_0, end_mask = var_41650_end_mask_0, x = k_371_cast_fp16)[name = tensor<string, []>("op_41650_cast_fp16")];
            tensor<int32, [4]> var_41654_begin_0 = const()[name = tensor<string, []>("op_41654_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_41654_end_0 = const()[name = tensor<string, []>("op_41654_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_41654_end_mask_0 = const()[name = tensor<string, []>("op_41654_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_41654_cast_fp16 = slice_by_index(begin = var_41654_begin_0, end = var_41654_end_0, end_mask = var_41654_end_mask_0, x = k_371_cast_fp16)[name = tensor<string, []>("op_41654_cast_fp16")];
            tensor<int32, [4]> var_41658_begin_0 = const()[name = tensor<string, []>("op_41658_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_41658_end_0 = const()[name = tensor<string, []>("op_41658_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_41658_end_mask_0 = const()[name = tensor<string, []>("op_41658_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_41658_cast_fp16 = slice_by_index(begin = var_41658_begin_0, end = var_41658_end_0, end_mask = var_41658_end_mask_0, x = k_371_cast_fp16)[name = tensor<string, []>("op_41658_cast_fp16")];
            tensor<int32, [4]> var_41662_begin_0 = const()[name = tensor<string, []>("op_41662_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_41662_end_0 = const()[name = tensor<string, []>("op_41662_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_41662_end_mask_0 = const()[name = tensor<string, []>("op_41662_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_41662_cast_fp16 = slice_by_index(begin = var_41662_begin_0, end = var_41662_end_0, end_mask = var_41662_end_mask_0, x = k_371_cast_fp16)[name = tensor<string, []>("op_41662_cast_fp16")];
            tensor<int32, [4]> var_41666_begin_0 = const()[name = tensor<string, []>("op_41666_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_41666_end_0 = const()[name = tensor<string, []>("op_41666_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_41666_end_mask_0 = const()[name = tensor<string, []>("op_41666_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_41666_cast_fp16 = slice_by_index(begin = var_41666_begin_0, end = var_41666_end_0, end_mask = var_41666_end_mask_0, x = k_371_cast_fp16)[name = tensor<string, []>("op_41666_cast_fp16")];
            tensor<int32, [4]> var_41670_begin_0 = const()[name = tensor<string, []>("op_41670_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_41670_end_0 = const()[name = tensor<string, []>("op_41670_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_41670_end_mask_0 = const()[name = tensor<string, []>("op_41670_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_41670_cast_fp16 = slice_by_index(begin = var_41670_begin_0, end = var_41670_end_0, end_mask = var_41670_end_mask_0, x = k_371_cast_fp16)[name = tensor<string, []>("op_41670_cast_fp16")];
            tensor<int32, [4]> var_41674_begin_0 = const()[name = tensor<string, []>("op_41674_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_41674_end_0 = const()[name = tensor<string, []>("op_41674_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_41674_end_mask_0 = const()[name = tensor<string, []>("op_41674_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_41674_cast_fp16 = slice_by_index(begin = var_41674_begin_0, end = var_41674_end_0, end_mask = var_41674_end_mask_0, x = k_371_cast_fp16)[name = tensor<string, []>("op_41674_cast_fp16")];
            tensor<int32, [4]> var_41678_begin_0 = const()[name = tensor<string, []>("op_41678_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_41678_end_0 = const()[name = tensor<string, []>("op_41678_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_41678_end_mask_0 = const()[name = tensor<string, []>("op_41678_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_41678_cast_fp16 = slice_by_index(begin = var_41678_begin_0, end = var_41678_end_0, end_mask = var_41678_end_mask_0, x = k_371_cast_fp16)[name = tensor<string, []>("op_41678_cast_fp16")];
            tensor<int32, [4]> var_41682_begin_0 = const()[name = tensor<string, []>("op_41682_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_41682_end_0 = const()[name = tensor<string, []>("op_41682_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_41682_end_mask_0 = const()[name = tensor<string, []>("op_41682_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_41682_cast_fp16 = slice_by_index(begin = var_41682_begin_0, end = var_41682_end_0, end_mask = var_41682_end_mask_0, x = k_371_cast_fp16)[name = tensor<string, []>("op_41682_cast_fp16")];
            tensor<int32, [4]> var_41686_begin_0 = const()[name = tensor<string, []>("op_41686_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_41686_end_0 = const()[name = tensor<string, []>("op_41686_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_41686_end_mask_0 = const()[name = tensor<string, []>("op_41686_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_41686_cast_fp16 = slice_by_index(begin = var_41686_begin_0, end = var_41686_end_0, end_mask = var_41686_end_mask_0, x = k_371_cast_fp16)[name = tensor<string, []>("op_41686_cast_fp16")];
            tensor<int32, [4]> var_41690_begin_0 = const()[name = tensor<string, []>("op_41690_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_41690_end_0 = const()[name = tensor<string, []>("op_41690_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_41690_end_mask_0 = const()[name = tensor<string, []>("op_41690_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_41690_cast_fp16 = slice_by_index(begin = var_41690_begin_0, end = var_41690_end_0, end_mask = var_41690_end_mask_0, x = k_371_cast_fp16)[name = tensor<string, []>("op_41690_cast_fp16")];
            tensor<int32, [4]> var_41694_begin_0 = const()[name = tensor<string, []>("op_41694_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_41694_end_0 = const()[name = tensor<string, []>("op_41694_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_41694_end_mask_0 = const()[name = tensor<string, []>("op_41694_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_41694_cast_fp16 = slice_by_index(begin = var_41694_begin_0, end = var_41694_end_0, end_mask = var_41694_end_mask_0, x = k_371_cast_fp16)[name = tensor<string, []>("op_41694_cast_fp16")];
            tensor<int32, [4]> var_41698_begin_0 = const()[name = tensor<string, []>("op_41698_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_41698_end_0 = const()[name = tensor<string, []>("op_41698_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_41698_end_mask_0 = const()[name = tensor<string, []>("op_41698_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_41698_cast_fp16 = slice_by_index(begin = var_41698_begin_0, end = var_41698_end_0, end_mask = var_41698_end_mask_0, x = k_371_cast_fp16)[name = tensor<string, []>("op_41698_cast_fp16")];
            tensor<int32, [4]> var_41700_begin_0 = const()[name = tensor<string, []>("op_41700_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_41700_end_0 = const()[name = tensor<string, []>("op_41700_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_41700_end_mask_0 = const()[name = tensor<string, []>("op_41700_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41700_cast_fp16 = slice_by_index(begin = var_41700_begin_0, end = var_41700_end_0, end_mask = var_41700_end_mask_0, x = v_185_cast_fp16)[name = tensor<string, []>("op_41700_cast_fp16")];
            tensor<int32, [4]> var_41704_begin_0 = const()[name = tensor<string, []>("op_41704_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_41704_end_0 = const()[name = tensor<string, []>("op_41704_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_41704_end_mask_0 = const()[name = tensor<string, []>("op_41704_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41704_cast_fp16 = slice_by_index(begin = var_41704_begin_0, end = var_41704_end_0, end_mask = var_41704_end_mask_0, x = v_185_cast_fp16)[name = tensor<string, []>("op_41704_cast_fp16")];
            tensor<int32, [4]> var_41708_begin_0 = const()[name = tensor<string, []>("op_41708_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_41708_end_0 = const()[name = tensor<string, []>("op_41708_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_41708_end_mask_0 = const()[name = tensor<string, []>("op_41708_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41708_cast_fp16 = slice_by_index(begin = var_41708_begin_0, end = var_41708_end_0, end_mask = var_41708_end_mask_0, x = v_185_cast_fp16)[name = tensor<string, []>("op_41708_cast_fp16")];
            tensor<int32, [4]> var_41712_begin_0 = const()[name = tensor<string, []>("op_41712_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_41712_end_0 = const()[name = tensor<string, []>("op_41712_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_41712_end_mask_0 = const()[name = tensor<string, []>("op_41712_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41712_cast_fp16 = slice_by_index(begin = var_41712_begin_0, end = var_41712_end_0, end_mask = var_41712_end_mask_0, x = v_185_cast_fp16)[name = tensor<string, []>("op_41712_cast_fp16")];
            tensor<int32, [4]> var_41716_begin_0 = const()[name = tensor<string, []>("op_41716_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_41716_end_0 = const()[name = tensor<string, []>("op_41716_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_41716_end_mask_0 = const()[name = tensor<string, []>("op_41716_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41716_cast_fp16 = slice_by_index(begin = var_41716_begin_0, end = var_41716_end_0, end_mask = var_41716_end_mask_0, x = v_185_cast_fp16)[name = tensor<string, []>("op_41716_cast_fp16")];
            tensor<int32, [4]> var_41720_begin_0 = const()[name = tensor<string, []>("op_41720_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_41720_end_0 = const()[name = tensor<string, []>("op_41720_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_41720_end_mask_0 = const()[name = tensor<string, []>("op_41720_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41720_cast_fp16 = slice_by_index(begin = var_41720_begin_0, end = var_41720_end_0, end_mask = var_41720_end_mask_0, x = v_185_cast_fp16)[name = tensor<string, []>("op_41720_cast_fp16")];
            tensor<int32, [4]> var_41724_begin_0 = const()[name = tensor<string, []>("op_41724_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_41724_end_0 = const()[name = tensor<string, []>("op_41724_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_41724_end_mask_0 = const()[name = tensor<string, []>("op_41724_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41724_cast_fp16 = slice_by_index(begin = var_41724_begin_0, end = var_41724_end_0, end_mask = var_41724_end_mask_0, x = v_185_cast_fp16)[name = tensor<string, []>("op_41724_cast_fp16")];
            tensor<int32, [4]> var_41728_begin_0 = const()[name = tensor<string, []>("op_41728_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_41728_end_0 = const()[name = tensor<string, []>("op_41728_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_41728_end_mask_0 = const()[name = tensor<string, []>("op_41728_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41728_cast_fp16 = slice_by_index(begin = var_41728_begin_0, end = var_41728_end_0, end_mask = var_41728_end_mask_0, x = v_185_cast_fp16)[name = tensor<string, []>("op_41728_cast_fp16")];
            tensor<int32, [4]> var_41732_begin_0 = const()[name = tensor<string, []>("op_41732_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_41732_end_0 = const()[name = tensor<string, []>("op_41732_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_41732_end_mask_0 = const()[name = tensor<string, []>("op_41732_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41732_cast_fp16 = slice_by_index(begin = var_41732_begin_0, end = var_41732_end_0, end_mask = var_41732_end_mask_0, x = v_185_cast_fp16)[name = tensor<string, []>("op_41732_cast_fp16")];
            tensor<int32, [4]> var_41736_begin_0 = const()[name = tensor<string, []>("op_41736_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_41736_end_0 = const()[name = tensor<string, []>("op_41736_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_41736_end_mask_0 = const()[name = tensor<string, []>("op_41736_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41736_cast_fp16 = slice_by_index(begin = var_41736_begin_0, end = var_41736_end_0, end_mask = var_41736_end_mask_0, x = v_185_cast_fp16)[name = tensor<string, []>("op_41736_cast_fp16")];
            tensor<int32, [4]> var_41740_begin_0 = const()[name = tensor<string, []>("op_41740_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_41740_end_0 = const()[name = tensor<string, []>("op_41740_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_41740_end_mask_0 = const()[name = tensor<string, []>("op_41740_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41740_cast_fp16 = slice_by_index(begin = var_41740_begin_0, end = var_41740_end_0, end_mask = var_41740_end_mask_0, x = v_185_cast_fp16)[name = tensor<string, []>("op_41740_cast_fp16")];
            tensor<int32, [4]> var_41744_begin_0 = const()[name = tensor<string, []>("op_41744_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_41744_end_0 = const()[name = tensor<string, []>("op_41744_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_41744_end_mask_0 = const()[name = tensor<string, []>("op_41744_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41744_cast_fp16 = slice_by_index(begin = var_41744_begin_0, end = var_41744_end_0, end_mask = var_41744_end_mask_0, x = v_185_cast_fp16)[name = tensor<string, []>("op_41744_cast_fp16")];
            tensor<int32, [4]> var_41748_begin_0 = const()[name = tensor<string, []>("op_41748_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_41748_end_0 = const()[name = tensor<string, []>("op_41748_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_41748_end_mask_0 = const()[name = tensor<string, []>("op_41748_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41748_cast_fp16 = slice_by_index(begin = var_41748_begin_0, end = var_41748_end_0, end_mask = var_41748_end_mask_0, x = v_185_cast_fp16)[name = tensor<string, []>("op_41748_cast_fp16")];
            tensor<int32, [4]> var_41752_begin_0 = const()[name = tensor<string, []>("op_41752_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_41752_end_0 = const()[name = tensor<string, []>("op_41752_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_41752_end_mask_0 = const()[name = tensor<string, []>("op_41752_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41752_cast_fp16 = slice_by_index(begin = var_41752_begin_0, end = var_41752_end_0, end_mask = var_41752_end_mask_0, x = v_185_cast_fp16)[name = tensor<string, []>("op_41752_cast_fp16")];
            tensor<int32, [4]> var_41756_begin_0 = const()[name = tensor<string, []>("op_41756_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_41756_end_0 = const()[name = tensor<string, []>("op_41756_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_41756_end_mask_0 = const()[name = tensor<string, []>("op_41756_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41756_cast_fp16 = slice_by_index(begin = var_41756_begin_0, end = var_41756_end_0, end_mask = var_41756_end_mask_0, x = v_185_cast_fp16)[name = tensor<string, []>("op_41756_cast_fp16")];
            tensor<int32, [4]> var_41760_begin_0 = const()[name = tensor<string, []>("op_41760_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_41760_end_0 = const()[name = tensor<string, []>("op_41760_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_41760_end_mask_0 = const()[name = tensor<string, []>("op_41760_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41760_cast_fp16 = slice_by_index(begin = var_41760_begin_0, end = var_41760_end_0, end_mask = var_41760_end_mask_0, x = v_185_cast_fp16)[name = tensor<string, []>("op_41760_cast_fp16")];
            tensor<int32, [4]> var_41764_begin_0 = const()[name = tensor<string, []>("op_41764_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_41764_end_0 = const()[name = tensor<string, []>("op_41764_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_41764_end_mask_0 = const()[name = tensor<string, []>("op_41764_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41764_cast_fp16 = slice_by_index(begin = var_41764_begin_0, end = var_41764_end_0, end_mask = var_41764_end_mask_0, x = v_185_cast_fp16)[name = tensor<string, []>("op_41764_cast_fp16")];
            tensor<int32, [4]> var_41768_begin_0 = const()[name = tensor<string, []>("op_41768_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_41768_end_0 = const()[name = tensor<string, []>("op_41768_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_41768_end_mask_0 = const()[name = tensor<string, []>("op_41768_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41768_cast_fp16 = slice_by_index(begin = var_41768_begin_0, end = var_41768_end_0, end_mask = var_41768_end_mask_0, x = v_185_cast_fp16)[name = tensor<string, []>("op_41768_cast_fp16")];
            tensor<int32, [4]> var_41772_begin_0 = const()[name = tensor<string, []>("op_41772_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_41772_end_0 = const()[name = tensor<string, []>("op_41772_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_41772_end_mask_0 = const()[name = tensor<string, []>("op_41772_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41772_cast_fp16 = slice_by_index(begin = var_41772_begin_0, end = var_41772_end_0, end_mask = var_41772_end_mask_0, x = v_185_cast_fp16)[name = tensor<string, []>("op_41772_cast_fp16")];
            tensor<int32, [4]> var_41776_begin_0 = const()[name = tensor<string, []>("op_41776_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_41776_end_0 = const()[name = tensor<string, []>("op_41776_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_41776_end_mask_0 = const()[name = tensor<string, []>("op_41776_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41776_cast_fp16 = slice_by_index(begin = var_41776_begin_0, end = var_41776_end_0, end_mask = var_41776_end_mask_0, x = v_185_cast_fp16)[name = tensor<string, []>("op_41776_cast_fp16")];
            tensor<string, []> var_41780_equation_0 = const()[name = tensor<string, []>("op_41780_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41780_cast_fp16 = einsum(equation = var_41780_equation_0, values = (var_41622_cast_fp16, var_41539_cast_fp16))[name = tensor<string, []>("op_41780_cast_fp16")];
            tensor<fp16, []> var_41781_to_fp16 = const()[name = tensor<string, []>("op_41781_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3521_cast_fp16 = mul(x = var_41780_cast_fp16, y = var_41781_to_fp16)[name = tensor<string, []>("aw_3521_cast_fp16")];
            tensor<string, []> var_41784_equation_0 = const()[name = tensor<string, []>("op_41784_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41784_cast_fp16 = einsum(equation = var_41784_equation_0, values = (var_41626_cast_fp16, var_41543_cast_fp16))[name = tensor<string, []>("op_41784_cast_fp16")];
            tensor<fp16, []> var_41785_to_fp16 = const()[name = tensor<string, []>("op_41785_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3523_cast_fp16 = mul(x = var_41784_cast_fp16, y = var_41785_to_fp16)[name = tensor<string, []>("aw_3523_cast_fp16")];
            tensor<string, []> var_41788_equation_0 = const()[name = tensor<string, []>("op_41788_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41788_cast_fp16 = einsum(equation = var_41788_equation_0, values = (var_41630_cast_fp16, var_41547_cast_fp16))[name = tensor<string, []>("op_41788_cast_fp16")];
            tensor<fp16, []> var_41789_to_fp16 = const()[name = tensor<string, []>("op_41789_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3525_cast_fp16 = mul(x = var_41788_cast_fp16, y = var_41789_to_fp16)[name = tensor<string, []>("aw_3525_cast_fp16")];
            tensor<string, []> var_41792_equation_0 = const()[name = tensor<string, []>("op_41792_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41792_cast_fp16 = einsum(equation = var_41792_equation_0, values = (var_41634_cast_fp16, var_41551_cast_fp16))[name = tensor<string, []>("op_41792_cast_fp16")];
            tensor<fp16, []> var_41793_to_fp16 = const()[name = tensor<string, []>("op_41793_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3527_cast_fp16 = mul(x = var_41792_cast_fp16, y = var_41793_to_fp16)[name = tensor<string, []>("aw_3527_cast_fp16")];
            tensor<string, []> var_41796_equation_0 = const()[name = tensor<string, []>("op_41796_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41796_cast_fp16 = einsum(equation = var_41796_equation_0, values = (var_41638_cast_fp16, var_41555_cast_fp16))[name = tensor<string, []>("op_41796_cast_fp16")];
            tensor<fp16, []> var_41797_to_fp16 = const()[name = tensor<string, []>("op_41797_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3529_cast_fp16 = mul(x = var_41796_cast_fp16, y = var_41797_to_fp16)[name = tensor<string, []>("aw_3529_cast_fp16")];
            tensor<string, []> var_41800_equation_0 = const()[name = tensor<string, []>("op_41800_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41800_cast_fp16 = einsum(equation = var_41800_equation_0, values = (var_41642_cast_fp16, var_41559_cast_fp16))[name = tensor<string, []>("op_41800_cast_fp16")];
            tensor<fp16, []> var_41801_to_fp16 = const()[name = tensor<string, []>("op_41801_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3531_cast_fp16 = mul(x = var_41800_cast_fp16, y = var_41801_to_fp16)[name = tensor<string, []>("aw_3531_cast_fp16")];
            tensor<string, []> var_41804_equation_0 = const()[name = tensor<string, []>("op_41804_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41804_cast_fp16 = einsum(equation = var_41804_equation_0, values = (var_41646_cast_fp16, var_41563_cast_fp16))[name = tensor<string, []>("op_41804_cast_fp16")];
            tensor<fp16, []> var_41805_to_fp16 = const()[name = tensor<string, []>("op_41805_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3533_cast_fp16 = mul(x = var_41804_cast_fp16, y = var_41805_to_fp16)[name = tensor<string, []>("aw_3533_cast_fp16")];
            tensor<string, []> var_41808_equation_0 = const()[name = tensor<string, []>("op_41808_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41808_cast_fp16 = einsum(equation = var_41808_equation_0, values = (var_41650_cast_fp16, var_41567_cast_fp16))[name = tensor<string, []>("op_41808_cast_fp16")];
            tensor<fp16, []> var_41809_to_fp16 = const()[name = tensor<string, []>("op_41809_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3535_cast_fp16 = mul(x = var_41808_cast_fp16, y = var_41809_to_fp16)[name = tensor<string, []>("aw_3535_cast_fp16")];
            tensor<string, []> var_41812_equation_0 = const()[name = tensor<string, []>("op_41812_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41812_cast_fp16 = einsum(equation = var_41812_equation_0, values = (var_41654_cast_fp16, var_41571_cast_fp16))[name = tensor<string, []>("op_41812_cast_fp16")];
            tensor<fp16, []> var_41813_to_fp16 = const()[name = tensor<string, []>("op_41813_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3537_cast_fp16 = mul(x = var_41812_cast_fp16, y = var_41813_to_fp16)[name = tensor<string, []>("aw_3537_cast_fp16")];
            tensor<string, []> var_41816_equation_0 = const()[name = tensor<string, []>("op_41816_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41816_cast_fp16 = einsum(equation = var_41816_equation_0, values = (var_41658_cast_fp16, var_41575_cast_fp16))[name = tensor<string, []>("op_41816_cast_fp16")];
            tensor<fp16, []> var_41817_to_fp16 = const()[name = tensor<string, []>("op_41817_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3539_cast_fp16 = mul(x = var_41816_cast_fp16, y = var_41817_to_fp16)[name = tensor<string, []>("aw_3539_cast_fp16")];
            tensor<string, []> var_41820_equation_0 = const()[name = tensor<string, []>("op_41820_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41820_cast_fp16 = einsum(equation = var_41820_equation_0, values = (var_41662_cast_fp16, var_41579_cast_fp16))[name = tensor<string, []>("op_41820_cast_fp16")];
            tensor<fp16, []> var_41821_to_fp16 = const()[name = tensor<string, []>("op_41821_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3541_cast_fp16 = mul(x = var_41820_cast_fp16, y = var_41821_to_fp16)[name = tensor<string, []>("aw_3541_cast_fp16")];
            tensor<string, []> var_41824_equation_0 = const()[name = tensor<string, []>("op_41824_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41824_cast_fp16 = einsum(equation = var_41824_equation_0, values = (var_41666_cast_fp16, var_41583_cast_fp16))[name = tensor<string, []>("op_41824_cast_fp16")];
            tensor<fp16, []> var_41825_to_fp16 = const()[name = tensor<string, []>("op_41825_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3543_cast_fp16 = mul(x = var_41824_cast_fp16, y = var_41825_to_fp16)[name = tensor<string, []>("aw_3543_cast_fp16")];
            tensor<string, []> var_41828_equation_0 = const()[name = tensor<string, []>("op_41828_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41828_cast_fp16 = einsum(equation = var_41828_equation_0, values = (var_41670_cast_fp16, var_41587_cast_fp16))[name = tensor<string, []>("op_41828_cast_fp16")];
            tensor<fp16, []> var_41829_to_fp16 = const()[name = tensor<string, []>("op_41829_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3545_cast_fp16 = mul(x = var_41828_cast_fp16, y = var_41829_to_fp16)[name = tensor<string, []>("aw_3545_cast_fp16")];
            tensor<string, []> var_41832_equation_0 = const()[name = tensor<string, []>("op_41832_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41832_cast_fp16 = einsum(equation = var_41832_equation_0, values = (var_41674_cast_fp16, var_41591_cast_fp16))[name = tensor<string, []>("op_41832_cast_fp16")];
            tensor<fp16, []> var_41833_to_fp16 = const()[name = tensor<string, []>("op_41833_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3547_cast_fp16 = mul(x = var_41832_cast_fp16, y = var_41833_to_fp16)[name = tensor<string, []>("aw_3547_cast_fp16")];
            tensor<string, []> var_41836_equation_0 = const()[name = tensor<string, []>("op_41836_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41836_cast_fp16 = einsum(equation = var_41836_equation_0, values = (var_41678_cast_fp16, var_41595_cast_fp16))[name = tensor<string, []>("op_41836_cast_fp16")];
            tensor<fp16, []> var_41837_to_fp16 = const()[name = tensor<string, []>("op_41837_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3549_cast_fp16 = mul(x = var_41836_cast_fp16, y = var_41837_to_fp16)[name = tensor<string, []>("aw_3549_cast_fp16")];
            tensor<string, []> var_41840_equation_0 = const()[name = tensor<string, []>("op_41840_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41840_cast_fp16 = einsum(equation = var_41840_equation_0, values = (var_41682_cast_fp16, var_41599_cast_fp16))[name = tensor<string, []>("op_41840_cast_fp16")];
            tensor<fp16, []> var_41841_to_fp16 = const()[name = tensor<string, []>("op_41841_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3551_cast_fp16 = mul(x = var_41840_cast_fp16, y = var_41841_to_fp16)[name = tensor<string, []>("aw_3551_cast_fp16")];
            tensor<string, []> var_41844_equation_0 = const()[name = tensor<string, []>("op_41844_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41844_cast_fp16 = einsum(equation = var_41844_equation_0, values = (var_41686_cast_fp16, var_41603_cast_fp16))[name = tensor<string, []>("op_41844_cast_fp16")];
            tensor<fp16, []> var_41845_to_fp16 = const()[name = tensor<string, []>("op_41845_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3553_cast_fp16 = mul(x = var_41844_cast_fp16, y = var_41845_to_fp16)[name = tensor<string, []>("aw_3553_cast_fp16")];
            tensor<string, []> var_41848_equation_0 = const()[name = tensor<string, []>("op_41848_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41848_cast_fp16 = einsum(equation = var_41848_equation_0, values = (var_41690_cast_fp16, var_41607_cast_fp16))[name = tensor<string, []>("op_41848_cast_fp16")];
            tensor<fp16, []> var_41849_to_fp16 = const()[name = tensor<string, []>("op_41849_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3555_cast_fp16 = mul(x = var_41848_cast_fp16, y = var_41849_to_fp16)[name = tensor<string, []>("aw_3555_cast_fp16")];
            tensor<string, []> var_41852_equation_0 = const()[name = tensor<string, []>("op_41852_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41852_cast_fp16 = einsum(equation = var_41852_equation_0, values = (var_41694_cast_fp16, var_41611_cast_fp16))[name = tensor<string, []>("op_41852_cast_fp16")];
            tensor<fp16, []> var_41853_to_fp16 = const()[name = tensor<string, []>("op_41853_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3557_cast_fp16 = mul(x = var_41852_cast_fp16, y = var_41853_to_fp16)[name = tensor<string, []>("aw_3557_cast_fp16")];
            tensor<string, []> var_41856_equation_0 = const()[name = tensor<string, []>("op_41856_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41856_cast_fp16 = einsum(equation = var_41856_equation_0, values = (var_41698_cast_fp16, var_41615_cast_fp16))[name = tensor<string, []>("op_41856_cast_fp16")];
            tensor<fp16, []> var_41857_to_fp16 = const()[name = tensor<string, []>("op_41857_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3559_cast_fp16 = mul(x = var_41856_cast_fp16, y = var_41857_to_fp16)[name = tensor<string, []>("aw_3559_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41859_cast_fp16 = softmax(axis = var_30355, x = aw_3521_cast_fp16)[name = tensor<string, []>("op_41859_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41860_cast_fp16 = softmax(axis = var_30355, x = aw_3523_cast_fp16)[name = tensor<string, []>("op_41860_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41861_cast_fp16 = softmax(axis = var_30355, x = aw_3525_cast_fp16)[name = tensor<string, []>("op_41861_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41862_cast_fp16 = softmax(axis = var_30355, x = aw_3527_cast_fp16)[name = tensor<string, []>("op_41862_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41863_cast_fp16 = softmax(axis = var_30355, x = aw_3529_cast_fp16)[name = tensor<string, []>("op_41863_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41864_cast_fp16 = softmax(axis = var_30355, x = aw_3531_cast_fp16)[name = tensor<string, []>("op_41864_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41865_cast_fp16 = softmax(axis = var_30355, x = aw_3533_cast_fp16)[name = tensor<string, []>("op_41865_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41866_cast_fp16 = softmax(axis = var_30355, x = aw_3535_cast_fp16)[name = tensor<string, []>("op_41866_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41867_cast_fp16 = softmax(axis = var_30355, x = aw_3537_cast_fp16)[name = tensor<string, []>("op_41867_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41868_cast_fp16 = softmax(axis = var_30355, x = aw_3539_cast_fp16)[name = tensor<string, []>("op_41868_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41869_cast_fp16 = softmax(axis = var_30355, x = aw_3541_cast_fp16)[name = tensor<string, []>("op_41869_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41870_cast_fp16 = softmax(axis = var_30355, x = aw_3543_cast_fp16)[name = tensor<string, []>("op_41870_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41871_cast_fp16 = softmax(axis = var_30355, x = aw_3545_cast_fp16)[name = tensor<string, []>("op_41871_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41872_cast_fp16 = softmax(axis = var_30355, x = aw_3547_cast_fp16)[name = tensor<string, []>("op_41872_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41873_cast_fp16 = softmax(axis = var_30355, x = aw_3549_cast_fp16)[name = tensor<string, []>("op_41873_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41874_cast_fp16 = softmax(axis = var_30355, x = aw_3551_cast_fp16)[name = tensor<string, []>("op_41874_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41875_cast_fp16 = softmax(axis = var_30355, x = aw_3553_cast_fp16)[name = tensor<string, []>("op_41875_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41876_cast_fp16 = softmax(axis = var_30355, x = aw_3555_cast_fp16)[name = tensor<string, []>("op_41876_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41877_cast_fp16 = softmax(axis = var_30355, x = aw_3557_cast_fp16)[name = tensor<string, []>("op_41877_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41878_cast_fp16 = softmax(axis = var_30355, x = aw_3559_cast_fp16)[name = tensor<string, []>("op_41878_cast_fp16")];
            tensor<string, []> var_41880_equation_0 = const()[name = tensor<string, []>("op_41880_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41880_cast_fp16 = einsum(equation = var_41880_equation_0, values = (var_41700_cast_fp16, var_41859_cast_fp16))[name = tensor<string, []>("op_41880_cast_fp16")];
            tensor<string, []> var_41882_equation_0 = const()[name = tensor<string, []>("op_41882_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41882_cast_fp16 = einsum(equation = var_41882_equation_0, values = (var_41704_cast_fp16, var_41860_cast_fp16))[name = tensor<string, []>("op_41882_cast_fp16")];
            tensor<string, []> var_41884_equation_0 = const()[name = tensor<string, []>("op_41884_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41884_cast_fp16 = einsum(equation = var_41884_equation_0, values = (var_41708_cast_fp16, var_41861_cast_fp16))[name = tensor<string, []>("op_41884_cast_fp16")];
            tensor<string, []> var_41886_equation_0 = const()[name = tensor<string, []>("op_41886_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41886_cast_fp16 = einsum(equation = var_41886_equation_0, values = (var_41712_cast_fp16, var_41862_cast_fp16))[name = tensor<string, []>("op_41886_cast_fp16")];
            tensor<string, []> var_41888_equation_0 = const()[name = tensor<string, []>("op_41888_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41888_cast_fp16 = einsum(equation = var_41888_equation_0, values = (var_41716_cast_fp16, var_41863_cast_fp16))[name = tensor<string, []>("op_41888_cast_fp16")];
            tensor<string, []> var_41890_equation_0 = const()[name = tensor<string, []>("op_41890_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41890_cast_fp16 = einsum(equation = var_41890_equation_0, values = (var_41720_cast_fp16, var_41864_cast_fp16))[name = tensor<string, []>("op_41890_cast_fp16")];
            tensor<string, []> var_41892_equation_0 = const()[name = tensor<string, []>("op_41892_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41892_cast_fp16 = einsum(equation = var_41892_equation_0, values = (var_41724_cast_fp16, var_41865_cast_fp16))[name = tensor<string, []>("op_41892_cast_fp16")];
            tensor<string, []> var_41894_equation_0 = const()[name = tensor<string, []>("op_41894_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41894_cast_fp16 = einsum(equation = var_41894_equation_0, values = (var_41728_cast_fp16, var_41866_cast_fp16))[name = tensor<string, []>("op_41894_cast_fp16")];
            tensor<string, []> var_41896_equation_0 = const()[name = tensor<string, []>("op_41896_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41896_cast_fp16 = einsum(equation = var_41896_equation_0, values = (var_41732_cast_fp16, var_41867_cast_fp16))[name = tensor<string, []>("op_41896_cast_fp16")];
            tensor<string, []> var_41898_equation_0 = const()[name = tensor<string, []>("op_41898_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41898_cast_fp16 = einsum(equation = var_41898_equation_0, values = (var_41736_cast_fp16, var_41868_cast_fp16))[name = tensor<string, []>("op_41898_cast_fp16")];
            tensor<string, []> var_41900_equation_0 = const()[name = tensor<string, []>("op_41900_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41900_cast_fp16 = einsum(equation = var_41900_equation_0, values = (var_41740_cast_fp16, var_41869_cast_fp16))[name = tensor<string, []>("op_41900_cast_fp16")];
            tensor<string, []> var_41902_equation_0 = const()[name = tensor<string, []>("op_41902_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41902_cast_fp16 = einsum(equation = var_41902_equation_0, values = (var_41744_cast_fp16, var_41870_cast_fp16))[name = tensor<string, []>("op_41902_cast_fp16")];
            tensor<string, []> var_41904_equation_0 = const()[name = tensor<string, []>("op_41904_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41904_cast_fp16 = einsum(equation = var_41904_equation_0, values = (var_41748_cast_fp16, var_41871_cast_fp16))[name = tensor<string, []>("op_41904_cast_fp16")];
            tensor<string, []> var_41906_equation_0 = const()[name = tensor<string, []>("op_41906_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41906_cast_fp16 = einsum(equation = var_41906_equation_0, values = (var_41752_cast_fp16, var_41872_cast_fp16))[name = tensor<string, []>("op_41906_cast_fp16")];
            tensor<string, []> var_41908_equation_0 = const()[name = tensor<string, []>("op_41908_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41908_cast_fp16 = einsum(equation = var_41908_equation_0, values = (var_41756_cast_fp16, var_41873_cast_fp16))[name = tensor<string, []>("op_41908_cast_fp16")];
            tensor<string, []> var_41910_equation_0 = const()[name = tensor<string, []>("op_41910_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41910_cast_fp16 = einsum(equation = var_41910_equation_0, values = (var_41760_cast_fp16, var_41874_cast_fp16))[name = tensor<string, []>("op_41910_cast_fp16")];
            tensor<string, []> var_41912_equation_0 = const()[name = tensor<string, []>("op_41912_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41912_cast_fp16 = einsum(equation = var_41912_equation_0, values = (var_41764_cast_fp16, var_41875_cast_fp16))[name = tensor<string, []>("op_41912_cast_fp16")];
            tensor<string, []> var_41914_equation_0 = const()[name = tensor<string, []>("op_41914_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41914_cast_fp16 = einsum(equation = var_41914_equation_0, values = (var_41768_cast_fp16, var_41876_cast_fp16))[name = tensor<string, []>("op_41914_cast_fp16")];
            tensor<string, []> var_41916_equation_0 = const()[name = tensor<string, []>("op_41916_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41916_cast_fp16 = einsum(equation = var_41916_equation_0, values = (var_41772_cast_fp16, var_41877_cast_fp16))[name = tensor<string, []>("op_41916_cast_fp16")];
            tensor<string, []> var_41918_equation_0 = const()[name = tensor<string, []>("op_41918_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41918_cast_fp16 = einsum(equation = var_41918_equation_0, values = (var_41776_cast_fp16, var_41878_cast_fp16))[name = tensor<string, []>("op_41918_cast_fp16")];
            tensor<bool, []> input_553_interleave_0 = const()[name = tensor<string, []>("input_553_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_553_cast_fp16 = concat(axis = var_30355, interleave = input_553_interleave_0, values = (var_41880_cast_fp16, var_41882_cast_fp16, var_41884_cast_fp16, var_41886_cast_fp16, var_41888_cast_fp16, var_41890_cast_fp16, var_41892_cast_fp16, var_41894_cast_fp16, var_41896_cast_fp16, var_41898_cast_fp16, var_41900_cast_fp16, var_41902_cast_fp16, var_41904_cast_fp16, var_41906_cast_fp16, var_41908_cast_fp16, var_41910_cast_fp16, var_41912_cast_fp16, var_41914_cast_fp16, var_41916_cast_fp16, var_41918_cast_fp16))[name = tensor<string, []>("input_553_cast_fp16")];
            tensor<string, []> var_41928_pad_type_0 = const()[name = tensor<string, []>("op_41928_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_41928_strides_0 = const()[name = tensor<string, []>("op_41928_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_41928_pad_0 = const()[name = tensor<string, []>("op_41928_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_41928_dilations_0 = const()[name = tensor<string, []>("op_41928_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_41928_groups_0 = const()[name = tensor<string, []>("op_41928_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_2_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(359936704))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(361165568))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_2_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_2_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_2_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(361165760)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_41928_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_2_attn1_to_out_0_bias_to_fp16, dilations = var_41928_dilations_0, groups = var_41928_groups_0, pad = var_41928_pad_0, pad_type = var_41928_pad_type_0, strides = var_41928_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_2_attn1_to_out_0_weight_to_fp16_palettized, x = input_553_cast_fp16)[name = tensor<string, []>("op_41928_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_279_cast_fp16 = add(x = var_41928_cast_fp16, y = inputs_277_cast_fp16)[name = tensor<string, []>("inputs_279_cast_fp16")];
            tensor<int32, [1]> hidden_states_375_axes_0 = const()[name = tensor<string, []>("hidden_states_375_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_375_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_375_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(361168384)))];
            tensor<fp16, [1280]> hidden_states_375_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_375_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(361171008)))];
            tensor<fp16, []> var_41938_to_fp16 = const()[name = tensor<string, []>("op_41938_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_375_cast_fp16 = layer_norm(axes = hidden_states_375_axes_0, beta = hidden_states_375_beta_0_to_fp16, epsilon = var_41938_to_fp16, gamma = hidden_states_375_gamma_0_to_fp16, x = inputs_279_cast_fp16)[name = tensor<string, []>("hidden_states_375_cast_fp16")];
            tensor<string, []> q_187_pad_type_0 = const()[name = tensor<string, []>("q_187_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_187_strides_0 = const()[name = tensor<string, []>("q_187_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_187_pad_0 = const()[name = tensor<string, []>("q_187_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_187_dilations_0 = const()[name = tensor<string, []>("q_187_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_187_groups_0 = const()[name = tensor<string, []>("q_187_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_2_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(361173632))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(362402496))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_2_attn2_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_187_cast_fp16 = conv(dilations = q_187_dilations_0, groups = q_187_groups_0, pad = q_187_pad_0, pad_type = q_187_pad_type_0, strides = q_187_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_2_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_375_cast_fp16)[name = tensor<string, []>("q_187_cast_fp16")];
            tensor<string, []> k_373_pad_type_0 = const()[name = tensor<string, []>("k_373_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_373_strides_0 = const()[name = tensor<string, []>("k_373_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_373_pad_0 = const()[name = tensor<string, []>("k_373_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_373_dilations_0 = const()[name = tensor<string, []>("k_373_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_373_groups_0 = const()[name = tensor<string, []>("k_373_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_2_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(362402688))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(364368832))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_2_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> k_373_cast_fp16 = conv(dilations = k_373_dilations_0, groups = k_373_groups_0, pad = k_373_pad_0, pad_type = k_373_pad_type_0, strides = k_373_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_2_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("k_373_cast_fp16")];
            tensor<string, []> v_187_pad_type_0 = const()[name = tensor<string, []>("v_187_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_187_strides_0 = const()[name = tensor<string, []>("v_187_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_187_pad_0 = const()[name = tensor<string, []>("v_187_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_187_dilations_0 = const()[name = tensor<string, []>("v_187_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_187_groups_0 = const()[name = tensor<string, []>("v_187_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_2_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(364369024))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(366335168))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_2_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> v_187_cast_fp16 = conv(dilations = v_187_dilations_0, groups = v_187_groups_0, pad = v_187_pad_0, pad_type = v_187_pad_type_0, strides = v_187_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_2_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("v_187_cast_fp16")];
            tensor<int32, [4]> var_41971_begin_0 = const()[name = tensor<string, []>("op_41971_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_41971_end_0 = const()[name = tensor<string, []>("op_41971_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_41971_end_mask_0 = const()[name = tensor<string, []>("op_41971_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41971_cast_fp16 = slice_by_index(begin = var_41971_begin_0, end = var_41971_end_0, end_mask = var_41971_end_mask_0, x = q_187_cast_fp16)[name = tensor<string, []>("op_41971_cast_fp16")];
            tensor<int32, [4]> var_41975_begin_0 = const()[name = tensor<string, []>("op_41975_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_41975_end_0 = const()[name = tensor<string, []>("op_41975_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_41975_end_mask_0 = const()[name = tensor<string, []>("op_41975_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41975_cast_fp16 = slice_by_index(begin = var_41975_begin_0, end = var_41975_end_0, end_mask = var_41975_end_mask_0, x = q_187_cast_fp16)[name = tensor<string, []>("op_41975_cast_fp16")];
            tensor<int32, [4]> var_41979_begin_0 = const()[name = tensor<string, []>("op_41979_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_41979_end_0 = const()[name = tensor<string, []>("op_41979_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_41979_end_mask_0 = const()[name = tensor<string, []>("op_41979_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41979_cast_fp16 = slice_by_index(begin = var_41979_begin_0, end = var_41979_end_0, end_mask = var_41979_end_mask_0, x = q_187_cast_fp16)[name = tensor<string, []>("op_41979_cast_fp16")];
            tensor<int32, [4]> var_41983_begin_0 = const()[name = tensor<string, []>("op_41983_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_41983_end_0 = const()[name = tensor<string, []>("op_41983_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_41983_end_mask_0 = const()[name = tensor<string, []>("op_41983_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41983_cast_fp16 = slice_by_index(begin = var_41983_begin_0, end = var_41983_end_0, end_mask = var_41983_end_mask_0, x = q_187_cast_fp16)[name = tensor<string, []>("op_41983_cast_fp16")];
            tensor<int32, [4]> var_41987_begin_0 = const()[name = tensor<string, []>("op_41987_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_41987_end_0 = const()[name = tensor<string, []>("op_41987_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_41987_end_mask_0 = const()[name = tensor<string, []>("op_41987_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41987_cast_fp16 = slice_by_index(begin = var_41987_begin_0, end = var_41987_end_0, end_mask = var_41987_end_mask_0, x = q_187_cast_fp16)[name = tensor<string, []>("op_41987_cast_fp16")];
            tensor<int32, [4]> var_41991_begin_0 = const()[name = tensor<string, []>("op_41991_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_41991_end_0 = const()[name = tensor<string, []>("op_41991_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_41991_end_mask_0 = const()[name = tensor<string, []>("op_41991_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41991_cast_fp16 = slice_by_index(begin = var_41991_begin_0, end = var_41991_end_0, end_mask = var_41991_end_mask_0, x = q_187_cast_fp16)[name = tensor<string, []>("op_41991_cast_fp16")];
            tensor<int32, [4]> var_41995_begin_0 = const()[name = tensor<string, []>("op_41995_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_41995_end_0 = const()[name = tensor<string, []>("op_41995_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_41995_end_mask_0 = const()[name = tensor<string, []>("op_41995_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41995_cast_fp16 = slice_by_index(begin = var_41995_begin_0, end = var_41995_end_0, end_mask = var_41995_end_mask_0, x = q_187_cast_fp16)[name = tensor<string, []>("op_41995_cast_fp16")];
            tensor<int32, [4]> var_41999_begin_0 = const()[name = tensor<string, []>("op_41999_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_41999_end_0 = const()[name = tensor<string, []>("op_41999_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_41999_end_mask_0 = const()[name = tensor<string, []>("op_41999_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41999_cast_fp16 = slice_by_index(begin = var_41999_begin_0, end = var_41999_end_0, end_mask = var_41999_end_mask_0, x = q_187_cast_fp16)[name = tensor<string, []>("op_41999_cast_fp16")];
            tensor<int32, [4]> var_42003_begin_0 = const()[name = tensor<string, []>("op_42003_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_42003_end_0 = const()[name = tensor<string, []>("op_42003_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_42003_end_mask_0 = const()[name = tensor<string, []>("op_42003_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42003_cast_fp16 = slice_by_index(begin = var_42003_begin_0, end = var_42003_end_0, end_mask = var_42003_end_mask_0, x = q_187_cast_fp16)[name = tensor<string, []>("op_42003_cast_fp16")];
            tensor<int32, [4]> var_42007_begin_0 = const()[name = tensor<string, []>("op_42007_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_42007_end_0 = const()[name = tensor<string, []>("op_42007_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_42007_end_mask_0 = const()[name = tensor<string, []>("op_42007_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42007_cast_fp16 = slice_by_index(begin = var_42007_begin_0, end = var_42007_end_0, end_mask = var_42007_end_mask_0, x = q_187_cast_fp16)[name = tensor<string, []>("op_42007_cast_fp16")];
            tensor<int32, [4]> var_42011_begin_0 = const()[name = tensor<string, []>("op_42011_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_42011_end_0 = const()[name = tensor<string, []>("op_42011_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_42011_end_mask_0 = const()[name = tensor<string, []>("op_42011_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42011_cast_fp16 = slice_by_index(begin = var_42011_begin_0, end = var_42011_end_0, end_mask = var_42011_end_mask_0, x = q_187_cast_fp16)[name = tensor<string, []>("op_42011_cast_fp16")];
            tensor<int32, [4]> var_42015_begin_0 = const()[name = tensor<string, []>("op_42015_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_42015_end_0 = const()[name = tensor<string, []>("op_42015_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_42015_end_mask_0 = const()[name = tensor<string, []>("op_42015_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42015_cast_fp16 = slice_by_index(begin = var_42015_begin_0, end = var_42015_end_0, end_mask = var_42015_end_mask_0, x = q_187_cast_fp16)[name = tensor<string, []>("op_42015_cast_fp16")];
            tensor<int32, [4]> var_42019_begin_0 = const()[name = tensor<string, []>("op_42019_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_42019_end_0 = const()[name = tensor<string, []>("op_42019_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_42019_end_mask_0 = const()[name = tensor<string, []>("op_42019_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42019_cast_fp16 = slice_by_index(begin = var_42019_begin_0, end = var_42019_end_0, end_mask = var_42019_end_mask_0, x = q_187_cast_fp16)[name = tensor<string, []>("op_42019_cast_fp16")];
            tensor<int32, [4]> var_42023_begin_0 = const()[name = tensor<string, []>("op_42023_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_42023_end_0 = const()[name = tensor<string, []>("op_42023_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_42023_end_mask_0 = const()[name = tensor<string, []>("op_42023_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42023_cast_fp16 = slice_by_index(begin = var_42023_begin_0, end = var_42023_end_0, end_mask = var_42023_end_mask_0, x = q_187_cast_fp16)[name = tensor<string, []>("op_42023_cast_fp16")];
            tensor<int32, [4]> var_42027_begin_0 = const()[name = tensor<string, []>("op_42027_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_42027_end_0 = const()[name = tensor<string, []>("op_42027_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_42027_end_mask_0 = const()[name = tensor<string, []>("op_42027_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42027_cast_fp16 = slice_by_index(begin = var_42027_begin_0, end = var_42027_end_0, end_mask = var_42027_end_mask_0, x = q_187_cast_fp16)[name = tensor<string, []>("op_42027_cast_fp16")];
            tensor<int32, [4]> var_42031_begin_0 = const()[name = tensor<string, []>("op_42031_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_42031_end_0 = const()[name = tensor<string, []>("op_42031_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_42031_end_mask_0 = const()[name = tensor<string, []>("op_42031_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42031_cast_fp16 = slice_by_index(begin = var_42031_begin_0, end = var_42031_end_0, end_mask = var_42031_end_mask_0, x = q_187_cast_fp16)[name = tensor<string, []>("op_42031_cast_fp16")];
            tensor<int32, [4]> var_42035_begin_0 = const()[name = tensor<string, []>("op_42035_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_42035_end_0 = const()[name = tensor<string, []>("op_42035_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_42035_end_mask_0 = const()[name = tensor<string, []>("op_42035_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42035_cast_fp16 = slice_by_index(begin = var_42035_begin_0, end = var_42035_end_0, end_mask = var_42035_end_mask_0, x = q_187_cast_fp16)[name = tensor<string, []>("op_42035_cast_fp16")];
            tensor<int32, [4]> var_42039_begin_0 = const()[name = tensor<string, []>("op_42039_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_42039_end_0 = const()[name = tensor<string, []>("op_42039_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_42039_end_mask_0 = const()[name = tensor<string, []>("op_42039_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42039_cast_fp16 = slice_by_index(begin = var_42039_begin_0, end = var_42039_end_0, end_mask = var_42039_end_mask_0, x = q_187_cast_fp16)[name = tensor<string, []>("op_42039_cast_fp16")];
            tensor<int32, [4]> var_42043_begin_0 = const()[name = tensor<string, []>("op_42043_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_42043_end_0 = const()[name = tensor<string, []>("op_42043_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_42043_end_mask_0 = const()[name = tensor<string, []>("op_42043_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42043_cast_fp16 = slice_by_index(begin = var_42043_begin_0, end = var_42043_end_0, end_mask = var_42043_end_mask_0, x = q_187_cast_fp16)[name = tensor<string, []>("op_42043_cast_fp16")];
            tensor<int32, [4]> var_42047_begin_0 = const()[name = tensor<string, []>("op_42047_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_42047_end_0 = const()[name = tensor<string, []>("op_42047_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_42047_end_mask_0 = const()[name = tensor<string, []>("op_42047_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42047_cast_fp16 = slice_by_index(begin = var_42047_begin_0, end = var_42047_end_0, end_mask = var_42047_end_mask_0, x = q_187_cast_fp16)[name = tensor<string, []>("op_42047_cast_fp16")];
            tensor<int32, [4]> k_375_perm_0 = const()[name = tensor<string, []>("k_375_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_42054_begin_0 = const()[name = tensor<string, []>("op_42054_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_42054_end_0 = const()[name = tensor<string, []>("op_42054_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_42054_end_mask_0 = const()[name = tensor<string, []>("op_42054_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> k_375_cast_fp16 = transpose(perm = k_375_perm_0, x = k_373_cast_fp16)[name = tensor<string, []>("transpose_46")];
            tensor<fp16, [2, 77, 1, 64]> var_42054_cast_fp16 = slice_by_index(begin = var_42054_begin_0, end = var_42054_end_0, end_mask = var_42054_end_mask_0, x = k_375_cast_fp16)[name = tensor<string, []>("op_42054_cast_fp16")];
            tensor<int32, [4]> var_42058_begin_0 = const()[name = tensor<string, []>("op_42058_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_42058_end_0 = const()[name = tensor<string, []>("op_42058_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_42058_end_mask_0 = const()[name = tensor<string, []>("op_42058_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_42058_cast_fp16 = slice_by_index(begin = var_42058_begin_0, end = var_42058_end_0, end_mask = var_42058_end_mask_0, x = k_375_cast_fp16)[name = tensor<string, []>("op_42058_cast_fp16")];
            tensor<int32, [4]> var_42062_begin_0 = const()[name = tensor<string, []>("op_42062_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_42062_end_0 = const()[name = tensor<string, []>("op_42062_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_42062_end_mask_0 = const()[name = tensor<string, []>("op_42062_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_42062_cast_fp16 = slice_by_index(begin = var_42062_begin_0, end = var_42062_end_0, end_mask = var_42062_end_mask_0, x = k_375_cast_fp16)[name = tensor<string, []>("op_42062_cast_fp16")];
            tensor<int32, [4]> var_42066_begin_0 = const()[name = tensor<string, []>("op_42066_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_42066_end_0 = const()[name = tensor<string, []>("op_42066_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_42066_end_mask_0 = const()[name = tensor<string, []>("op_42066_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_42066_cast_fp16 = slice_by_index(begin = var_42066_begin_0, end = var_42066_end_0, end_mask = var_42066_end_mask_0, x = k_375_cast_fp16)[name = tensor<string, []>("op_42066_cast_fp16")];
            tensor<int32, [4]> var_42070_begin_0 = const()[name = tensor<string, []>("op_42070_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_42070_end_0 = const()[name = tensor<string, []>("op_42070_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_42070_end_mask_0 = const()[name = tensor<string, []>("op_42070_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_42070_cast_fp16 = slice_by_index(begin = var_42070_begin_0, end = var_42070_end_0, end_mask = var_42070_end_mask_0, x = k_375_cast_fp16)[name = tensor<string, []>("op_42070_cast_fp16")];
            tensor<int32, [4]> var_42074_begin_0 = const()[name = tensor<string, []>("op_42074_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_42074_end_0 = const()[name = tensor<string, []>("op_42074_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_42074_end_mask_0 = const()[name = tensor<string, []>("op_42074_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_42074_cast_fp16 = slice_by_index(begin = var_42074_begin_0, end = var_42074_end_0, end_mask = var_42074_end_mask_0, x = k_375_cast_fp16)[name = tensor<string, []>("op_42074_cast_fp16")];
            tensor<int32, [4]> var_42078_begin_0 = const()[name = tensor<string, []>("op_42078_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_42078_end_0 = const()[name = tensor<string, []>("op_42078_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_42078_end_mask_0 = const()[name = tensor<string, []>("op_42078_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_42078_cast_fp16 = slice_by_index(begin = var_42078_begin_0, end = var_42078_end_0, end_mask = var_42078_end_mask_0, x = k_375_cast_fp16)[name = tensor<string, []>("op_42078_cast_fp16")];
            tensor<int32, [4]> var_42082_begin_0 = const()[name = tensor<string, []>("op_42082_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_42082_end_0 = const()[name = tensor<string, []>("op_42082_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_42082_end_mask_0 = const()[name = tensor<string, []>("op_42082_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_42082_cast_fp16 = slice_by_index(begin = var_42082_begin_0, end = var_42082_end_0, end_mask = var_42082_end_mask_0, x = k_375_cast_fp16)[name = tensor<string, []>("op_42082_cast_fp16")];
            tensor<int32, [4]> var_42086_begin_0 = const()[name = tensor<string, []>("op_42086_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_42086_end_0 = const()[name = tensor<string, []>("op_42086_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_42086_end_mask_0 = const()[name = tensor<string, []>("op_42086_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_42086_cast_fp16 = slice_by_index(begin = var_42086_begin_0, end = var_42086_end_0, end_mask = var_42086_end_mask_0, x = k_375_cast_fp16)[name = tensor<string, []>("op_42086_cast_fp16")];
            tensor<int32, [4]> var_42090_begin_0 = const()[name = tensor<string, []>("op_42090_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_42090_end_0 = const()[name = tensor<string, []>("op_42090_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_42090_end_mask_0 = const()[name = tensor<string, []>("op_42090_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_42090_cast_fp16 = slice_by_index(begin = var_42090_begin_0, end = var_42090_end_0, end_mask = var_42090_end_mask_0, x = k_375_cast_fp16)[name = tensor<string, []>("op_42090_cast_fp16")];
            tensor<int32, [4]> var_42094_begin_0 = const()[name = tensor<string, []>("op_42094_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_42094_end_0 = const()[name = tensor<string, []>("op_42094_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_42094_end_mask_0 = const()[name = tensor<string, []>("op_42094_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_42094_cast_fp16 = slice_by_index(begin = var_42094_begin_0, end = var_42094_end_0, end_mask = var_42094_end_mask_0, x = k_375_cast_fp16)[name = tensor<string, []>("op_42094_cast_fp16")];
            tensor<int32, [4]> var_42098_begin_0 = const()[name = tensor<string, []>("op_42098_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_42098_end_0 = const()[name = tensor<string, []>("op_42098_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_42098_end_mask_0 = const()[name = tensor<string, []>("op_42098_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_42098_cast_fp16 = slice_by_index(begin = var_42098_begin_0, end = var_42098_end_0, end_mask = var_42098_end_mask_0, x = k_375_cast_fp16)[name = tensor<string, []>("op_42098_cast_fp16")];
            tensor<int32, [4]> var_42102_begin_0 = const()[name = tensor<string, []>("op_42102_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_42102_end_0 = const()[name = tensor<string, []>("op_42102_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_42102_end_mask_0 = const()[name = tensor<string, []>("op_42102_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_42102_cast_fp16 = slice_by_index(begin = var_42102_begin_0, end = var_42102_end_0, end_mask = var_42102_end_mask_0, x = k_375_cast_fp16)[name = tensor<string, []>("op_42102_cast_fp16")];
            tensor<int32, [4]> var_42106_begin_0 = const()[name = tensor<string, []>("op_42106_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_42106_end_0 = const()[name = tensor<string, []>("op_42106_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_42106_end_mask_0 = const()[name = tensor<string, []>("op_42106_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_42106_cast_fp16 = slice_by_index(begin = var_42106_begin_0, end = var_42106_end_0, end_mask = var_42106_end_mask_0, x = k_375_cast_fp16)[name = tensor<string, []>("op_42106_cast_fp16")];
            tensor<int32, [4]> var_42110_begin_0 = const()[name = tensor<string, []>("op_42110_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_42110_end_0 = const()[name = tensor<string, []>("op_42110_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_42110_end_mask_0 = const()[name = tensor<string, []>("op_42110_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_42110_cast_fp16 = slice_by_index(begin = var_42110_begin_0, end = var_42110_end_0, end_mask = var_42110_end_mask_0, x = k_375_cast_fp16)[name = tensor<string, []>("op_42110_cast_fp16")];
            tensor<int32, [4]> var_42114_begin_0 = const()[name = tensor<string, []>("op_42114_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_42114_end_0 = const()[name = tensor<string, []>("op_42114_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_42114_end_mask_0 = const()[name = tensor<string, []>("op_42114_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_42114_cast_fp16 = slice_by_index(begin = var_42114_begin_0, end = var_42114_end_0, end_mask = var_42114_end_mask_0, x = k_375_cast_fp16)[name = tensor<string, []>("op_42114_cast_fp16")];
            tensor<int32, [4]> var_42118_begin_0 = const()[name = tensor<string, []>("op_42118_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_42118_end_0 = const()[name = tensor<string, []>("op_42118_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_42118_end_mask_0 = const()[name = tensor<string, []>("op_42118_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_42118_cast_fp16 = slice_by_index(begin = var_42118_begin_0, end = var_42118_end_0, end_mask = var_42118_end_mask_0, x = k_375_cast_fp16)[name = tensor<string, []>("op_42118_cast_fp16")];
            tensor<int32, [4]> var_42122_begin_0 = const()[name = tensor<string, []>("op_42122_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_42122_end_0 = const()[name = tensor<string, []>("op_42122_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_42122_end_mask_0 = const()[name = tensor<string, []>("op_42122_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_42122_cast_fp16 = slice_by_index(begin = var_42122_begin_0, end = var_42122_end_0, end_mask = var_42122_end_mask_0, x = k_375_cast_fp16)[name = tensor<string, []>("op_42122_cast_fp16")];
            tensor<int32, [4]> var_42126_begin_0 = const()[name = tensor<string, []>("op_42126_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_42126_end_0 = const()[name = tensor<string, []>("op_42126_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_42126_end_mask_0 = const()[name = tensor<string, []>("op_42126_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_42126_cast_fp16 = slice_by_index(begin = var_42126_begin_0, end = var_42126_end_0, end_mask = var_42126_end_mask_0, x = k_375_cast_fp16)[name = tensor<string, []>("op_42126_cast_fp16")];
            tensor<int32, [4]> var_42130_begin_0 = const()[name = tensor<string, []>("op_42130_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_42130_end_0 = const()[name = tensor<string, []>("op_42130_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_42130_end_mask_0 = const()[name = tensor<string, []>("op_42130_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_42130_cast_fp16 = slice_by_index(begin = var_42130_begin_0, end = var_42130_end_0, end_mask = var_42130_end_mask_0, x = k_375_cast_fp16)[name = tensor<string, []>("op_42130_cast_fp16")];
            tensor<int32, [4]> var_42132_begin_0 = const()[name = tensor<string, []>("op_42132_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_42132_end_0 = const()[name = tensor<string, []>("op_42132_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_42132_end_mask_0 = const()[name = tensor<string, []>("op_42132_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_42132_cast_fp16 = slice_by_index(begin = var_42132_begin_0, end = var_42132_end_0, end_mask = var_42132_end_mask_0, x = v_187_cast_fp16)[name = tensor<string, []>("op_42132_cast_fp16")];
            tensor<int32, [4]> var_42136_begin_0 = const()[name = tensor<string, []>("op_42136_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_42136_end_0 = const()[name = tensor<string, []>("op_42136_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_42136_end_mask_0 = const()[name = tensor<string, []>("op_42136_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_42136_cast_fp16 = slice_by_index(begin = var_42136_begin_0, end = var_42136_end_0, end_mask = var_42136_end_mask_0, x = v_187_cast_fp16)[name = tensor<string, []>("op_42136_cast_fp16")];
            tensor<int32, [4]> var_42140_begin_0 = const()[name = tensor<string, []>("op_42140_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_42140_end_0 = const()[name = tensor<string, []>("op_42140_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_42140_end_mask_0 = const()[name = tensor<string, []>("op_42140_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_42140_cast_fp16 = slice_by_index(begin = var_42140_begin_0, end = var_42140_end_0, end_mask = var_42140_end_mask_0, x = v_187_cast_fp16)[name = tensor<string, []>("op_42140_cast_fp16")];
            tensor<int32, [4]> var_42144_begin_0 = const()[name = tensor<string, []>("op_42144_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_42144_end_0 = const()[name = tensor<string, []>("op_42144_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_42144_end_mask_0 = const()[name = tensor<string, []>("op_42144_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_42144_cast_fp16 = slice_by_index(begin = var_42144_begin_0, end = var_42144_end_0, end_mask = var_42144_end_mask_0, x = v_187_cast_fp16)[name = tensor<string, []>("op_42144_cast_fp16")];
            tensor<int32, [4]> var_42148_begin_0 = const()[name = tensor<string, []>("op_42148_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_42148_end_0 = const()[name = tensor<string, []>("op_42148_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_42148_end_mask_0 = const()[name = tensor<string, []>("op_42148_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_42148_cast_fp16 = slice_by_index(begin = var_42148_begin_0, end = var_42148_end_0, end_mask = var_42148_end_mask_0, x = v_187_cast_fp16)[name = tensor<string, []>("op_42148_cast_fp16")];
            tensor<int32, [4]> var_42152_begin_0 = const()[name = tensor<string, []>("op_42152_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_42152_end_0 = const()[name = tensor<string, []>("op_42152_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_42152_end_mask_0 = const()[name = tensor<string, []>("op_42152_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_42152_cast_fp16 = slice_by_index(begin = var_42152_begin_0, end = var_42152_end_0, end_mask = var_42152_end_mask_0, x = v_187_cast_fp16)[name = tensor<string, []>("op_42152_cast_fp16")];
            tensor<int32, [4]> var_42156_begin_0 = const()[name = tensor<string, []>("op_42156_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_42156_end_0 = const()[name = tensor<string, []>("op_42156_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_42156_end_mask_0 = const()[name = tensor<string, []>("op_42156_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_42156_cast_fp16 = slice_by_index(begin = var_42156_begin_0, end = var_42156_end_0, end_mask = var_42156_end_mask_0, x = v_187_cast_fp16)[name = tensor<string, []>("op_42156_cast_fp16")];
            tensor<int32, [4]> var_42160_begin_0 = const()[name = tensor<string, []>("op_42160_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_42160_end_0 = const()[name = tensor<string, []>("op_42160_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_42160_end_mask_0 = const()[name = tensor<string, []>("op_42160_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_42160_cast_fp16 = slice_by_index(begin = var_42160_begin_0, end = var_42160_end_0, end_mask = var_42160_end_mask_0, x = v_187_cast_fp16)[name = tensor<string, []>("op_42160_cast_fp16")];
            tensor<int32, [4]> var_42164_begin_0 = const()[name = tensor<string, []>("op_42164_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_42164_end_0 = const()[name = tensor<string, []>("op_42164_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_42164_end_mask_0 = const()[name = tensor<string, []>("op_42164_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_42164_cast_fp16 = slice_by_index(begin = var_42164_begin_0, end = var_42164_end_0, end_mask = var_42164_end_mask_0, x = v_187_cast_fp16)[name = tensor<string, []>("op_42164_cast_fp16")];
            tensor<int32, [4]> var_42168_begin_0 = const()[name = tensor<string, []>("op_42168_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_42168_end_0 = const()[name = tensor<string, []>("op_42168_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_42168_end_mask_0 = const()[name = tensor<string, []>("op_42168_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_42168_cast_fp16 = slice_by_index(begin = var_42168_begin_0, end = var_42168_end_0, end_mask = var_42168_end_mask_0, x = v_187_cast_fp16)[name = tensor<string, []>("op_42168_cast_fp16")];
            tensor<int32, [4]> var_42172_begin_0 = const()[name = tensor<string, []>("op_42172_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_42172_end_0 = const()[name = tensor<string, []>("op_42172_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_42172_end_mask_0 = const()[name = tensor<string, []>("op_42172_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_42172_cast_fp16 = slice_by_index(begin = var_42172_begin_0, end = var_42172_end_0, end_mask = var_42172_end_mask_0, x = v_187_cast_fp16)[name = tensor<string, []>("op_42172_cast_fp16")];
            tensor<int32, [4]> var_42176_begin_0 = const()[name = tensor<string, []>("op_42176_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_42176_end_0 = const()[name = tensor<string, []>("op_42176_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_42176_end_mask_0 = const()[name = tensor<string, []>("op_42176_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_42176_cast_fp16 = slice_by_index(begin = var_42176_begin_0, end = var_42176_end_0, end_mask = var_42176_end_mask_0, x = v_187_cast_fp16)[name = tensor<string, []>("op_42176_cast_fp16")];
            tensor<int32, [4]> var_42180_begin_0 = const()[name = tensor<string, []>("op_42180_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_42180_end_0 = const()[name = tensor<string, []>("op_42180_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_42180_end_mask_0 = const()[name = tensor<string, []>("op_42180_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_42180_cast_fp16 = slice_by_index(begin = var_42180_begin_0, end = var_42180_end_0, end_mask = var_42180_end_mask_0, x = v_187_cast_fp16)[name = tensor<string, []>("op_42180_cast_fp16")];
            tensor<int32, [4]> var_42184_begin_0 = const()[name = tensor<string, []>("op_42184_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_42184_end_0 = const()[name = tensor<string, []>("op_42184_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_42184_end_mask_0 = const()[name = tensor<string, []>("op_42184_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_42184_cast_fp16 = slice_by_index(begin = var_42184_begin_0, end = var_42184_end_0, end_mask = var_42184_end_mask_0, x = v_187_cast_fp16)[name = tensor<string, []>("op_42184_cast_fp16")];
            tensor<int32, [4]> var_42188_begin_0 = const()[name = tensor<string, []>("op_42188_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_42188_end_0 = const()[name = tensor<string, []>("op_42188_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_42188_end_mask_0 = const()[name = tensor<string, []>("op_42188_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_42188_cast_fp16 = slice_by_index(begin = var_42188_begin_0, end = var_42188_end_0, end_mask = var_42188_end_mask_0, x = v_187_cast_fp16)[name = tensor<string, []>("op_42188_cast_fp16")];
            tensor<int32, [4]> var_42192_begin_0 = const()[name = tensor<string, []>("op_42192_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_42192_end_0 = const()[name = tensor<string, []>("op_42192_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_42192_end_mask_0 = const()[name = tensor<string, []>("op_42192_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_42192_cast_fp16 = slice_by_index(begin = var_42192_begin_0, end = var_42192_end_0, end_mask = var_42192_end_mask_0, x = v_187_cast_fp16)[name = tensor<string, []>("op_42192_cast_fp16")];
            tensor<int32, [4]> var_42196_begin_0 = const()[name = tensor<string, []>("op_42196_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_42196_end_0 = const()[name = tensor<string, []>("op_42196_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_42196_end_mask_0 = const()[name = tensor<string, []>("op_42196_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_42196_cast_fp16 = slice_by_index(begin = var_42196_begin_0, end = var_42196_end_0, end_mask = var_42196_end_mask_0, x = v_187_cast_fp16)[name = tensor<string, []>("op_42196_cast_fp16")];
            tensor<int32, [4]> var_42200_begin_0 = const()[name = tensor<string, []>("op_42200_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_42200_end_0 = const()[name = tensor<string, []>("op_42200_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_42200_end_mask_0 = const()[name = tensor<string, []>("op_42200_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_42200_cast_fp16 = slice_by_index(begin = var_42200_begin_0, end = var_42200_end_0, end_mask = var_42200_end_mask_0, x = v_187_cast_fp16)[name = tensor<string, []>("op_42200_cast_fp16")];
            tensor<int32, [4]> var_42204_begin_0 = const()[name = tensor<string, []>("op_42204_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_42204_end_0 = const()[name = tensor<string, []>("op_42204_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_42204_end_mask_0 = const()[name = tensor<string, []>("op_42204_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_42204_cast_fp16 = slice_by_index(begin = var_42204_begin_0, end = var_42204_end_0, end_mask = var_42204_end_mask_0, x = v_187_cast_fp16)[name = tensor<string, []>("op_42204_cast_fp16")];
            tensor<int32, [4]> var_42208_begin_0 = const()[name = tensor<string, []>("op_42208_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_42208_end_0 = const()[name = tensor<string, []>("op_42208_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_42208_end_mask_0 = const()[name = tensor<string, []>("op_42208_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_42208_cast_fp16 = slice_by_index(begin = var_42208_begin_0, end = var_42208_end_0, end_mask = var_42208_end_mask_0, x = v_187_cast_fp16)[name = tensor<string, []>("op_42208_cast_fp16")];
            tensor<string, []> var_42212_equation_0 = const()[name = tensor<string, []>("op_42212_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_42212_cast_fp16 = einsum(equation = var_42212_equation_0, values = (var_42054_cast_fp16, var_41971_cast_fp16))[name = tensor<string, []>("op_42212_cast_fp16")];
            tensor<fp16, []> var_42213_to_fp16 = const()[name = tensor<string, []>("op_42213_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3561_cast_fp16 = mul(x = var_42212_cast_fp16, y = var_42213_to_fp16)[name = tensor<string, []>("aw_3561_cast_fp16")];
            tensor<string, []> var_42216_equation_0 = const()[name = tensor<string, []>("op_42216_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_42216_cast_fp16 = einsum(equation = var_42216_equation_0, values = (var_42058_cast_fp16, var_41975_cast_fp16))[name = tensor<string, []>("op_42216_cast_fp16")];
            tensor<fp16, []> var_42217_to_fp16 = const()[name = tensor<string, []>("op_42217_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3563_cast_fp16 = mul(x = var_42216_cast_fp16, y = var_42217_to_fp16)[name = tensor<string, []>("aw_3563_cast_fp16")];
            tensor<string, []> var_42220_equation_0 = const()[name = tensor<string, []>("op_42220_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_42220_cast_fp16 = einsum(equation = var_42220_equation_0, values = (var_42062_cast_fp16, var_41979_cast_fp16))[name = tensor<string, []>("op_42220_cast_fp16")];
            tensor<fp16, []> var_42221_to_fp16 = const()[name = tensor<string, []>("op_42221_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3565_cast_fp16 = mul(x = var_42220_cast_fp16, y = var_42221_to_fp16)[name = tensor<string, []>("aw_3565_cast_fp16")];
            tensor<string, []> var_42224_equation_0 = const()[name = tensor<string, []>("op_42224_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_42224_cast_fp16 = einsum(equation = var_42224_equation_0, values = (var_42066_cast_fp16, var_41983_cast_fp16))[name = tensor<string, []>("op_42224_cast_fp16")];
            tensor<fp16, []> var_42225_to_fp16 = const()[name = tensor<string, []>("op_42225_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3567_cast_fp16 = mul(x = var_42224_cast_fp16, y = var_42225_to_fp16)[name = tensor<string, []>("aw_3567_cast_fp16")];
            tensor<string, []> var_42228_equation_0 = const()[name = tensor<string, []>("op_42228_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_42228_cast_fp16 = einsum(equation = var_42228_equation_0, values = (var_42070_cast_fp16, var_41987_cast_fp16))[name = tensor<string, []>("op_42228_cast_fp16")];
            tensor<fp16, []> var_42229_to_fp16 = const()[name = tensor<string, []>("op_42229_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3569_cast_fp16 = mul(x = var_42228_cast_fp16, y = var_42229_to_fp16)[name = tensor<string, []>("aw_3569_cast_fp16")];
            tensor<string, []> var_42232_equation_0 = const()[name = tensor<string, []>("op_42232_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_42232_cast_fp16 = einsum(equation = var_42232_equation_0, values = (var_42074_cast_fp16, var_41991_cast_fp16))[name = tensor<string, []>("op_42232_cast_fp16")];
            tensor<fp16, []> var_42233_to_fp16 = const()[name = tensor<string, []>("op_42233_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3571_cast_fp16 = mul(x = var_42232_cast_fp16, y = var_42233_to_fp16)[name = tensor<string, []>("aw_3571_cast_fp16")];
            tensor<string, []> var_42236_equation_0 = const()[name = tensor<string, []>("op_42236_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_42236_cast_fp16 = einsum(equation = var_42236_equation_0, values = (var_42078_cast_fp16, var_41995_cast_fp16))[name = tensor<string, []>("op_42236_cast_fp16")];
            tensor<fp16, []> var_42237_to_fp16 = const()[name = tensor<string, []>("op_42237_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3573_cast_fp16 = mul(x = var_42236_cast_fp16, y = var_42237_to_fp16)[name = tensor<string, []>("aw_3573_cast_fp16")];
            tensor<string, []> var_42240_equation_0 = const()[name = tensor<string, []>("op_42240_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_42240_cast_fp16 = einsum(equation = var_42240_equation_0, values = (var_42082_cast_fp16, var_41999_cast_fp16))[name = tensor<string, []>("op_42240_cast_fp16")];
            tensor<fp16, []> var_42241_to_fp16 = const()[name = tensor<string, []>("op_42241_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3575_cast_fp16 = mul(x = var_42240_cast_fp16, y = var_42241_to_fp16)[name = tensor<string, []>("aw_3575_cast_fp16")];
            tensor<string, []> var_42244_equation_0 = const()[name = tensor<string, []>("op_42244_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_42244_cast_fp16 = einsum(equation = var_42244_equation_0, values = (var_42086_cast_fp16, var_42003_cast_fp16))[name = tensor<string, []>("op_42244_cast_fp16")];
            tensor<fp16, []> var_42245_to_fp16 = const()[name = tensor<string, []>("op_42245_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3577_cast_fp16 = mul(x = var_42244_cast_fp16, y = var_42245_to_fp16)[name = tensor<string, []>("aw_3577_cast_fp16")];
            tensor<string, []> var_42248_equation_0 = const()[name = tensor<string, []>("op_42248_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_42248_cast_fp16 = einsum(equation = var_42248_equation_0, values = (var_42090_cast_fp16, var_42007_cast_fp16))[name = tensor<string, []>("op_42248_cast_fp16")];
            tensor<fp16, []> var_42249_to_fp16 = const()[name = tensor<string, []>("op_42249_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3579_cast_fp16 = mul(x = var_42248_cast_fp16, y = var_42249_to_fp16)[name = tensor<string, []>("aw_3579_cast_fp16")];
            tensor<string, []> var_42252_equation_0 = const()[name = tensor<string, []>("op_42252_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_42252_cast_fp16 = einsum(equation = var_42252_equation_0, values = (var_42094_cast_fp16, var_42011_cast_fp16))[name = tensor<string, []>("op_42252_cast_fp16")];
            tensor<fp16, []> var_42253_to_fp16 = const()[name = tensor<string, []>("op_42253_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3581_cast_fp16 = mul(x = var_42252_cast_fp16, y = var_42253_to_fp16)[name = tensor<string, []>("aw_3581_cast_fp16")];
            tensor<string, []> var_42256_equation_0 = const()[name = tensor<string, []>("op_42256_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_42256_cast_fp16 = einsum(equation = var_42256_equation_0, values = (var_42098_cast_fp16, var_42015_cast_fp16))[name = tensor<string, []>("op_42256_cast_fp16")];
            tensor<fp16, []> var_42257_to_fp16 = const()[name = tensor<string, []>("op_42257_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3583_cast_fp16 = mul(x = var_42256_cast_fp16, y = var_42257_to_fp16)[name = tensor<string, []>("aw_3583_cast_fp16")];
            tensor<string, []> var_42260_equation_0 = const()[name = tensor<string, []>("op_42260_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_42260_cast_fp16 = einsum(equation = var_42260_equation_0, values = (var_42102_cast_fp16, var_42019_cast_fp16))[name = tensor<string, []>("op_42260_cast_fp16")];
            tensor<fp16, []> var_42261_to_fp16 = const()[name = tensor<string, []>("op_42261_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3585_cast_fp16 = mul(x = var_42260_cast_fp16, y = var_42261_to_fp16)[name = tensor<string, []>("aw_3585_cast_fp16")];
            tensor<string, []> var_42264_equation_0 = const()[name = tensor<string, []>("op_42264_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_42264_cast_fp16 = einsum(equation = var_42264_equation_0, values = (var_42106_cast_fp16, var_42023_cast_fp16))[name = tensor<string, []>("op_42264_cast_fp16")];
            tensor<fp16, []> var_42265_to_fp16 = const()[name = tensor<string, []>("op_42265_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3587_cast_fp16 = mul(x = var_42264_cast_fp16, y = var_42265_to_fp16)[name = tensor<string, []>("aw_3587_cast_fp16")];
            tensor<string, []> var_42268_equation_0 = const()[name = tensor<string, []>("op_42268_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_42268_cast_fp16 = einsum(equation = var_42268_equation_0, values = (var_42110_cast_fp16, var_42027_cast_fp16))[name = tensor<string, []>("op_42268_cast_fp16")];
            tensor<fp16, []> var_42269_to_fp16 = const()[name = tensor<string, []>("op_42269_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3589_cast_fp16 = mul(x = var_42268_cast_fp16, y = var_42269_to_fp16)[name = tensor<string, []>("aw_3589_cast_fp16")];
            tensor<string, []> var_42272_equation_0 = const()[name = tensor<string, []>("op_42272_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_42272_cast_fp16 = einsum(equation = var_42272_equation_0, values = (var_42114_cast_fp16, var_42031_cast_fp16))[name = tensor<string, []>("op_42272_cast_fp16")];
            tensor<fp16, []> var_42273_to_fp16 = const()[name = tensor<string, []>("op_42273_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3591_cast_fp16 = mul(x = var_42272_cast_fp16, y = var_42273_to_fp16)[name = tensor<string, []>("aw_3591_cast_fp16")];
            tensor<string, []> var_42276_equation_0 = const()[name = tensor<string, []>("op_42276_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_42276_cast_fp16 = einsum(equation = var_42276_equation_0, values = (var_42118_cast_fp16, var_42035_cast_fp16))[name = tensor<string, []>("op_42276_cast_fp16")];
            tensor<fp16, []> var_42277_to_fp16 = const()[name = tensor<string, []>("op_42277_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3593_cast_fp16 = mul(x = var_42276_cast_fp16, y = var_42277_to_fp16)[name = tensor<string, []>("aw_3593_cast_fp16")];
            tensor<string, []> var_42280_equation_0 = const()[name = tensor<string, []>("op_42280_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_42280_cast_fp16 = einsum(equation = var_42280_equation_0, values = (var_42122_cast_fp16, var_42039_cast_fp16))[name = tensor<string, []>("op_42280_cast_fp16")];
            tensor<fp16, []> var_42281_to_fp16 = const()[name = tensor<string, []>("op_42281_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3595_cast_fp16 = mul(x = var_42280_cast_fp16, y = var_42281_to_fp16)[name = tensor<string, []>("aw_3595_cast_fp16")];
            tensor<string, []> var_42284_equation_0 = const()[name = tensor<string, []>("op_42284_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_42284_cast_fp16 = einsum(equation = var_42284_equation_0, values = (var_42126_cast_fp16, var_42043_cast_fp16))[name = tensor<string, []>("op_42284_cast_fp16")];
            tensor<fp16, []> var_42285_to_fp16 = const()[name = tensor<string, []>("op_42285_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3597_cast_fp16 = mul(x = var_42284_cast_fp16, y = var_42285_to_fp16)[name = tensor<string, []>("aw_3597_cast_fp16")];
            tensor<string, []> var_42288_equation_0 = const()[name = tensor<string, []>("op_42288_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_42288_cast_fp16 = einsum(equation = var_42288_equation_0, values = (var_42130_cast_fp16, var_42047_cast_fp16))[name = tensor<string, []>("op_42288_cast_fp16")];
            tensor<fp16, []> var_42289_to_fp16 = const()[name = tensor<string, []>("op_42289_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3599_cast_fp16 = mul(x = var_42288_cast_fp16, y = var_42289_to_fp16)[name = tensor<string, []>("aw_3599_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_42291_cast_fp16 = softmax(axis = var_30355, x = aw_3561_cast_fp16)[name = tensor<string, []>("op_42291_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_42292_cast_fp16 = softmax(axis = var_30355, x = aw_3563_cast_fp16)[name = tensor<string, []>("op_42292_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_42293_cast_fp16 = softmax(axis = var_30355, x = aw_3565_cast_fp16)[name = tensor<string, []>("op_42293_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_42294_cast_fp16 = softmax(axis = var_30355, x = aw_3567_cast_fp16)[name = tensor<string, []>("op_42294_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_42295_cast_fp16 = softmax(axis = var_30355, x = aw_3569_cast_fp16)[name = tensor<string, []>("op_42295_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_42296_cast_fp16 = softmax(axis = var_30355, x = aw_3571_cast_fp16)[name = tensor<string, []>("op_42296_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_42297_cast_fp16 = softmax(axis = var_30355, x = aw_3573_cast_fp16)[name = tensor<string, []>("op_42297_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_42298_cast_fp16 = softmax(axis = var_30355, x = aw_3575_cast_fp16)[name = tensor<string, []>("op_42298_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_42299_cast_fp16 = softmax(axis = var_30355, x = aw_3577_cast_fp16)[name = tensor<string, []>("op_42299_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_42300_cast_fp16 = softmax(axis = var_30355, x = aw_3579_cast_fp16)[name = tensor<string, []>("op_42300_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_42301_cast_fp16 = softmax(axis = var_30355, x = aw_3581_cast_fp16)[name = tensor<string, []>("op_42301_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_42302_cast_fp16 = softmax(axis = var_30355, x = aw_3583_cast_fp16)[name = tensor<string, []>("op_42302_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_42303_cast_fp16 = softmax(axis = var_30355, x = aw_3585_cast_fp16)[name = tensor<string, []>("op_42303_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_42304_cast_fp16 = softmax(axis = var_30355, x = aw_3587_cast_fp16)[name = tensor<string, []>("op_42304_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_42305_cast_fp16 = softmax(axis = var_30355, x = aw_3589_cast_fp16)[name = tensor<string, []>("op_42305_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_42306_cast_fp16 = softmax(axis = var_30355, x = aw_3591_cast_fp16)[name = tensor<string, []>("op_42306_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_42307_cast_fp16 = softmax(axis = var_30355, x = aw_3593_cast_fp16)[name = tensor<string, []>("op_42307_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_42308_cast_fp16 = softmax(axis = var_30355, x = aw_3595_cast_fp16)[name = tensor<string, []>("op_42308_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_42309_cast_fp16 = softmax(axis = var_30355, x = aw_3597_cast_fp16)[name = tensor<string, []>("op_42309_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_42310_cast_fp16 = softmax(axis = var_30355, x = aw_3599_cast_fp16)[name = tensor<string, []>("op_42310_cast_fp16")];
            tensor<string, []> var_42312_equation_0 = const()[name = tensor<string, []>("op_42312_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42312_cast_fp16 = einsum(equation = var_42312_equation_0, values = (var_42132_cast_fp16, var_42291_cast_fp16))[name = tensor<string, []>("op_42312_cast_fp16")];
            tensor<string, []> var_42314_equation_0 = const()[name = tensor<string, []>("op_42314_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42314_cast_fp16 = einsum(equation = var_42314_equation_0, values = (var_42136_cast_fp16, var_42292_cast_fp16))[name = tensor<string, []>("op_42314_cast_fp16")];
            tensor<string, []> var_42316_equation_0 = const()[name = tensor<string, []>("op_42316_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42316_cast_fp16 = einsum(equation = var_42316_equation_0, values = (var_42140_cast_fp16, var_42293_cast_fp16))[name = tensor<string, []>("op_42316_cast_fp16")];
            tensor<string, []> var_42318_equation_0 = const()[name = tensor<string, []>("op_42318_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42318_cast_fp16 = einsum(equation = var_42318_equation_0, values = (var_42144_cast_fp16, var_42294_cast_fp16))[name = tensor<string, []>("op_42318_cast_fp16")];
            tensor<string, []> var_42320_equation_0 = const()[name = tensor<string, []>("op_42320_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42320_cast_fp16 = einsum(equation = var_42320_equation_0, values = (var_42148_cast_fp16, var_42295_cast_fp16))[name = tensor<string, []>("op_42320_cast_fp16")];
            tensor<string, []> var_42322_equation_0 = const()[name = tensor<string, []>("op_42322_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42322_cast_fp16 = einsum(equation = var_42322_equation_0, values = (var_42152_cast_fp16, var_42296_cast_fp16))[name = tensor<string, []>("op_42322_cast_fp16")];
            tensor<string, []> var_42324_equation_0 = const()[name = tensor<string, []>("op_42324_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42324_cast_fp16 = einsum(equation = var_42324_equation_0, values = (var_42156_cast_fp16, var_42297_cast_fp16))[name = tensor<string, []>("op_42324_cast_fp16")];
            tensor<string, []> var_42326_equation_0 = const()[name = tensor<string, []>("op_42326_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42326_cast_fp16 = einsum(equation = var_42326_equation_0, values = (var_42160_cast_fp16, var_42298_cast_fp16))[name = tensor<string, []>("op_42326_cast_fp16")];
            tensor<string, []> var_42328_equation_0 = const()[name = tensor<string, []>("op_42328_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42328_cast_fp16 = einsum(equation = var_42328_equation_0, values = (var_42164_cast_fp16, var_42299_cast_fp16))[name = tensor<string, []>("op_42328_cast_fp16")];
            tensor<string, []> var_42330_equation_0 = const()[name = tensor<string, []>("op_42330_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42330_cast_fp16 = einsum(equation = var_42330_equation_0, values = (var_42168_cast_fp16, var_42300_cast_fp16))[name = tensor<string, []>("op_42330_cast_fp16")];
            tensor<string, []> var_42332_equation_0 = const()[name = tensor<string, []>("op_42332_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42332_cast_fp16 = einsum(equation = var_42332_equation_0, values = (var_42172_cast_fp16, var_42301_cast_fp16))[name = tensor<string, []>("op_42332_cast_fp16")];
            tensor<string, []> var_42334_equation_0 = const()[name = tensor<string, []>("op_42334_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42334_cast_fp16 = einsum(equation = var_42334_equation_0, values = (var_42176_cast_fp16, var_42302_cast_fp16))[name = tensor<string, []>("op_42334_cast_fp16")];
            tensor<string, []> var_42336_equation_0 = const()[name = tensor<string, []>("op_42336_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42336_cast_fp16 = einsum(equation = var_42336_equation_0, values = (var_42180_cast_fp16, var_42303_cast_fp16))[name = tensor<string, []>("op_42336_cast_fp16")];
            tensor<string, []> var_42338_equation_0 = const()[name = tensor<string, []>("op_42338_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42338_cast_fp16 = einsum(equation = var_42338_equation_0, values = (var_42184_cast_fp16, var_42304_cast_fp16))[name = tensor<string, []>("op_42338_cast_fp16")];
            tensor<string, []> var_42340_equation_0 = const()[name = tensor<string, []>("op_42340_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42340_cast_fp16 = einsum(equation = var_42340_equation_0, values = (var_42188_cast_fp16, var_42305_cast_fp16))[name = tensor<string, []>("op_42340_cast_fp16")];
            tensor<string, []> var_42342_equation_0 = const()[name = tensor<string, []>("op_42342_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42342_cast_fp16 = einsum(equation = var_42342_equation_0, values = (var_42192_cast_fp16, var_42306_cast_fp16))[name = tensor<string, []>("op_42342_cast_fp16")];
            tensor<string, []> var_42344_equation_0 = const()[name = tensor<string, []>("op_42344_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42344_cast_fp16 = einsum(equation = var_42344_equation_0, values = (var_42196_cast_fp16, var_42307_cast_fp16))[name = tensor<string, []>("op_42344_cast_fp16")];
            tensor<string, []> var_42346_equation_0 = const()[name = tensor<string, []>("op_42346_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42346_cast_fp16 = einsum(equation = var_42346_equation_0, values = (var_42200_cast_fp16, var_42308_cast_fp16))[name = tensor<string, []>("op_42346_cast_fp16")];
            tensor<string, []> var_42348_equation_0 = const()[name = tensor<string, []>("op_42348_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42348_cast_fp16 = einsum(equation = var_42348_equation_0, values = (var_42204_cast_fp16, var_42309_cast_fp16))[name = tensor<string, []>("op_42348_cast_fp16")];
            tensor<string, []> var_42350_equation_0 = const()[name = tensor<string, []>("op_42350_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42350_cast_fp16 = einsum(equation = var_42350_equation_0, values = (var_42208_cast_fp16, var_42310_cast_fp16))[name = tensor<string, []>("op_42350_cast_fp16")];
            tensor<bool, []> input_555_interleave_0 = const()[name = tensor<string, []>("input_555_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_555_cast_fp16 = concat(axis = var_30355, interleave = input_555_interleave_0, values = (var_42312_cast_fp16, var_42314_cast_fp16, var_42316_cast_fp16, var_42318_cast_fp16, var_42320_cast_fp16, var_42322_cast_fp16, var_42324_cast_fp16, var_42326_cast_fp16, var_42328_cast_fp16, var_42330_cast_fp16, var_42332_cast_fp16, var_42334_cast_fp16, var_42336_cast_fp16, var_42338_cast_fp16, var_42340_cast_fp16, var_42342_cast_fp16, var_42344_cast_fp16, var_42346_cast_fp16, var_42348_cast_fp16, var_42350_cast_fp16))[name = tensor<string, []>("input_555_cast_fp16")];
            tensor<string, []> var_42360_pad_type_0 = const()[name = tensor<string, []>("op_42360_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_42360_strides_0 = const()[name = tensor<string, []>("op_42360_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_42360_pad_0 = const()[name = tensor<string, []>("op_42360_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_42360_dilations_0 = const()[name = tensor<string, []>("op_42360_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_42360_groups_0 = const()[name = tensor<string, []>("op_42360_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_2_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(366335360))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(367564224))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_2_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_2_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_2_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(367564416)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_42360_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_2_attn2_to_out_0_bias_to_fp16, dilations = var_42360_dilations_0, groups = var_42360_groups_0, pad = var_42360_pad_0, pad_type = var_42360_pad_type_0, strides = var_42360_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_2_attn2_to_out_0_weight_to_fp16_palettized, x = input_555_cast_fp16)[name = tensor<string, []>("op_42360_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_281_cast_fp16 = add(x = var_42360_cast_fp16, y = inputs_279_cast_fp16)[name = tensor<string, []>("inputs_281_cast_fp16")];
            tensor<int32, [1]> input_557_axes_0 = const()[name = tensor<string, []>("input_557_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> input_557_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_557_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(367567040)))];
            tensor<fp16, [1280]> input_557_beta_0_to_fp16 = const()[name = tensor<string, []>("input_557_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(367569664)))];
            tensor<fp16, []> var_42370_to_fp16 = const()[name = tensor<string, []>("op_42370_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> input_557_cast_fp16 = layer_norm(axes = input_557_axes_0, beta = input_557_beta_0_to_fp16, epsilon = var_42370_to_fp16, gamma = input_557_gamma_0_to_fp16, x = inputs_281_cast_fp16)[name = tensor<string, []>("input_557_cast_fp16")];
            tensor<string, []> var_42390_pad_type_0 = const()[name = tensor<string, []>("op_42390_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_42390_strides_0 = const()[name = tensor<string, []>("op_42390_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_42390_pad_0 = const()[name = tensor<string, []>("op_42390_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_42390_dilations_0 = const()[name = tensor<string, []>("op_42390_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_42390_groups_0 = const()[name = tensor<string, []>("op_42390_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_2_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [9830400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(367572288))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(377402752))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_2_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([10240, 1280, 1, 1])];
            tensor<fp16, [10240]> up_blocks_0_attentions_1_transformer_blocks_2_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_2_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(377402944)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_42390_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_2_ff_net_0_proj_bias_to_fp16, dilations = var_42390_dilations_0, groups = var_42390_groups_0, pad = var_42390_pad_0, pad_type = var_42390_pad_type_0, strides = var_42390_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_2_ff_net_0_proj_weight_to_fp16_palettized, x = input_557_cast_fp16)[name = tensor<string, []>("op_42390_cast_fp16")];
            tensor<int32, [2]> var_42391_split_sizes_0 = const()[name = tensor<string, []>("op_42391_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_42391_axis_0 = const()[name = tensor<string, []>("op_42391_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_42391_cast_fp16_0, tensor<fp16, [2, 5120, 1, 1024]> var_42391_cast_fp16_1 = split(axis = var_42391_axis_0, split_sizes = var_42391_split_sizes_0, x = var_42390_cast_fp16)[name = tensor<string, []>("op_42391_cast_fp16")];
            tensor<string, []> var_42393_mode_0 = const()[name = tensor<string, []>("op_42393_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_42393_cast_fp16 = gelu(mode = var_42393_mode_0, x = var_42391_cast_fp16_1)[name = tensor<string, []>("op_42393_cast_fp16")];
            tensor<fp16, [2, 5120, 1, 1024]> input_559_cast_fp16 = mul(x = var_42391_cast_fp16_0, y = var_42393_cast_fp16)[name = tensor<string, []>("input_559_cast_fp16")];
            tensor<string, []> var_42401_pad_type_0 = const()[name = tensor<string, []>("op_42401_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_42401_strides_0 = const()[name = tensor<string, []>("op_42401_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_42401_pad_0 = const()[name = tensor<string, []>("op_42401_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_42401_dilations_0 = const()[name = tensor<string, []>("op_42401_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_42401_groups_0 = const()[name = tensor<string, []>("op_42401_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_2_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [4915200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(377423488))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(382338752))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_2_ff_net_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 5120, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_2_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_2_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(382338944)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_42401_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_2_ff_net_2_bias_to_fp16, dilations = var_42401_dilations_0, groups = var_42401_groups_0, pad = var_42401_pad_0, pad_type = var_42401_pad_type_0, strides = var_42401_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_2_ff_net_2_weight_to_fp16_palettized, x = input_559_cast_fp16)[name = tensor<string, []>("op_42401_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_283_cast_fp16 = add(x = var_42401_cast_fp16, y = inputs_281_cast_fp16)[name = tensor<string, []>("inputs_283_cast_fp16")];
            tensor<int32, [1]> hidden_states_379_axes_0 = const()[name = tensor<string, []>("hidden_states_379_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_379_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_379_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(382341568)))];
            tensor<fp16, [1280]> hidden_states_379_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_379_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(382344192)))];
            tensor<fp16, []> var_42417_to_fp16 = const()[name = tensor<string, []>("op_42417_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_379_cast_fp16 = layer_norm(axes = hidden_states_379_axes_0, beta = hidden_states_379_beta_0_to_fp16, epsilon = var_42417_to_fp16, gamma = hidden_states_379_gamma_0_to_fp16, x = inputs_283_cast_fp16)[name = tensor<string, []>("hidden_states_379_cast_fp16")];
            tensor<string, []> q_189_pad_type_0 = const()[name = tensor<string, []>("q_189_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_189_strides_0 = const()[name = tensor<string, []>("q_189_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_189_pad_0 = const()[name = tensor<string, []>("q_189_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_189_dilations_0 = const()[name = tensor<string, []>("q_189_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_189_groups_0 = const()[name = tensor<string, []>("q_189_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_3_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(382346816))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(383575680))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_3_attn1_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_189_cast_fp16 = conv(dilations = q_189_dilations_0, groups = q_189_groups_0, pad = q_189_pad_0, pad_type = q_189_pad_type_0, strides = q_189_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_3_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_379_cast_fp16)[name = tensor<string, []>("q_189_cast_fp16")];
            tensor<string, []> k_377_pad_type_0 = const()[name = tensor<string, []>("k_377_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_377_strides_0 = const()[name = tensor<string, []>("k_377_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_377_pad_0 = const()[name = tensor<string, []>("k_377_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_377_dilations_0 = const()[name = tensor<string, []>("k_377_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_377_groups_0 = const()[name = tensor<string, []>("k_377_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_3_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(383575872))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(384804736))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_3_attn1_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> k_377_cast_fp16 = conv(dilations = k_377_dilations_0, groups = k_377_groups_0, pad = k_377_pad_0, pad_type = k_377_pad_type_0, strides = k_377_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_3_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_379_cast_fp16)[name = tensor<string, []>("k_377_cast_fp16")];
            tensor<string, []> v_189_pad_type_0 = const()[name = tensor<string, []>("v_189_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_189_strides_0 = const()[name = tensor<string, []>("v_189_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_189_pad_0 = const()[name = tensor<string, []>("v_189_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_189_dilations_0 = const()[name = tensor<string, []>("v_189_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_189_groups_0 = const()[name = tensor<string, []>("v_189_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_3_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(384804928))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(386033792))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_3_attn1_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> v_189_cast_fp16 = conv(dilations = v_189_dilations_0, groups = v_189_groups_0, pad = v_189_pad_0, pad_type = v_189_pad_type_0, strides = v_189_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_3_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_379_cast_fp16)[name = tensor<string, []>("v_189_cast_fp16")];
            tensor<int32, [4]> var_42450_begin_0 = const()[name = tensor<string, []>("op_42450_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_42450_end_0 = const()[name = tensor<string, []>("op_42450_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_42450_end_mask_0 = const()[name = tensor<string, []>("op_42450_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42450_cast_fp16 = slice_by_index(begin = var_42450_begin_0, end = var_42450_end_0, end_mask = var_42450_end_mask_0, x = q_189_cast_fp16)[name = tensor<string, []>("op_42450_cast_fp16")];
            tensor<int32, [4]> var_42454_begin_0 = const()[name = tensor<string, []>("op_42454_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_42454_end_0 = const()[name = tensor<string, []>("op_42454_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_42454_end_mask_0 = const()[name = tensor<string, []>("op_42454_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42454_cast_fp16 = slice_by_index(begin = var_42454_begin_0, end = var_42454_end_0, end_mask = var_42454_end_mask_0, x = q_189_cast_fp16)[name = tensor<string, []>("op_42454_cast_fp16")];
            tensor<int32, [4]> var_42458_begin_0 = const()[name = tensor<string, []>("op_42458_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_42458_end_0 = const()[name = tensor<string, []>("op_42458_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_42458_end_mask_0 = const()[name = tensor<string, []>("op_42458_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42458_cast_fp16 = slice_by_index(begin = var_42458_begin_0, end = var_42458_end_0, end_mask = var_42458_end_mask_0, x = q_189_cast_fp16)[name = tensor<string, []>("op_42458_cast_fp16")];
            tensor<int32, [4]> var_42462_begin_0 = const()[name = tensor<string, []>("op_42462_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_42462_end_0 = const()[name = tensor<string, []>("op_42462_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_42462_end_mask_0 = const()[name = tensor<string, []>("op_42462_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42462_cast_fp16 = slice_by_index(begin = var_42462_begin_0, end = var_42462_end_0, end_mask = var_42462_end_mask_0, x = q_189_cast_fp16)[name = tensor<string, []>("op_42462_cast_fp16")];
            tensor<int32, [4]> var_42466_begin_0 = const()[name = tensor<string, []>("op_42466_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_42466_end_0 = const()[name = tensor<string, []>("op_42466_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_42466_end_mask_0 = const()[name = tensor<string, []>("op_42466_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42466_cast_fp16 = slice_by_index(begin = var_42466_begin_0, end = var_42466_end_0, end_mask = var_42466_end_mask_0, x = q_189_cast_fp16)[name = tensor<string, []>("op_42466_cast_fp16")];
            tensor<int32, [4]> var_42470_begin_0 = const()[name = tensor<string, []>("op_42470_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_42470_end_0 = const()[name = tensor<string, []>("op_42470_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_42470_end_mask_0 = const()[name = tensor<string, []>("op_42470_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42470_cast_fp16 = slice_by_index(begin = var_42470_begin_0, end = var_42470_end_0, end_mask = var_42470_end_mask_0, x = q_189_cast_fp16)[name = tensor<string, []>("op_42470_cast_fp16")];
            tensor<int32, [4]> var_42474_begin_0 = const()[name = tensor<string, []>("op_42474_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_42474_end_0 = const()[name = tensor<string, []>("op_42474_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_42474_end_mask_0 = const()[name = tensor<string, []>("op_42474_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42474_cast_fp16 = slice_by_index(begin = var_42474_begin_0, end = var_42474_end_0, end_mask = var_42474_end_mask_0, x = q_189_cast_fp16)[name = tensor<string, []>("op_42474_cast_fp16")];
            tensor<int32, [4]> var_42478_begin_0 = const()[name = tensor<string, []>("op_42478_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_42478_end_0 = const()[name = tensor<string, []>("op_42478_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_42478_end_mask_0 = const()[name = tensor<string, []>("op_42478_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42478_cast_fp16 = slice_by_index(begin = var_42478_begin_0, end = var_42478_end_0, end_mask = var_42478_end_mask_0, x = q_189_cast_fp16)[name = tensor<string, []>("op_42478_cast_fp16")];
            tensor<int32, [4]> var_42482_begin_0 = const()[name = tensor<string, []>("op_42482_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_42482_end_0 = const()[name = tensor<string, []>("op_42482_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_42482_end_mask_0 = const()[name = tensor<string, []>("op_42482_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42482_cast_fp16 = slice_by_index(begin = var_42482_begin_0, end = var_42482_end_0, end_mask = var_42482_end_mask_0, x = q_189_cast_fp16)[name = tensor<string, []>("op_42482_cast_fp16")];
            tensor<int32, [4]> var_42486_begin_0 = const()[name = tensor<string, []>("op_42486_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_42486_end_0 = const()[name = tensor<string, []>("op_42486_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_42486_end_mask_0 = const()[name = tensor<string, []>("op_42486_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42486_cast_fp16 = slice_by_index(begin = var_42486_begin_0, end = var_42486_end_0, end_mask = var_42486_end_mask_0, x = q_189_cast_fp16)[name = tensor<string, []>("op_42486_cast_fp16")];
            tensor<int32, [4]> var_42490_begin_0 = const()[name = tensor<string, []>("op_42490_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_42490_end_0 = const()[name = tensor<string, []>("op_42490_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_42490_end_mask_0 = const()[name = tensor<string, []>("op_42490_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42490_cast_fp16 = slice_by_index(begin = var_42490_begin_0, end = var_42490_end_0, end_mask = var_42490_end_mask_0, x = q_189_cast_fp16)[name = tensor<string, []>("op_42490_cast_fp16")];
            tensor<int32, [4]> var_42494_begin_0 = const()[name = tensor<string, []>("op_42494_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_42494_end_0 = const()[name = tensor<string, []>("op_42494_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_42494_end_mask_0 = const()[name = tensor<string, []>("op_42494_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42494_cast_fp16 = slice_by_index(begin = var_42494_begin_0, end = var_42494_end_0, end_mask = var_42494_end_mask_0, x = q_189_cast_fp16)[name = tensor<string, []>("op_42494_cast_fp16")];
            tensor<int32, [4]> var_42498_begin_0 = const()[name = tensor<string, []>("op_42498_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_42498_end_0 = const()[name = tensor<string, []>("op_42498_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_42498_end_mask_0 = const()[name = tensor<string, []>("op_42498_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42498_cast_fp16 = slice_by_index(begin = var_42498_begin_0, end = var_42498_end_0, end_mask = var_42498_end_mask_0, x = q_189_cast_fp16)[name = tensor<string, []>("op_42498_cast_fp16")];
            tensor<int32, [4]> var_42502_begin_0 = const()[name = tensor<string, []>("op_42502_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_42502_end_0 = const()[name = tensor<string, []>("op_42502_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_42502_end_mask_0 = const()[name = tensor<string, []>("op_42502_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42502_cast_fp16 = slice_by_index(begin = var_42502_begin_0, end = var_42502_end_0, end_mask = var_42502_end_mask_0, x = q_189_cast_fp16)[name = tensor<string, []>("op_42502_cast_fp16")];
            tensor<int32, [4]> var_42506_begin_0 = const()[name = tensor<string, []>("op_42506_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_42506_end_0 = const()[name = tensor<string, []>("op_42506_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_42506_end_mask_0 = const()[name = tensor<string, []>("op_42506_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42506_cast_fp16 = slice_by_index(begin = var_42506_begin_0, end = var_42506_end_0, end_mask = var_42506_end_mask_0, x = q_189_cast_fp16)[name = tensor<string, []>("op_42506_cast_fp16")];
            tensor<int32, [4]> var_42510_begin_0 = const()[name = tensor<string, []>("op_42510_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_42510_end_0 = const()[name = tensor<string, []>("op_42510_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_42510_end_mask_0 = const()[name = tensor<string, []>("op_42510_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42510_cast_fp16 = slice_by_index(begin = var_42510_begin_0, end = var_42510_end_0, end_mask = var_42510_end_mask_0, x = q_189_cast_fp16)[name = tensor<string, []>("op_42510_cast_fp16")];
            tensor<int32, [4]> var_42514_begin_0 = const()[name = tensor<string, []>("op_42514_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_42514_end_0 = const()[name = tensor<string, []>("op_42514_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_42514_end_mask_0 = const()[name = tensor<string, []>("op_42514_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42514_cast_fp16 = slice_by_index(begin = var_42514_begin_0, end = var_42514_end_0, end_mask = var_42514_end_mask_0, x = q_189_cast_fp16)[name = tensor<string, []>("op_42514_cast_fp16")];
            tensor<int32, [4]> var_42518_begin_0 = const()[name = tensor<string, []>("op_42518_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_42518_end_0 = const()[name = tensor<string, []>("op_42518_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_42518_end_mask_0 = const()[name = tensor<string, []>("op_42518_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42518_cast_fp16 = slice_by_index(begin = var_42518_begin_0, end = var_42518_end_0, end_mask = var_42518_end_mask_0, x = q_189_cast_fp16)[name = tensor<string, []>("op_42518_cast_fp16")];
            tensor<int32, [4]> var_42522_begin_0 = const()[name = tensor<string, []>("op_42522_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_42522_end_0 = const()[name = tensor<string, []>("op_42522_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_42522_end_mask_0 = const()[name = tensor<string, []>("op_42522_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42522_cast_fp16 = slice_by_index(begin = var_42522_begin_0, end = var_42522_end_0, end_mask = var_42522_end_mask_0, x = q_189_cast_fp16)[name = tensor<string, []>("op_42522_cast_fp16")];
            tensor<int32, [4]> var_42526_begin_0 = const()[name = tensor<string, []>("op_42526_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_42526_end_0 = const()[name = tensor<string, []>("op_42526_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_42526_end_mask_0 = const()[name = tensor<string, []>("op_42526_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42526_cast_fp16 = slice_by_index(begin = var_42526_begin_0, end = var_42526_end_0, end_mask = var_42526_end_mask_0, x = q_189_cast_fp16)[name = tensor<string, []>("op_42526_cast_fp16")];
            tensor<int32, [4]> k_379_perm_0 = const()[name = tensor<string, []>("k_379_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_42533_begin_0 = const()[name = tensor<string, []>("op_42533_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_42533_end_0 = const()[name = tensor<string, []>("op_42533_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_42533_end_mask_0 = const()[name = tensor<string, []>("op_42533_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> k_379_cast_fp16 = transpose(perm = k_379_perm_0, x = k_377_cast_fp16)[name = tensor<string, []>("transpose_45")];
            tensor<fp16, [2, 1024, 1, 64]> var_42533_cast_fp16 = slice_by_index(begin = var_42533_begin_0, end = var_42533_end_0, end_mask = var_42533_end_mask_0, x = k_379_cast_fp16)[name = tensor<string, []>("op_42533_cast_fp16")];
            tensor<int32, [4]> var_42537_begin_0 = const()[name = tensor<string, []>("op_42537_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_42537_end_0 = const()[name = tensor<string, []>("op_42537_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_42537_end_mask_0 = const()[name = tensor<string, []>("op_42537_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_42537_cast_fp16 = slice_by_index(begin = var_42537_begin_0, end = var_42537_end_0, end_mask = var_42537_end_mask_0, x = k_379_cast_fp16)[name = tensor<string, []>("op_42537_cast_fp16")];
            tensor<int32, [4]> var_42541_begin_0 = const()[name = tensor<string, []>("op_42541_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_42541_end_0 = const()[name = tensor<string, []>("op_42541_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_42541_end_mask_0 = const()[name = tensor<string, []>("op_42541_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_42541_cast_fp16 = slice_by_index(begin = var_42541_begin_0, end = var_42541_end_0, end_mask = var_42541_end_mask_0, x = k_379_cast_fp16)[name = tensor<string, []>("op_42541_cast_fp16")];
            tensor<int32, [4]> var_42545_begin_0 = const()[name = tensor<string, []>("op_42545_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_42545_end_0 = const()[name = tensor<string, []>("op_42545_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_42545_end_mask_0 = const()[name = tensor<string, []>("op_42545_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_42545_cast_fp16 = slice_by_index(begin = var_42545_begin_0, end = var_42545_end_0, end_mask = var_42545_end_mask_0, x = k_379_cast_fp16)[name = tensor<string, []>("op_42545_cast_fp16")];
            tensor<int32, [4]> var_42549_begin_0 = const()[name = tensor<string, []>("op_42549_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_42549_end_0 = const()[name = tensor<string, []>("op_42549_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_42549_end_mask_0 = const()[name = tensor<string, []>("op_42549_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_42549_cast_fp16 = slice_by_index(begin = var_42549_begin_0, end = var_42549_end_0, end_mask = var_42549_end_mask_0, x = k_379_cast_fp16)[name = tensor<string, []>("op_42549_cast_fp16")];
            tensor<int32, [4]> var_42553_begin_0 = const()[name = tensor<string, []>("op_42553_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_42553_end_0 = const()[name = tensor<string, []>("op_42553_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_42553_end_mask_0 = const()[name = tensor<string, []>("op_42553_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_42553_cast_fp16 = slice_by_index(begin = var_42553_begin_0, end = var_42553_end_0, end_mask = var_42553_end_mask_0, x = k_379_cast_fp16)[name = tensor<string, []>("op_42553_cast_fp16")];
            tensor<int32, [4]> var_42557_begin_0 = const()[name = tensor<string, []>("op_42557_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_42557_end_0 = const()[name = tensor<string, []>("op_42557_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_42557_end_mask_0 = const()[name = tensor<string, []>("op_42557_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_42557_cast_fp16 = slice_by_index(begin = var_42557_begin_0, end = var_42557_end_0, end_mask = var_42557_end_mask_0, x = k_379_cast_fp16)[name = tensor<string, []>("op_42557_cast_fp16")];
            tensor<int32, [4]> var_42561_begin_0 = const()[name = tensor<string, []>("op_42561_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_42561_end_0 = const()[name = tensor<string, []>("op_42561_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_42561_end_mask_0 = const()[name = tensor<string, []>("op_42561_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_42561_cast_fp16 = slice_by_index(begin = var_42561_begin_0, end = var_42561_end_0, end_mask = var_42561_end_mask_0, x = k_379_cast_fp16)[name = tensor<string, []>("op_42561_cast_fp16")];
            tensor<int32, [4]> var_42565_begin_0 = const()[name = tensor<string, []>("op_42565_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_42565_end_0 = const()[name = tensor<string, []>("op_42565_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_42565_end_mask_0 = const()[name = tensor<string, []>("op_42565_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_42565_cast_fp16 = slice_by_index(begin = var_42565_begin_0, end = var_42565_end_0, end_mask = var_42565_end_mask_0, x = k_379_cast_fp16)[name = tensor<string, []>("op_42565_cast_fp16")];
            tensor<int32, [4]> var_42569_begin_0 = const()[name = tensor<string, []>("op_42569_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_42569_end_0 = const()[name = tensor<string, []>("op_42569_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_42569_end_mask_0 = const()[name = tensor<string, []>("op_42569_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_42569_cast_fp16 = slice_by_index(begin = var_42569_begin_0, end = var_42569_end_0, end_mask = var_42569_end_mask_0, x = k_379_cast_fp16)[name = tensor<string, []>("op_42569_cast_fp16")];
            tensor<int32, [4]> var_42573_begin_0 = const()[name = tensor<string, []>("op_42573_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_42573_end_0 = const()[name = tensor<string, []>("op_42573_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_42573_end_mask_0 = const()[name = tensor<string, []>("op_42573_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_42573_cast_fp16 = slice_by_index(begin = var_42573_begin_0, end = var_42573_end_0, end_mask = var_42573_end_mask_0, x = k_379_cast_fp16)[name = tensor<string, []>("op_42573_cast_fp16")];
            tensor<int32, [4]> var_42577_begin_0 = const()[name = tensor<string, []>("op_42577_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_42577_end_0 = const()[name = tensor<string, []>("op_42577_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_42577_end_mask_0 = const()[name = tensor<string, []>("op_42577_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_42577_cast_fp16 = slice_by_index(begin = var_42577_begin_0, end = var_42577_end_0, end_mask = var_42577_end_mask_0, x = k_379_cast_fp16)[name = tensor<string, []>("op_42577_cast_fp16")];
            tensor<int32, [4]> var_42581_begin_0 = const()[name = tensor<string, []>("op_42581_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_42581_end_0 = const()[name = tensor<string, []>("op_42581_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_42581_end_mask_0 = const()[name = tensor<string, []>("op_42581_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_42581_cast_fp16 = slice_by_index(begin = var_42581_begin_0, end = var_42581_end_0, end_mask = var_42581_end_mask_0, x = k_379_cast_fp16)[name = tensor<string, []>("op_42581_cast_fp16")];
            tensor<int32, [4]> var_42585_begin_0 = const()[name = tensor<string, []>("op_42585_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_42585_end_0 = const()[name = tensor<string, []>("op_42585_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_42585_end_mask_0 = const()[name = tensor<string, []>("op_42585_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_42585_cast_fp16 = slice_by_index(begin = var_42585_begin_0, end = var_42585_end_0, end_mask = var_42585_end_mask_0, x = k_379_cast_fp16)[name = tensor<string, []>("op_42585_cast_fp16")];
            tensor<int32, [4]> var_42589_begin_0 = const()[name = tensor<string, []>("op_42589_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_42589_end_0 = const()[name = tensor<string, []>("op_42589_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_42589_end_mask_0 = const()[name = tensor<string, []>("op_42589_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_42589_cast_fp16 = slice_by_index(begin = var_42589_begin_0, end = var_42589_end_0, end_mask = var_42589_end_mask_0, x = k_379_cast_fp16)[name = tensor<string, []>("op_42589_cast_fp16")];
            tensor<int32, [4]> var_42593_begin_0 = const()[name = tensor<string, []>("op_42593_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_42593_end_0 = const()[name = tensor<string, []>("op_42593_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_42593_end_mask_0 = const()[name = tensor<string, []>("op_42593_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_42593_cast_fp16 = slice_by_index(begin = var_42593_begin_0, end = var_42593_end_0, end_mask = var_42593_end_mask_0, x = k_379_cast_fp16)[name = tensor<string, []>("op_42593_cast_fp16")];
            tensor<int32, [4]> var_42597_begin_0 = const()[name = tensor<string, []>("op_42597_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_42597_end_0 = const()[name = tensor<string, []>("op_42597_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_42597_end_mask_0 = const()[name = tensor<string, []>("op_42597_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_42597_cast_fp16 = slice_by_index(begin = var_42597_begin_0, end = var_42597_end_0, end_mask = var_42597_end_mask_0, x = k_379_cast_fp16)[name = tensor<string, []>("op_42597_cast_fp16")];
            tensor<int32, [4]> var_42601_begin_0 = const()[name = tensor<string, []>("op_42601_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_42601_end_0 = const()[name = tensor<string, []>("op_42601_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_42601_end_mask_0 = const()[name = tensor<string, []>("op_42601_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_42601_cast_fp16 = slice_by_index(begin = var_42601_begin_0, end = var_42601_end_0, end_mask = var_42601_end_mask_0, x = k_379_cast_fp16)[name = tensor<string, []>("op_42601_cast_fp16")];
            tensor<int32, [4]> var_42605_begin_0 = const()[name = tensor<string, []>("op_42605_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_42605_end_0 = const()[name = tensor<string, []>("op_42605_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_42605_end_mask_0 = const()[name = tensor<string, []>("op_42605_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_42605_cast_fp16 = slice_by_index(begin = var_42605_begin_0, end = var_42605_end_0, end_mask = var_42605_end_mask_0, x = k_379_cast_fp16)[name = tensor<string, []>("op_42605_cast_fp16")];
            tensor<int32, [4]> var_42609_begin_0 = const()[name = tensor<string, []>("op_42609_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_42609_end_0 = const()[name = tensor<string, []>("op_42609_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_42609_end_mask_0 = const()[name = tensor<string, []>("op_42609_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_42609_cast_fp16 = slice_by_index(begin = var_42609_begin_0, end = var_42609_end_0, end_mask = var_42609_end_mask_0, x = k_379_cast_fp16)[name = tensor<string, []>("op_42609_cast_fp16")];
            tensor<int32, [4]> var_42611_begin_0 = const()[name = tensor<string, []>("op_42611_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_42611_end_0 = const()[name = tensor<string, []>("op_42611_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_42611_end_mask_0 = const()[name = tensor<string, []>("op_42611_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42611_cast_fp16 = slice_by_index(begin = var_42611_begin_0, end = var_42611_end_0, end_mask = var_42611_end_mask_0, x = v_189_cast_fp16)[name = tensor<string, []>("op_42611_cast_fp16")];
            tensor<int32, [4]> var_42615_begin_0 = const()[name = tensor<string, []>("op_42615_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_42615_end_0 = const()[name = tensor<string, []>("op_42615_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_42615_end_mask_0 = const()[name = tensor<string, []>("op_42615_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42615_cast_fp16 = slice_by_index(begin = var_42615_begin_0, end = var_42615_end_0, end_mask = var_42615_end_mask_0, x = v_189_cast_fp16)[name = tensor<string, []>("op_42615_cast_fp16")];
            tensor<int32, [4]> var_42619_begin_0 = const()[name = tensor<string, []>("op_42619_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_42619_end_0 = const()[name = tensor<string, []>("op_42619_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_42619_end_mask_0 = const()[name = tensor<string, []>("op_42619_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42619_cast_fp16 = slice_by_index(begin = var_42619_begin_0, end = var_42619_end_0, end_mask = var_42619_end_mask_0, x = v_189_cast_fp16)[name = tensor<string, []>("op_42619_cast_fp16")];
            tensor<int32, [4]> var_42623_begin_0 = const()[name = tensor<string, []>("op_42623_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_42623_end_0 = const()[name = tensor<string, []>("op_42623_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_42623_end_mask_0 = const()[name = tensor<string, []>("op_42623_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42623_cast_fp16 = slice_by_index(begin = var_42623_begin_0, end = var_42623_end_0, end_mask = var_42623_end_mask_0, x = v_189_cast_fp16)[name = tensor<string, []>("op_42623_cast_fp16")];
            tensor<int32, [4]> var_42627_begin_0 = const()[name = tensor<string, []>("op_42627_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_42627_end_0 = const()[name = tensor<string, []>("op_42627_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_42627_end_mask_0 = const()[name = tensor<string, []>("op_42627_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42627_cast_fp16 = slice_by_index(begin = var_42627_begin_0, end = var_42627_end_0, end_mask = var_42627_end_mask_0, x = v_189_cast_fp16)[name = tensor<string, []>("op_42627_cast_fp16")];
            tensor<int32, [4]> var_42631_begin_0 = const()[name = tensor<string, []>("op_42631_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_42631_end_0 = const()[name = tensor<string, []>("op_42631_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_42631_end_mask_0 = const()[name = tensor<string, []>("op_42631_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42631_cast_fp16 = slice_by_index(begin = var_42631_begin_0, end = var_42631_end_0, end_mask = var_42631_end_mask_0, x = v_189_cast_fp16)[name = tensor<string, []>("op_42631_cast_fp16")];
            tensor<int32, [4]> var_42635_begin_0 = const()[name = tensor<string, []>("op_42635_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_42635_end_0 = const()[name = tensor<string, []>("op_42635_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_42635_end_mask_0 = const()[name = tensor<string, []>("op_42635_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42635_cast_fp16 = slice_by_index(begin = var_42635_begin_0, end = var_42635_end_0, end_mask = var_42635_end_mask_0, x = v_189_cast_fp16)[name = tensor<string, []>("op_42635_cast_fp16")];
            tensor<int32, [4]> var_42639_begin_0 = const()[name = tensor<string, []>("op_42639_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_42639_end_0 = const()[name = tensor<string, []>("op_42639_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_42639_end_mask_0 = const()[name = tensor<string, []>("op_42639_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42639_cast_fp16 = slice_by_index(begin = var_42639_begin_0, end = var_42639_end_0, end_mask = var_42639_end_mask_0, x = v_189_cast_fp16)[name = tensor<string, []>("op_42639_cast_fp16")];
            tensor<int32, [4]> var_42643_begin_0 = const()[name = tensor<string, []>("op_42643_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_42643_end_0 = const()[name = tensor<string, []>("op_42643_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_42643_end_mask_0 = const()[name = tensor<string, []>("op_42643_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42643_cast_fp16 = slice_by_index(begin = var_42643_begin_0, end = var_42643_end_0, end_mask = var_42643_end_mask_0, x = v_189_cast_fp16)[name = tensor<string, []>("op_42643_cast_fp16")];
            tensor<int32, [4]> var_42647_begin_0 = const()[name = tensor<string, []>("op_42647_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_42647_end_0 = const()[name = tensor<string, []>("op_42647_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_42647_end_mask_0 = const()[name = tensor<string, []>("op_42647_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42647_cast_fp16 = slice_by_index(begin = var_42647_begin_0, end = var_42647_end_0, end_mask = var_42647_end_mask_0, x = v_189_cast_fp16)[name = tensor<string, []>("op_42647_cast_fp16")];
            tensor<int32, [4]> var_42651_begin_0 = const()[name = tensor<string, []>("op_42651_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_42651_end_0 = const()[name = tensor<string, []>("op_42651_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_42651_end_mask_0 = const()[name = tensor<string, []>("op_42651_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42651_cast_fp16 = slice_by_index(begin = var_42651_begin_0, end = var_42651_end_0, end_mask = var_42651_end_mask_0, x = v_189_cast_fp16)[name = tensor<string, []>("op_42651_cast_fp16")];
            tensor<int32, [4]> var_42655_begin_0 = const()[name = tensor<string, []>("op_42655_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_42655_end_0 = const()[name = tensor<string, []>("op_42655_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_42655_end_mask_0 = const()[name = tensor<string, []>("op_42655_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42655_cast_fp16 = slice_by_index(begin = var_42655_begin_0, end = var_42655_end_0, end_mask = var_42655_end_mask_0, x = v_189_cast_fp16)[name = tensor<string, []>("op_42655_cast_fp16")];
            tensor<int32, [4]> var_42659_begin_0 = const()[name = tensor<string, []>("op_42659_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_42659_end_0 = const()[name = tensor<string, []>("op_42659_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_42659_end_mask_0 = const()[name = tensor<string, []>("op_42659_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42659_cast_fp16 = slice_by_index(begin = var_42659_begin_0, end = var_42659_end_0, end_mask = var_42659_end_mask_0, x = v_189_cast_fp16)[name = tensor<string, []>("op_42659_cast_fp16")];
            tensor<int32, [4]> var_42663_begin_0 = const()[name = tensor<string, []>("op_42663_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_42663_end_0 = const()[name = tensor<string, []>("op_42663_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_42663_end_mask_0 = const()[name = tensor<string, []>("op_42663_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42663_cast_fp16 = slice_by_index(begin = var_42663_begin_0, end = var_42663_end_0, end_mask = var_42663_end_mask_0, x = v_189_cast_fp16)[name = tensor<string, []>("op_42663_cast_fp16")];
            tensor<int32, [4]> var_42667_begin_0 = const()[name = tensor<string, []>("op_42667_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_42667_end_0 = const()[name = tensor<string, []>("op_42667_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_42667_end_mask_0 = const()[name = tensor<string, []>("op_42667_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42667_cast_fp16 = slice_by_index(begin = var_42667_begin_0, end = var_42667_end_0, end_mask = var_42667_end_mask_0, x = v_189_cast_fp16)[name = tensor<string, []>("op_42667_cast_fp16")];
            tensor<int32, [4]> var_42671_begin_0 = const()[name = tensor<string, []>("op_42671_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_42671_end_0 = const()[name = tensor<string, []>("op_42671_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_42671_end_mask_0 = const()[name = tensor<string, []>("op_42671_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42671_cast_fp16 = slice_by_index(begin = var_42671_begin_0, end = var_42671_end_0, end_mask = var_42671_end_mask_0, x = v_189_cast_fp16)[name = tensor<string, []>("op_42671_cast_fp16")];
            tensor<int32, [4]> var_42675_begin_0 = const()[name = tensor<string, []>("op_42675_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_42675_end_0 = const()[name = tensor<string, []>("op_42675_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_42675_end_mask_0 = const()[name = tensor<string, []>("op_42675_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42675_cast_fp16 = slice_by_index(begin = var_42675_begin_0, end = var_42675_end_0, end_mask = var_42675_end_mask_0, x = v_189_cast_fp16)[name = tensor<string, []>("op_42675_cast_fp16")];
            tensor<int32, [4]> var_42679_begin_0 = const()[name = tensor<string, []>("op_42679_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_42679_end_0 = const()[name = tensor<string, []>("op_42679_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_42679_end_mask_0 = const()[name = tensor<string, []>("op_42679_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42679_cast_fp16 = slice_by_index(begin = var_42679_begin_0, end = var_42679_end_0, end_mask = var_42679_end_mask_0, x = v_189_cast_fp16)[name = tensor<string, []>("op_42679_cast_fp16")];
            tensor<int32, [4]> var_42683_begin_0 = const()[name = tensor<string, []>("op_42683_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_42683_end_0 = const()[name = tensor<string, []>("op_42683_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_42683_end_mask_0 = const()[name = tensor<string, []>("op_42683_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42683_cast_fp16 = slice_by_index(begin = var_42683_begin_0, end = var_42683_end_0, end_mask = var_42683_end_mask_0, x = v_189_cast_fp16)[name = tensor<string, []>("op_42683_cast_fp16")];
            tensor<int32, [4]> var_42687_begin_0 = const()[name = tensor<string, []>("op_42687_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_42687_end_0 = const()[name = tensor<string, []>("op_42687_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_42687_end_mask_0 = const()[name = tensor<string, []>("op_42687_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42687_cast_fp16 = slice_by_index(begin = var_42687_begin_0, end = var_42687_end_0, end_mask = var_42687_end_mask_0, x = v_189_cast_fp16)[name = tensor<string, []>("op_42687_cast_fp16")];
            tensor<string, []> var_42691_equation_0 = const()[name = tensor<string, []>("op_42691_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42691_cast_fp16 = einsum(equation = var_42691_equation_0, values = (var_42533_cast_fp16, var_42450_cast_fp16))[name = tensor<string, []>("op_42691_cast_fp16")];
            tensor<fp16, []> var_42692_to_fp16 = const()[name = tensor<string, []>("op_42692_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3601_cast_fp16 = mul(x = var_42691_cast_fp16, y = var_42692_to_fp16)[name = tensor<string, []>("aw_3601_cast_fp16")];
            tensor<string, []> var_42695_equation_0 = const()[name = tensor<string, []>("op_42695_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42695_cast_fp16 = einsum(equation = var_42695_equation_0, values = (var_42537_cast_fp16, var_42454_cast_fp16))[name = tensor<string, []>("op_42695_cast_fp16")];
            tensor<fp16, []> var_42696_to_fp16 = const()[name = tensor<string, []>("op_42696_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3603_cast_fp16 = mul(x = var_42695_cast_fp16, y = var_42696_to_fp16)[name = tensor<string, []>("aw_3603_cast_fp16")];
            tensor<string, []> var_42699_equation_0 = const()[name = tensor<string, []>("op_42699_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42699_cast_fp16 = einsum(equation = var_42699_equation_0, values = (var_42541_cast_fp16, var_42458_cast_fp16))[name = tensor<string, []>("op_42699_cast_fp16")];
            tensor<fp16, []> var_42700_to_fp16 = const()[name = tensor<string, []>("op_42700_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3605_cast_fp16 = mul(x = var_42699_cast_fp16, y = var_42700_to_fp16)[name = tensor<string, []>("aw_3605_cast_fp16")];
            tensor<string, []> var_42703_equation_0 = const()[name = tensor<string, []>("op_42703_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42703_cast_fp16 = einsum(equation = var_42703_equation_0, values = (var_42545_cast_fp16, var_42462_cast_fp16))[name = tensor<string, []>("op_42703_cast_fp16")];
            tensor<fp16, []> var_42704_to_fp16 = const()[name = tensor<string, []>("op_42704_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3607_cast_fp16 = mul(x = var_42703_cast_fp16, y = var_42704_to_fp16)[name = tensor<string, []>("aw_3607_cast_fp16")];
            tensor<string, []> var_42707_equation_0 = const()[name = tensor<string, []>("op_42707_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42707_cast_fp16 = einsum(equation = var_42707_equation_0, values = (var_42549_cast_fp16, var_42466_cast_fp16))[name = tensor<string, []>("op_42707_cast_fp16")];
            tensor<fp16, []> var_42708_to_fp16 = const()[name = tensor<string, []>("op_42708_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3609_cast_fp16 = mul(x = var_42707_cast_fp16, y = var_42708_to_fp16)[name = tensor<string, []>("aw_3609_cast_fp16")];
            tensor<string, []> var_42711_equation_0 = const()[name = tensor<string, []>("op_42711_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42711_cast_fp16 = einsum(equation = var_42711_equation_0, values = (var_42553_cast_fp16, var_42470_cast_fp16))[name = tensor<string, []>("op_42711_cast_fp16")];
            tensor<fp16, []> var_42712_to_fp16 = const()[name = tensor<string, []>("op_42712_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3611_cast_fp16 = mul(x = var_42711_cast_fp16, y = var_42712_to_fp16)[name = tensor<string, []>("aw_3611_cast_fp16")];
            tensor<string, []> var_42715_equation_0 = const()[name = tensor<string, []>("op_42715_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42715_cast_fp16 = einsum(equation = var_42715_equation_0, values = (var_42557_cast_fp16, var_42474_cast_fp16))[name = tensor<string, []>("op_42715_cast_fp16")];
            tensor<fp16, []> var_42716_to_fp16 = const()[name = tensor<string, []>("op_42716_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3613_cast_fp16 = mul(x = var_42715_cast_fp16, y = var_42716_to_fp16)[name = tensor<string, []>("aw_3613_cast_fp16")];
            tensor<string, []> var_42719_equation_0 = const()[name = tensor<string, []>("op_42719_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42719_cast_fp16 = einsum(equation = var_42719_equation_0, values = (var_42561_cast_fp16, var_42478_cast_fp16))[name = tensor<string, []>("op_42719_cast_fp16")];
            tensor<fp16, []> var_42720_to_fp16 = const()[name = tensor<string, []>("op_42720_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3615_cast_fp16 = mul(x = var_42719_cast_fp16, y = var_42720_to_fp16)[name = tensor<string, []>("aw_3615_cast_fp16")];
            tensor<string, []> var_42723_equation_0 = const()[name = tensor<string, []>("op_42723_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42723_cast_fp16 = einsum(equation = var_42723_equation_0, values = (var_42565_cast_fp16, var_42482_cast_fp16))[name = tensor<string, []>("op_42723_cast_fp16")];
            tensor<fp16, []> var_42724_to_fp16 = const()[name = tensor<string, []>("op_42724_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3617_cast_fp16 = mul(x = var_42723_cast_fp16, y = var_42724_to_fp16)[name = tensor<string, []>("aw_3617_cast_fp16")];
            tensor<string, []> var_42727_equation_0 = const()[name = tensor<string, []>("op_42727_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42727_cast_fp16 = einsum(equation = var_42727_equation_0, values = (var_42569_cast_fp16, var_42486_cast_fp16))[name = tensor<string, []>("op_42727_cast_fp16")];
            tensor<fp16, []> var_42728_to_fp16 = const()[name = tensor<string, []>("op_42728_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3619_cast_fp16 = mul(x = var_42727_cast_fp16, y = var_42728_to_fp16)[name = tensor<string, []>("aw_3619_cast_fp16")];
            tensor<string, []> var_42731_equation_0 = const()[name = tensor<string, []>("op_42731_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42731_cast_fp16 = einsum(equation = var_42731_equation_0, values = (var_42573_cast_fp16, var_42490_cast_fp16))[name = tensor<string, []>("op_42731_cast_fp16")];
            tensor<fp16, []> var_42732_to_fp16 = const()[name = tensor<string, []>("op_42732_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3621_cast_fp16 = mul(x = var_42731_cast_fp16, y = var_42732_to_fp16)[name = tensor<string, []>("aw_3621_cast_fp16")];
            tensor<string, []> var_42735_equation_0 = const()[name = tensor<string, []>("op_42735_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42735_cast_fp16 = einsum(equation = var_42735_equation_0, values = (var_42577_cast_fp16, var_42494_cast_fp16))[name = tensor<string, []>("op_42735_cast_fp16")];
            tensor<fp16, []> var_42736_to_fp16 = const()[name = tensor<string, []>("op_42736_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3623_cast_fp16 = mul(x = var_42735_cast_fp16, y = var_42736_to_fp16)[name = tensor<string, []>("aw_3623_cast_fp16")];
            tensor<string, []> var_42739_equation_0 = const()[name = tensor<string, []>("op_42739_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42739_cast_fp16 = einsum(equation = var_42739_equation_0, values = (var_42581_cast_fp16, var_42498_cast_fp16))[name = tensor<string, []>("op_42739_cast_fp16")];
            tensor<fp16, []> var_42740_to_fp16 = const()[name = tensor<string, []>("op_42740_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3625_cast_fp16 = mul(x = var_42739_cast_fp16, y = var_42740_to_fp16)[name = tensor<string, []>("aw_3625_cast_fp16")];
            tensor<string, []> var_42743_equation_0 = const()[name = tensor<string, []>("op_42743_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42743_cast_fp16 = einsum(equation = var_42743_equation_0, values = (var_42585_cast_fp16, var_42502_cast_fp16))[name = tensor<string, []>("op_42743_cast_fp16")];
            tensor<fp16, []> var_42744_to_fp16 = const()[name = tensor<string, []>("op_42744_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3627_cast_fp16 = mul(x = var_42743_cast_fp16, y = var_42744_to_fp16)[name = tensor<string, []>("aw_3627_cast_fp16")];
            tensor<string, []> var_42747_equation_0 = const()[name = tensor<string, []>("op_42747_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42747_cast_fp16 = einsum(equation = var_42747_equation_0, values = (var_42589_cast_fp16, var_42506_cast_fp16))[name = tensor<string, []>("op_42747_cast_fp16")];
            tensor<fp16, []> var_42748_to_fp16 = const()[name = tensor<string, []>("op_42748_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3629_cast_fp16 = mul(x = var_42747_cast_fp16, y = var_42748_to_fp16)[name = tensor<string, []>("aw_3629_cast_fp16")];
            tensor<string, []> var_42751_equation_0 = const()[name = tensor<string, []>("op_42751_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42751_cast_fp16 = einsum(equation = var_42751_equation_0, values = (var_42593_cast_fp16, var_42510_cast_fp16))[name = tensor<string, []>("op_42751_cast_fp16")];
            tensor<fp16, []> var_42752_to_fp16 = const()[name = tensor<string, []>("op_42752_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3631_cast_fp16 = mul(x = var_42751_cast_fp16, y = var_42752_to_fp16)[name = tensor<string, []>("aw_3631_cast_fp16")];
            tensor<string, []> var_42755_equation_0 = const()[name = tensor<string, []>("op_42755_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42755_cast_fp16 = einsum(equation = var_42755_equation_0, values = (var_42597_cast_fp16, var_42514_cast_fp16))[name = tensor<string, []>("op_42755_cast_fp16")];
            tensor<fp16, []> var_42756_to_fp16 = const()[name = tensor<string, []>("op_42756_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3633_cast_fp16 = mul(x = var_42755_cast_fp16, y = var_42756_to_fp16)[name = tensor<string, []>("aw_3633_cast_fp16")];
            tensor<string, []> var_42759_equation_0 = const()[name = tensor<string, []>("op_42759_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42759_cast_fp16 = einsum(equation = var_42759_equation_0, values = (var_42601_cast_fp16, var_42518_cast_fp16))[name = tensor<string, []>("op_42759_cast_fp16")];
            tensor<fp16, []> var_42760_to_fp16 = const()[name = tensor<string, []>("op_42760_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3635_cast_fp16 = mul(x = var_42759_cast_fp16, y = var_42760_to_fp16)[name = tensor<string, []>("aw_3635_cast_fp16")];
            tensor<string, []> var_42763_equation_0 = const()[name = tensor<string, []>("op_42763_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42763_cast_fp16 = einsum(equation = var_42763_equation_0, values = (var_42605_cast_fp16, var_42522_cast_fp16))[name = tensor<string, []>("op_42763_cast_fp16")];
            tensor<fp16, []> var_42764_to_fp16 = const()[name = tensor<string, []>("op_42764_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3637_cast_fp16 = mul(x = var_42763_cast_fp16, y = var_42764_to_fp16)[name = tensor<string, []>("aw_3637_cast_fp16")];
            tensor<string, []> var_42767_equation_0 = const()[name = tensor<string, []>("op_42767_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42767_cast_fp16 = einsum(equation = var_42767_equation_0, values = (var_42609_cast_fp16, var_42526_cast_fp16))[name = tensor<string, []>("op_42767_cast_fp16")];
            tensor<fp16, []> var_42768_to_fp16 = const()[name = tensor<string, []>("op_42768_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3639_cast_fp16 = mul(x = var_42767_cast_fp16, y = var_42768_to_fp16)[name = tensor<string, []>("aw_3639_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42770_cast_fp16 = softmax(axis = var_30355, x = aw_3601_cast_fp16)[name = tensor<string, []>("op_42770_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42771_cast_fp16 = softmax(axis = var_30355, x = aw_3603_cast_fp16)[name = tensor<string, []>("op_42771_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42772_cast_fp16 = softmax(axis = var_30355, x = aw_3605_cast_fp16)[name = tensor<string, []>("op_42772_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42773_cast_fp16 = softmax(axis = var_30355, x = aw_3607_cast_fp16)[name = tensor<string, []>("op_42773_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42774_cast_fp16 = softmax(axis = var_30355, x = aw_3609_cast_fp16)[name = tensor<string, []>("op_42774_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42775_cast_fp16 = softmax(axis = var_30355, x = aw_3611_cast_fp16)[name = tensor<string, []>("op_42775_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42776_cast_fp16 = softmax(axis = var_30355, x = aw_3613_cast_fp16)[name = tensor<string, []>("op_42776_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42777_cast_fp16 = softmax(axis = var_30355, x = aw_3615_cast_fp16)[name = tensor<string, []>("op_42777_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42778_cast_fp16 = softmax(axis = var_30355, x = aw_3617_cast_fp16)[name = tensor<string, []>("op_42778_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42779_cast_fp16 = softmax(axis = var_30355, x = aw_3619_cast_fp16)[name = tensor<string, []>("op_42779_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42780_cast_fp16 = softmax(axis = var_30355, x = aw_3621_cast_fp16)[name = tensor<string, []>("op_42780_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42781_cast_fp16 = softmax(axis = var_30355, x = aw_3623_cast_fp16)[name = tensor<string, []>("op_42781_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42782_cast_fp16 = softmax(axis = var_30355, x = aw_3625_cast_fp16)[name = tensor<string, []>("op_42782_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42783_cast_fp16 = softmax(axis = var_30355, x = aw_3627_cast_fp16)[name = tensor<string, []>("op_42783_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42784_cast_fp16 = softmax(axis = var_30355, x = aw_3629_cast_fp16)[name = tensor<string, []>("op_42784_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42785_cast_fp16 = softmax(axis = var_30355, x = aw_3631_cast_fp16)[name = tensor<string, []>("op_42785_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42786_cast_fp16 = softmax(axis = var_30355, x = aw_3633_cast_fp16)[name = tensor<string, []>("op_42786_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42787_cast_fp16 = softmax(axis = var_30355, x = aw_3635_cast_fp16)[name = tensor<string, []>("op_42787_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42788_cast_fp16 = softmax(axis = var_30355, x = aw_3637_cast_fp16)[name = tensor<string, []>("op_42788_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42789_cast_fp16 = softmax(axis = var_30355, x = aw_3639_cast_fp16)[name = tensor<string, []>("op_42789_cast_fp16")];
            tensor<string, []> var_42791_equation_0 = const()[name = tensor<string, []>("op_42791_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42791_cast_fp16 = einsum(equation = var_42791_equation_0, values = (var_42611_cast_fp16, var_42770_cast_fp16))[name = tensor<string, []>("op_42791_cast_fp16")];
            tensor<string, []> var_42793_equation_0 = const()[name = tensor<string, []>("op_42793_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42793_cast_fp16 = einsum(equation = var_42793_equation_0, values = (var_42615_cast_fp16, var_42771_cast_fp16))[name = tensor<string, []>("op_42793_cast_fp16")];
            tensor<string, []> var_42795_equation_0 = const()[name = tensor<string, []>("op_42795_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42795_cast_fp16 = einsum(equation = var_42795_equation_0, values = (var_42619_cast_fp16, var_42772_cast_fp16))[name = tensor<string, []>("op_42795_cast_fp16")];
            tensor<string, []> var_42797_equation_0 = const()[name = tensor<string, []>("op_42797_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42797_cast_fp16 = einsum(equation = var_42797_equation_0, values = (var_42623_cast_fp16, var_42773_cast_fp16))[name = tensor<string, []>("op_42797_cast_fp16")];
            tensor<string, []> var_42799_equation_0 = const()[name = tensor<string, []>("op_42799_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42799_cast_fp16 = einsum(equation = var_42799_equation_0, values = (var_42627_cast_fp16, var_42774_cast_fp16))[name = tensor<string, []>("op_42799_cast_fp16")];
            tensor<string, []> var_42801_equation_0 = const()[name = tensor<string, []>("op_42801_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42801_cast_fp16 = einsum(equation = var_42801_equation_0, values = (var_42631_cast_fp16, var_42775_cast_fp16))[name = tensor<string, []>("op_42801_cast_fp16")];
            tensor<string, []> var_42803_equation_0 = const()[name = tensor<string, []>("op_42803_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42803_cast_fp16 = einsum(equation = var_42803_equation_0, values = (var_42635_cast_fp16, var_42776_cast_fp16))[name = tensor<string, []>("op_42803_cast_fp16")];
            tensor<string, []> var_42805_equation_0 = const()[name = tensor<string, []>("op_42805_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42805_cast_fp16 = einsum(equation = var_42805_equation_0, values = (var_42639_cast_fp16, var_42777_cast_fp16))[name = tensor<string, []>("op_42805_cast_fp16")];
            tensor<string, []> var_42807_equation_0 = const()[name = tensor<string, []>("op_42807_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42807_cast_fp16 = einsum(equation = var_42807_equation_0, values = (var_42643_cast_fp16, var_42778_cast_fp16))[name = tensor<string, []>("op_42807_cast_fp16")];
            tensor<string, []> var_42809_equation_0 = const()[name = tensor<string, []>("op_42809_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42809_cast_fp16 = einsum(equation = var_42809_equation_0, values = (var_42647_cast_fp16, var_42779_cast_fp16))[name = tensor<string, []>("op_42809_cast_fp16")];
            tensor<string, []> var_42811_equation_0 = const()[name = tensor<string, []>("op_42811_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42811_cast_fp16 = einsum(equation = var_42811_equation_0, values = (var_42651_cast_fp16, var_42780_cast_fp16))[name = tensor<string, []>("op_42811_cast_fp16")];
            tensor<string, []> var_42813_equation_0 = const()[name = tensor<string, []>("op_42813_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42813_cast_fp16 = einsum(equation = var_42813_equation_0, values = (var_42655_cast_fp16, var_42781_cast_fp16))[name = tensor<string, []>("op_42813_cast_fp16")];
            tensor<string, []> var_42815_equation_0 = const()[name = tensor<string, []>("op_42815_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42815_cast_fp16 = einsum(equation = var_42815_equation_0, values = (var_42659_cast_fp16, var_42782_cast_fp16))[name = tensor<string, []>("op_42815_cast_fp16")];
            tensor<string, []> var_42817_equation_0 = const()[name = tensor<string, []>("op_42817_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42817_cast_fp16 = einsum(equation = var_42817_equation_0, values = (var_42663_cast_fp16, var_42783_cast_fp16))[name = tensor<string, []>("op_42817_cast_fp16")];
            tensor<string, []> var_42819_equation_0 = const()[name = tensor<string, []>("op_42819_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42819_cast_fp16 = einsum(equation = var_42819_equation_0, values = (var_42667_cast_fp16, var_42784_cast_fp16))[name = tensor<string, []>("op_42819_cast_fp16")];
            tensor<string, []> var_42821_equation_0 = const()[name = tensor<string, []>("op_42821_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42821_cast_fp16 = einsum(equation = var_42821_equation_0, values = (var_42671_cast_fp16, var_42785_cast_fp16))[name = tensor<string, []>("op_42821_cast_fp16")];
            tensor<string, []> var_42823_equation_0 = const()[name = tensor<string, []>("op_42823_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42823_cast_fp16 = einsum(equation = var_42823_equation_0, values = (var_42675_cast_fp16, var_42786_cast_fp16))[name = tensor<string, []>("op_42823_cast_fp16")];
            tensor<string, []> var_42825_equation_0 = const()[name = tensor<string, []>("op_42825_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42825_cast_fp16 = einsum(equation = var_42825_equation_0, values = (var_42679_cast_fp16, var_42787_cast_fp16))[name = tensor<string, []>("op_42825_cast_fp16")];
            tensor<string, []> var_42827_equation_0 = const()[name = tensor<string, []>("op_42827_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42827_cast_fp16 = einsum(equation = var_42827_equation_0, values = (var_42683_cast_fp16, var_42788_cast_fp16))[name = tensor<string, []>("op_42827_cast_fp16")];
            tensor<string, []> var_42829_equation_0 = const()[name = tensor<string, []>("op_42829_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42829_cast_fp16 = einsum(equation = var_42829_equation_0, values = (var_42687_cast_fp16, var_42789_cast_fp16))[name = tensor<string, []>("op_42829_cast_fp16")];
            tensor<bool, []> input_561_interleave_0 = const()[name = tensor<string, []>("input_561_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_561_cast_fp16 = concat(axis = var_30355, interleave = input_561_interleave_0, values = (var_42791_cast_fp16, var_42793_cast_fp16, var_42795_cast_fp16, var_42797_cast_fp16, var_42799_cast_fp16, var_42801_cast_fp16, var_42803_cast_fp16, var_42805_cast_fp16, var_42807_cast_fp16, var_42809_cast_fp16, var_42811_cast_fp16, var_42813_cast_fp16, var_42815_cast_fp16, var_42817_cast_fp16, var_42819_cast_fp16, var_42821_cast_fp16, var_42823_cast_fp16, var_42825_cast_fp16, var_42827_cast_fp16, var_42829_cast_fp16))[name = tensor<string, []>("input_561_cast_fp16")];
            tensor<string, []> var_42839_pad_type_0 = const()[name = tensor<string, []>("op_42839_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_42839_strides_0 = const()[name = tensor<string, []>("op_42839_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_42839_pad_0 = const()[name = tensor<string, []>("op_42839_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_42839_dilations_0 = const()[name = tensor<string, []>("op_42839_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_42839_groups_0 = const()[name = tensor<string, []>("op_42839_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_3_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(386033984))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(387262848))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_3_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_3_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_3_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(387263040)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_42839_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_3_attn1_to_out_0_bias_to_fp16, dilations = var_42839_dilations_0, groups = var_42839_groups_0, pad = var_42839_pad_0, pad_type = var_42839_pad_type_0, strides = var_42839_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_3_attn1_to_out_0_weight_to_fp16_palettized, x = input_561_cast_fp16)[name = tensor<string, []>("op_42839_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_285_cast_fp16 = add(x = var_42839_cast_fp16, y = inputs_283_cast_fp16)[name = tensor<string, []>("inputs_285_cast_fp16")];
            tensor<int32, [1]> hidden_states_381_axes_0 = const()[name = tensor<string, []>("hidden_states_381_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_381_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_381_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(387265664)))];
            tensor<fp16, [1280]> hidden_states_381_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_381_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(387268288)))];
            tensor<fp16, []> var_42849_to_fp16 = const()[name = tensor<string, []>("op_42849_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_381_cast_fp16 = layer_norm(axes = hidden_states_381_axes_0, beta = hidden_states_381_beta_0_to_fp16, epsilon = var_42849_to_fp16, gamma = hidden_states_381_gamma_0_to_fp16, x = inputs_285_cast_fp16)[name = tensor<string, []>("hidden_states_381_cast_fp16")];
            tensor<string, []> q_191_pad_type_0 = const()[name = tensor<string, []>("q_191_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_191_strides_0 = const()[name = tensor<string, []>("q_191_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_191_pad_0 = const()[name = tensor<string, []>("q_191_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_191_dilations_0 = const()[name = tensor<string, []>("q_191_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_191_groups_0 = const()[name = tensor<string, []>("q_191_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_3_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(387270912))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(388499776))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_3_attn2_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_191_cast_fp16 = conv(dilations = q_191_dilations_0, groups = q_191_groups_0, pad = q_191_pad_0, pad_type = q_191_pad_type_0, strides = q_191_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_3_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_381_cast_fp16)[name = tensor<string, []>("q_191_cast_fp16")];
            tensor<string, []> k_381_pad_type_0 = const()[name = tensor<string, []>("k_381_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_381_strides_0 = const()[name = tensor<string, []>("k_381_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_381_pad_0 = const()[name = tensor<string, []>("k_381_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_381_dilations_0 = const()[name = tensor<string, []>("k_381_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_381_groups_0 = const()[name = tensor<string, []>("k_381_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_3_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(388499968))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(390466112))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_3_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> k_381_cast_fp16 = conv(dilations = k_381_dilations_0, groups = k_381_groups_0, pad = k_381_pad_0, pad_type = k_381_pad_type_0, strides = k_381_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_3_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("k_381_cast_fp16")];
            tensor<string, []> v_191_pad_type_0 = const()[name = tensor<string, []>("v_191_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_191_strides_0 = const()[name = tensor<string, []>("v_191_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_191_pad_0 = const()[name = tensor<string, []>("v_191_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_191_dilations_0 = const()[name = tensor<string, []>("v_191_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_191_groups_0 = const()[name = tensor<string, []>("v_191_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_3_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(390466304))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(392432448))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_3_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> v_191_cast_fp16 = conv(dilations = v_191_dilations_0, groups = v_191_groups_0, pad = v_191_pad_0, pad_type = v_191_pad_type_0, strides = v_191_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_3_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("v_191_cast_fp16")];
            tensor<int32, [4]> var_42882_begin_0 = const()[name = tensor<string, []>("op_42882_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_42882_end_0 = const()[name = tensor<string, []>("op_42882_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_42882_end_mask_0 = const()[name = tensor<string, []>("op_42882_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42882_cast_fp16 = slice_by_index(begin = var_42882_begin_0, end = var_42882_end_0, end_mask = var_42882_end_mask_0, x = q_191_cast_fp16)[name = tensor<string, []>("op_42882_cast_fp16")];
            tensor<int32, [4]> var_42886_begin_0 = const()[name = tensor<string, []>("op_42886_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_42886_end_0 = const()[name = tensor<string, []>("op_42886_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_42886_end_mask_0 = const()[name = tensor<string, []>("op_42886_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42886_cast_fp16 = slice_by_index(begin = var_42886_begin_0, end = var_42886_end_0, end_mask = var_42886_end_mask_0, x = q_191_cast_fp16)[name = tensor<string, []>("op_42886_cast_fp16")];
            tensor<int32, [4]> var_42890_begin_0 = const()[name = tensor<string, []>("op_42890_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_42890_end_0 = const()[name = tensor<string, []>("op_42890_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_42890_end_mask_0 = const()[name = tensor<string, []>("op_42890_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42890_cast_fp16 = slice_by_index(begin = var_42890_begin_0, end = var_42890_end_0, end_mask = var_42890_end_mask_0, x = q_191_cast_fp16)[name = tensor<string, []>("op_42890_cast_fp16")];
            tensor<int32, [4]> var_42894_begin_0 = const()[name = tensor<string, []>("op_42894_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_42894_end_0 = const()[name = tensor<string, []>("op_42894_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_42894_end_mask_0 = const()[name = tensor<string, []>("op_42894_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42894_cast_fp16 = slice_by_index(begin = var_42894_begin_0, end = var_42894_end_0, end_mask = var_42894_end_mask_0, x = q_191_cast_fp16)[name = tensor<string, []>("op_42894_cast_fp16")];
            tensor<int32, [4]> var_42898_begin_0 = const()[name = tensor<string, []>("op_42898_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_42898_end_0 = const()[name = tensor<string, []>("op_42898_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_42898_end_mask_0 = const()[name = tensor<string, []>("op_42898_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42898_cast_fp16 = slice_by_index(begin = var_42898_begin_0, end = var_42898_end_0, end_mask = var_42898_end_mask_0, x = q_191_cast_fp16)[name = tensor<string, []>("op_42898_cast_fp16")];
            tensor<int32, [4]> var_42902_begin_0 = const()[name = tensor<string, []>("op_42902_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_42902_end_0 = const()[name = tensor<string, []>("op_42902_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_42902_end_mask_0 = const()[name = tensor<string, []>("op_42902_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42902_cast_fp16 = slice_by_index(begin = var_42902_begin_0, end = var_42902_end_0, end_mask = var_42902_end_mask_0, x = q_191_cast_fp16)[name = tensor<string, []>("op_42902_cast_fp16")];
            tensor<int32, [4]> var_42906_begin_0 = const()[name = tensor<string, []>("op_42906_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_42906_end_0 = const()[name = tensor<string, []>("op_42906_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_42906_end_mask_0 = const()[name = tensor<string, []>("op_42906_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42906_cast_fp16 = slice_by_index(begin = var_42906_begin_0, end = var_42906_end_0, end_mask = var_42906_end_mask_0, x = q_191_cast_fp16)[name = tensor<string, []>("op_42906_cast_fp16")];
            tensor<int32, [4]> var_42910_begin_0 = const()[name = tensor<string, []>("op_42910_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_42910_end_0 = const()[name = tensor<string, []>("op_42910_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_42910_end_mask_0 = const()[name = tensor<string, []>("op_42910_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42910_cast_fp16 = slice_by_index(begin = var_42910_begin_0, end = var_42910_end_0, end_mask = var_42910_end_mask_0, x = q_191_cast_fp16)[name = tensor<string, []>("op_42910_cast_fp16")];
            tensor<int32, [4]> var_42914_begin_0 = const()[name = tensor<string, []>("op_42914_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_42914_end_0 = const()[name = tensor<string, []>("op_42914_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_42914_end_mask_0 = const()[name = tensor<string, []>("op_42914_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42914_cast_fp16 = slice_by_index(begin = var_42914_begin_0, end = var_42914_end_0, end_mask = var_42914_end_mask_0, x = q_191_cast_fp16)[name = tensor<string, []>("op_42914_cast_fp16")];
            tensor<int32, [4]> var_42918_begin_0 = const()[name = tensor<string, []>("op_42918_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_42918_end_0 = const()[name = tensor<string, []>("op_42918_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_42918_end_mask_0 = const()[name = tensor<string, []>("op_42918_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42918_cast_fp16 = slice_by_index(begin = var_42918_begin_0, end = var_42918_end_0, end_mask = var_42918_end_mask_0, x = q_191_cast_fp16)[name = tensor<string, []>("op_42918_cast_fp16")];
            tensor<int32, [4]> var_42922_begin_0 = const()[name = tensor<string, []>("op_42922_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_42922_end_0 = const()[name = tensor<string, []>("op_42922_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_42922_end_mask_0 = const()[name = tensor<string, []>("op_42922_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42922_cast_fp16 = slice_by_index(begin = var_42922_begin_0, end = var_42922_end_0, end_mask = var_42922_end_mask_0, x = q_191_cast_fp16)[name = tensor<string, []>("op_42922_cast_fp16")];
            tensor<int32, [4]> var_42926_begin_0 = const()[name = tensor<string, []>("op_42926_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_42926_end_0 = const()[name = tensor<string, []>("op_42926_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_42926_end_mask_0 = const()[name = tensor<string, []>("op_42926_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42926_cast_fp16 = slice_by_index(begin = var_42926_begin_0, end = var_42926_end_0, end_mask = var_42926_end_mask_0, x = q_191_cast_fp16)[name = tensor<string, []>("op_42926_cast_fp16")];
            tensor<int32, [4]> var_42930_begin_0 = const()[name = tensor<string, []>("op_42930_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_42930_end_0 = const()[name = tensor<string, []>("op_42930_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_42930_end_mask_0 = const()[name = tensor<string, []>("op_42930_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42930_cast_fp16 = slice_by_index(begin = var_42930_begin_0, end = var_42930_end_0, end_mask = var_42930_end_mask_0, x = q_191_cast_fp16)[name = tensor<string, []>("op_42930_cast_fp16")];
            tensor<int32, [4]> var_42934_begin_0 = const()[name = tensor<string, []>("op_42934_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_42934_end_0 = const()[name = tensor<string, []>("op_42934_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_42934_end_mask_0 = const()[name = tensor<string, []>("op_42934_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42934_cast_fp16 = slice_by_index(begin = var_42934_begin_0, end = var_42934_end_0, end_mask = var_42934_end_mask_0, x = q_191_cast_fp16)[name = tensor<string, []>("op_42934_cast_fp16")];
            tensor<int32, [4]> var_42938_begin_0 = const()[name = tensor<string, []>("op_42938_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_42938_end_0 = const()[name = tensor<string, []>("op_42938_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_42938_end_mask_0 = const()[name = tensor<string, []>("op_42938_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42938_cast_fp16 = slice_by_index(begin = var_42938_begin_0, end = var_42938_end_0, end_mask = var_42938_end_mask_0, x = q_191_cast_fp16)[name = tensor<string, []>("op_42938_cast_fp16")];
            tensor<int32, [4]> var_42942_begin_0 = const()[name = tensor<string, []>("op_42942_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_42942_end_0 = const()[name = tensor<string, []>("op_42942_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_42942_end_mask_0 = const()[name = tensor<string, []>("op_42942_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42942_cast_fp16 = slice_by_index(begin = var_42942_begin_0, end = var_42942_end_0, end_mask = var_42942_end_mask_0, x = q_191_cast_fp16)[name = tensor<string, []>("op_42942_cast_fp16")];
            tensor<int32, [4]> var_42946_begin_0 = const()[name = tensor<string, []>("op_42946_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_42946_end_0 = const()[name = tensor<string, []>("op_42946_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_42946_end_mask_0 = const()[name = tensor<string, []>("op_42946_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42946_cast_fp16 = slice_by_index(begin = var_42946_begin_0, end = var_42946_end_0, end_mask = var_42946_end_mask_0, x = q_191_cast_fp16)[name = tensor<string, []>("op_42946_cast_fp16")];
            tensor<int32, [4]> var_42950_begin_0 = const()[name = tensor<string, []>("op_42950_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_42950_end_0 = const()[name = tensor<string, []>("op_42950_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_42950_end_mask_0 = const()[name = tensor<string, []>("op_42950_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42950_cast_fp16 = slice_by_index(begin = var_42950_begin_0, end = var_42950_end_0, end_mask = var_42950_end_mask_0, x = q_191_cast_fp16)[name = tensor<string, []>("op_42950_cast_fp16")];
            tensor<int32, [4]> var_42954_begin_0 = const()[name = tensor<string, []>("op_42954_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_42954_end_0 = const()[name = tensor<string, []>("op_42954_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_42954_end_mask_0 = const()[name = tensor<string, []>("op_42954_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42954_cast_fp16 = slice_by_index(begin = var_42954_begin_0, end = var_42954_end_0, end_mask = var_42954_end_mask_0, x = q_191_cast_fp16)[name = tensor<string, []>("op_42954_cast_fp16")];
            tensor<int32, [4]> var_42958_begin_0 = const()[name = tensor<string, []>("op_42958_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_42958_end_0 = const()[name = tensor<string, []>("op_42958_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_42958_end_mask_0 = const()[name = tensor<string, []>("op_42958_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42958_cast_fp16 = slice_by_index(begin = var_42958_begin_0, end = var_42958_end_0, end_mask = var_42958_end_mask_0, x = q_191_cast_fp16)[name = tensor<string, []>("op_42958_cast_fp16")];
            tensor<int32, [4]> k_383_perm_0 = const()[name = tensor<string, []>("k_383_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_42965_begin_0 = const()[name = tensor<string, []>("op_42965_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_42965_end_0 = const()[name = tensor<string, []>("op_42965_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_42965_end_mask_0 = const()[name = tensor<string, []>("op_42965_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> k_383_cast_fp16 = transpose(perm = k_383_perm_0, x = k_381_cast_fp16)[name = tensor<string, []>("transpose_44")];
            tensor<fp16, [2, 77, 1, 64]> var_42965_cast_fp16 = slice_by_index(begin = var_42965_begin_0, end = var_42965_end_0, end_mask = var_42965_end_mask_0, x = k_383_cast_fp16)[name = tensor<string, []>("op_42965_cast_fp16")];
            tensor<int32, [4]> var_42969_begin_0 = const()[name = tensor<string, []>("op_42969_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_42969_end_0 = const()[name = tensor<string, []>("op_42969_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_42969_end_mask_0 = const()[name = tensor<string, []>("op_42969_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_42969_cast_fp16 = slice_by_index(begin = var_42969_begin_0, end = var_42969_end_0, end_mask = var_42969_end_mask_0, x = k_383_cast_fp16)[name = tensor<string, []>("op_42969_cast_fp16")];
            tensor<int32, [4]> var_42973_begin_0 = const()[name = tensor<string, []>("op_42973_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_42973_end_0 = const()[name = tensor<string, []>("op_42973_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_42973_end_mask_0 = const()[name = tensor<string, []>("op_42973_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_42973_cast_fp16 = slice_by_index(begin = var_42973_begin_0, end = var_42973_end_0, end_mask = var_42973_end_mask_0, x = k_383_cast_fp16)[name = tensor<string, []>("op_42973_cast_fp16")];
            tensor<int32, [4]> var_42977_begin_0 = const()[name = tensor<string, []>("op_42977_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_42977_end_0 = const()[name = tensor<string, []>("op_42977_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_42977_end_mask_0 = const()[name = tensor<string, []>("op_42977_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_42977_cast_fp16 = slice_by_index(begin = var_42977_begin_0, end = var_42977_end_0, end_mask = var_42977_end_mask_0, x = k_383_cast_fp16)[name = tensor<string, []>("op_42977_cast_fp16")];
            tensor<int32, [4]> var_42981_begin_0 = const()[name = tensor<string, []>("op_42981_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_42981_end_0 = const()[name = tensor<string, []>("op_42981_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_42981_end_mask_0 = const()[name = tensor<string, []>("op_42981_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_42981_cast_fp16 = slice_by_index(begin = var_42981_begin_0, end = var_42981_end_0, end_mask = var_42981_end_mask_0, x = k_383_cast_fp16)[name = tensor<string, []>("op_42981_cast_fp16")];
            tensor<int32, [4]> var_42985_begin_0 = const()[name = tensor<string, []>("op_42985_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_42985_end_0 = const()[name = tensor<string, []>("op_42985_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_42985_end_mask_0 = const()[name = tensor<string, []>("op_42985_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_42985_cast_fp16 = slice_by_index(begin = var_42985_begin_0, end = var_42985_end_0, end_mask = var_42985_end_mask_0, x = k_383_cast_fp16)[name = tensor<string, []>("op_42985_cast_fp16")];
            tensor<int32, [4]> var_42989_begin_0 = const()[name = tensor<string, []>("op_42989_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_42989_end_0 = const()[name = tensor<string, []>("op_42989_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_42989_end_mask_0 = const()[name = tensor<string, []>("op_42989_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_42989_cast_fp16 = slice_by_index(begin = var_42989_begin_0, end = var_42989_end_0, end_mask = var_42989_end_mask_0, x = k_383_cast_fp16)[name = tensor<string, []>("op_42989_cast_fp16")];
            tensor<int32, [4]> var_42993_begin_0 = const()[name = tensor<string, []>("op_42993_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_42993_end_0 = const()[name = tensor<string, []>("op_42993_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_42993_end_mask_0 = const()[name = tensor<string, []>("op_42993_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_42993_cast_fp16 = slice_by_index(begin = var_42993_begin_0, end = var_42993_end_0, end_mask = var_42993_end_mask_0, x = k_383_cast_fp16)[name = tensor<string, []>("op_42993_cast_fp16")];
            tensor<int32, [4]> var_42997_begin_0 = const()[name = tensor<string, []>("op_42997_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_42997_end_0 = const()[name = tensor<string, []>("op_42997_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_42997_end_mask_0 = const()[name = tensor<string, []>("op_42997_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_42997_cast_fp16 = slice_by_index(begin = var_42997_begin_0, end = var_42997_end_0, end_mask = var_42997_end_mask_0, x = k_383_cast_fp16)[name = tensor<string, []>("op_42997_cast_fp16")];
            tensor<int32, [4]> var_43001_begin_0 = const()[name = tensor<string, []>("op_43001_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_43001_end_0 = const()[name = tensor<string, []>("op_43001_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_43001_end_mask_0 = const()[name = tensor<string, []>("op_43001_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43001_cast_fp16 = slice_by_index(begin = var_43001_begin_0, end = var_43001_end_0, end_mask = var_43001_end_mask_0, x = k_383_cast_fp16)[name = tensor<string, []>("op_43001_cast_fp16")];
            tensor<int32, [4]> var_43005_begin_0 = const()[name = tensor<string, []>("op_43005_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_43005_end_0 = const()[name = tensor<string, []>("op_43005_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_43005_end_mask_0 = const()[name = tensor<string, []>("op_43005_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43005_cast_fp16 = slice_by_index(begin = var_43005_begin_0, end = var_43005_end_0, end_mask = var_43005_end_mask_0, x = k_383_cast_fp16)[name = tensor<string, []>("op_43005_cast_fp16")];
            tensor<int32, [4]> var_43009_begin_0 = const()[name = tensor<string, []>("op_43009_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_43009_end_0 = const()[name = tensor<string, []>("op_43009_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_43009_end_mask_0 = const()[name = tensor<string, []>("op_43009_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43009_cast_fp16 = slice_by_index(begin = var_43009_begin_0, end = var_43009_end_0, end_mask = var_43009_end_mask_0, x = k_383_cast_fp16)[name = tensor<string, []>("op_43009_cast_fp16")];
            tensor<int32, [4]> var_43013_begin_0 = const()[name = tensor<string, []>("op_43013_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_43013_end_0 = const()[name = tensor<string, []>("op_43013_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_43013_end_mask_0 = const()[name = tensor<string, []>("op_43013_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43013_cast_fp16 = slice_by_index(begin = var_43013_begin_0, end = var_43013_end_0, end_mask = var_43013_end_mask_0, x = k_383_cast_fp16)[name = tensor<string, []>("op_43013_cast_fp16")];
            tensor<int32, [4]> var_43017_begin_0 = const()[name = tensor<string, []>("op_43017_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_43017_end_0 = const()[name = tensor<string, []>("op_43017_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_43017_end_mask_0 = const()[name = tensor<string, []>("op_43017_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43017_cast_fp16 = slice_by_index(begin = var_43017_begin_0, end = var_43017_end_0, end_mask = var_43017_end_mask_0, x = k_383_cast_fp16)[name = tensor<string, []>("op_43017_cast_fp16")];
            tensor<int32, [4]> var_43021_begin_0 = const()[name = tensor<string, []>("op_43021_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_43021_end_0 = const()[name = tensor<string, []>("op_43021_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_43021_end_mask_0 = const()[name = tensor<string, []>("op_43021_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43021_cast_fp16 = slice_by_index(begin = var_43021_begin_0, end = var_43021_end_0, end_mask = var_43021_end_mask_0, x = k_383_cast_fp16)[name = tensor<string, []>("op_43021_cast_fp16")];
            tensor<int32, [4]> var_43025_begin_0 = const()[name = tensor<string, []>("op_43025_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_43025_end_0 = const()[name = tensor<string, []>("op_43025_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_43025_end_mask_0 = const()[name = tensor<string, []>("op_43025_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43025_cast_fp16 = slice_by_index(begin = var_43025_begin_0, end = var_43025_end_0, end_mask = var_43025_end_mask_0, x = k_383_cast_fp16)[name = tensor<string, []>("op_43025_cast_fp16")];
            tensor<int32, [4]> var_43029_begin_0 = const()[name = tensor<string, []>("op_43029_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_43029_end_0 = const()[name = tensor<string, []>("op_43029_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_43029_end_mask_0 = const()[name = tensor<string, []>("op_43029_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43029_cast_fp16 = slice_by_index(begin = var_43029_begin_0, end = var_43029_end_0, end_mask = var_43029_end_mask_0, x = k_383_cast_fp16)[name = tensor<string, []>("op_43029_cast_fp16")];
            tensor<int32, [4]> var_43033_begin_0 = const()[name = tensor<string, []>("op_43033_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_43033_end_0 = const()[name = tensor<string, []>("op_43033_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_43033_end_mask_0 = const()[name = tensor<string, []>("op_43033_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43033_cast_fp16 = slice_by_index(begin = var_43033_begin_0, end = var_43033_end_0, end_mask = var_43033_end_mask_0, x = k_383_cast_fp16)[name = tensor<string, []>("op_43033_cast_fp16")];
            tensor<int32, [4]> var_43037_begin_0 = const()[name = tensor<string, []>("op_43037_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_43037_end_0 = const()[name = tensor<string, []>("op_43037_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_43037_end_mask_0 = const()[name = tensor<string, []>("op_43037_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43037_cast_fp16 = slice_by_index(begin = var_43037_begin_0, end = var_43037_end_0, end_mask = var_43037_end_mask_0, x = k_383_cast_fp16)[name = tensor<string, []>("op_43037_cast_fp16")];
            tensor<int32, [4]> var_43041_begin_0 = const()[name = tensor<string, []>("op_43041_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_43041_end_0 = const()[name = tensor<string, []>("op_43041_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_43041_end_mask_0 = const()[name = tensor<string, []>("op_43041_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43041_cast_fp16 = slice_by_index(begin = var_43041_begin_0, end = var_43041_end_0, end_mask = var_43041_end_mask_0, x = k_383_cast_fp16)[name = tensor<string, []>("op_43041_cast_fp16")];
            tensor<int32, [4]> var_43043_begin_0 = const()[name = tensor<string, []>("op_43043_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_43043_end_0 = const()[name = tensor<string, []>("op_43043_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_43043_end_mask_0 = const()[name = tensor<string, []>("op_43043_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43043_cast_fp16 = slice_by_index(begin = var_43043_begin_0, end = var_43043_end_0, end_mask = var_43043_end_mask_0, x = v_191_cast_fp16)[name = tensor<string, []>("op_43043_cast_fp16")];
            tensor<int32, [4]> var_43047_begin_0 = const()[name = tensor<string, []>("op_43047_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_43047_end_0 = const()[name = tensor<string, []>("op_43047_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_43047_end_mask_0 = const()[name = tensor<string, []>("op_43047_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43047_cast_fp16 = slice_by_index(begin = var_43047_begin_0, end = var_43047_end_0, end_mask = var_43047_end_mask_0, x = v_191_cast_fp16)[name = tensor<string, []>("op_43047_cast_fp16")];
            tensor<int32, [4]> var_43051_begin_0 = const()[name = tensor<string, []>("op_43051_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_43051_end_0 = const()[name = tensor<string, []>("op_43051_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_43051_end_mask_0 = const()[name = tensor<string, []>("op_43051_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43051_cast_fp16 = slice_by_index(begin = var_43051_begin_0, end = var_43051_end_0, end_mask = var_43051_end_mask_0, x = v_191_cast_fp16)[name = tensor<string, []>("op_43051_cast_fp16")];
            tensor<int32, [4]> var_43055_begin_0 = const()[name = tensor<string, []>("op_43055_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_43055_end_0 = const()[name = tensor<string, []>("op_43055_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_43055_end_mask_0 = const()[name = tensor<string, []>("op_43055_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43055_cast_fp16 = slice_by_index(begin = var_43055_begin_0, end = var_43055_end_0, end_mask = var_43055_end_mask_0, x = v_191_cast_fp16)[name = tensor<string, []>("op_43055_cast_fp16")];
            tensor<int32, [4]> var_43059_begin_0 = const()[name = tensor<string, []>("op_43059_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_43059_end_0 = const()[name = tensor<string, []>("op_43059_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_43059_end_mask_0 = const()[name = tensor<string, []>("op_43059_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43059_cast_fp16 = slice_by_index(begin = var_43059_begin_0, end = var_43059_end_0, end_mask = var_43059_end_mask_0, x = v_191_cast_fp16)[name = tensor<string, []>("op_43059_cast_fp16")];
            tensor<int32, [4]> var_43063_begin_0 = const()[name = tensor<string, []>("op_43063_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_43063_end_0 = const()[name = tensor<string, []>("op_43063_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_43063_end_mask_0 = const()[name = tensor<string, []>("op_43063_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43063_cast_fp16 = slice_by_index(begin = var_43063_begin_0, end = var_43063_end_0, end_mask = var_43063_end_mask_0, x = v_191_cast_fp16)[name = tensor<string, []>("op_43063_cast_fp16")];
            tensor<int32, [4]> var_43067_begin_0 = const()[name = tensor<string, []>("op_43067_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_43067_end_0 = const()[name = tensor<string, []>("op_43067_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_43067_end_mask_0 = const()[name = tensor<string, []>("op_43067_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43067_cast_fp16 = slice_by_index(begin = var_43067_begin_0, end = var_43067_end_0, end_mask = var_43067_end_mask_0, x = v_191_cast_fp16)[name = tensor<string, []>("op_43067_cast_fp16")];
            tensor<int32, [4]> var_43071_begin_0 = const()[name = tensor<string, []>("op_43071_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_43071_end_0 = const()[name = tensor<string, []>("op_43071_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_43071_end_mask_0 = const()[name = tensor<string, []>("op_43071_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43071_cast_fp16 = slice_by_index(begin = var_43071_begin_0, end = var_43071_end_0, end_mask = var_43071_end_mask_0, x = v_191_cast_fp16)[name = tensor<string, []>("op_43071_cast_fp16")];
            tensor<int32, [4]> var_43075_begin_0 = const()[name = tensor<string, []>("op_43075_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_43075_end_0 = const()[name = tensor<string, []>("op_43075_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_43075_end_mask_0 = const()[name = tensor<string, []>("op_43075_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43075_cast_fp16 = slice_by_index(begin = var_43075_begin_0, end = var_43075_end_0, end_mask = var_43075_end_mask_0, x = v_191_cast_fp16)[name = tensor<string, []>("op_43075_cast_fp16")];
            tensor<int32, [4]> var_43079_begin_0 = const()[name = tensor<string, []>("op_43079_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_43079_end_0 = const()[name = tensor<string, []>("op_43079_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_43079_end_mask_0 = const()[name = tensor<string, []>("op_43079_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43079_cast_fp16 = slice_by_index(begin = var_43079_begin_0, end = var_43079_end_0, end_mask = var_43079_end_mask_0, x = v_191_cast_fp16)[name = tensor<string, []>("op_43079_cast_fp16")];
            tensor<int32, [4]> var_43083_begin_0 = const()[name = tensor<string, []>("op_43083_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_43083_end_0 = const()[name = tensor<string, []>("op_43083_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_43083_end_mask_0 = const()[name = tensor<string, []>("op_43083_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43083_cast_fp16 = slice_by_index(begin = var_43083_begin_0, end = var_43083_end_0, end_mask = var_43083_end_mask_0, x = v_191_cast_fp16)[name = tensor<string, []>("op_43083_cast_fp16")];
            tensor<int32, [4]> var_43087_begin_0 = const()[name = tensor<string, []>("op_43087_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_43087_end_0 = const()[name = tensor<string, []>("op_43087_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_43087_end_mask_0 = const()[name = tensor<string, []>("op_43087_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43087_cast_fp16 = slice_by_index(begin = var_43087_begin_0, end = var_43087_end_0, end_mask = var_43087_end_mask_0, x = v_191_cast_fp16)[name = tensor<string, []>("op_43087_cast_fp16")];
            tensor<int32, [4]> var_43091_begin_0 = const()[name = tensor<string, []>("op_43091_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_43091_end_0 = const()[name = tensor<string, []>("op_43091_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_43091_end_mask_0 = const()[name = tensor<string, []>("op_43091_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43091_cast_fp16 = slice_by_index(begin = var_43091_begin_0, end = var_43091_end_0, end_mask = var_43091_end_mask_0, x = v_191_cast_fp16)[name = tensor<string, []>("op_43091_cast_fp16")];
            tensor<int32, [4]> var_43095_begin_0 = const()[name = tensor<string, []>("op_43095_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_43095_end_0 = const()[name = tensor<string, []>("op_43095_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_43095_end_mask_0 = const()[name = tensor<string, []>("op_43095_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43095_cast_fp16 = slice_by_index(begin = var_43095_begin_0, end = var_43095_end_0, end_mask = var_43095_end_mask_0, x = v_191_cast_fp16)[name = tensor<string, []>("op_43095_cast_fp16")];
            tensor<int32, [4]> var_43099_begin_0 = const()[name = tensor<string, []>("op_43099_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_43099_end_0 = const()[name = tensor<string, []>("op_43099_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_43099_end_mask_0 = const()[name = tensor<string, []>("op_43099_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43099_cast_fp16 = slice_by_index(begin = var_43099_begin_0, end = var_43099_end_0, end_mask = var_43099_end_mask_0, x = v_191_cast_fp16)[name = tensor<string, []>("op_43099_cast_fp16")];
            tensor<int32, [4]> var_43103_begin_0 = const()[name = tensor<string, []>("op_43103_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_43103_end_0 = const()[name = tensor<string, []>("op_43103_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_43103_end_mask_0 = const()[name = tensor<string, []>("op_43103_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43103_cast_fp16 = slice_by_index(begin = var_43103_begin_0, end = var_43103_end_0, end_mask = var_43103_end_mask_0, x = v_191_cast_fp16)[name = tensor<string, []>("op_43103_cast_fp16")];
            tensor<int32, [4]> var_43107_begin_0 = const()[name = tensor<string, []>("op_43107_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_43107_end_0 = const()[name = tensor<string, []>("op_43107_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_43107_end_mask_0 = const()[name = tensor<string, []>("op_43107_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43107_cast_fp16 = slice_by_index(begin = var_43107_begin_0, end = var_43107_end_0, end_mask = var_43107_end_mask_0, x = v_191_cast_fp16)[name = tensor<string, []>("op_43107_cast_fp16")];
            tensor<int32, [4]> var_43111_begin_0 = const()[name = tensor<string, []>("op_43111_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_43111_end_0 = const()[name = tensor<string, []>("op_43111_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_43111_end_mask_0 = const()[name = tensor<string, []>("op_43111_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43111_cast_fp16 = slice_by_index(begin = var_43111_begin_0, end = var_43111_end_0, end_mask = var_43111_end_mask_0, x = v_191_cast_fp16)[name = tensor<string, []>("op_43111_cast_fp16")];
            tensor<int32, [4]> var_43115_begin_0 = const()[name = tensor<string, []>("op_43115_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_43115_end_0 = const()[name = tensor<string, []>("op_43115_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_43115_end_mask_0 = const()[name = tensor<string, []>("op_43115_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43115_cast_fp16 = slice_by_index(begin = var_43115_begin_0, end = var_43115_end_0, end_mask = var_43115_end_mask_0, x = v_191_cast_fp16)[name = tensor<string, []>("op_43115_cast_fp16")];
            tensor<int32, [4]> var_43119_begin_0 = const()[name = tensor<string, []>("op_43119_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_43119_end_0 = const()[name = tensor<string, []>("op_43119_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_43119_end_mask_0 = const()[name = tensor<string, []>("op_43119_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43119_cast_fp16 = slice_by_index(begin = var_43119_begin_0, end = var_43119_end_0, end_mask = var_43119_end_mask_0, x = v_191_cast_fp16)[name = tensor<string, []>("op_43119_cast_fp16")];
            tensor<string, []> var_43123_equation_0 = const()[name = tensor<string, []>("op_43123_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_43123_cast_fp16 = einsum(equation = var_43123_equation_0, values = (var_42965_cast_fp16, var_42882_cast_fp16))[name = tensor<string, []>("op_43123_cast_fp16")];
            tensor<fp16, []> var_43124_to_fp16 = const()[name = tensor<string, []>("op_43124_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3641_cast_fp16 = mul(x = var_43123_cast_fp16, y = var_43124_to_fp16)[name = tensor<string, []>("aw_3641_cast_fp16")];
            tensor<string, []> var_43127_equation_0 = const()[name = tensor<string, []>("op_43127_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_43127_cast_fp16 = einsum(equation = var_43127_equation_0, values = (var_42969_cast_fp16, var_42886_cast_fp16))[name = tensor<string, []>("op_43127_cast_fp16")];
            tensor<fp16, []> var_43128_to_fp16 = const()[name = tensor<string, []>("op_43128_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3643_cast_fp16 = mul(x = var_43127_cast_fp16, y = var_43128_to_fp16)[name = tensor<string, []>("aw_3643_cast_fp16")];
            tensor<string, []> var_43131_equation_0 = const()[name = tensor<string, []>("op_43131_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_43131_cast_fp16 = einsum(equation = var_43131_equation_0, values = (var_42973_cast_fp16, var_42890_cast_fp16))[name = tensor<string, []>("op_43131_cast_fp16")];
            tensor<fp16, []> var_43132_to_fp16 = const()[name = tensor<string, []>("op_43132_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3645_cast_fp16 = mul(x = var_43131_cast_fp16, y = var_43132_to_fp16)[name = tensor<string, []>("aw_3645_cast_fp16")];
            tensor<string, []> var_43135_equation_0 = const()[name = tensor<string, []>("op_43135_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_43135_cast_fp16 = einsum(equation = var_43135_equation_0, values = (var_42977_cast_fp16, var_42894_cast_fp16))[name = tensor<string, []>("op_43135_cast_fp16")];
            tensor<fp16, []> var_43136_to_fp16 = const()[name = tensor<string, []>("op_43136_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3647_cast_fp16 = mul(x = var_43135_cast_fp16, y = var_43136_to_fp16)[name = tensor<string, []>("aw_3647_cast_fp16")];
            tensor<string, []> var_43139_equation_0 = const()[name = tensor<string, []>("op_43139_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_43139_cast_fp16 = einsum(equation = var_43139_equation_0, values = (var_42981_cast_fp16, var_42898_cast_fp16))[name = tensor<string, []>("op_43139_cast_fp16")];
            tensor<fp16, []> var_43140_to_fp16 = const()[name = tensor<string, []>("op_43140_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3649_cast_fp16 = mul(x = var_43139_cast_fp16, y = var_43140_to_fp16)[name = tensor<string, []>("aw_3649_cast_fp16")];
            tensor<string, []> var_43143_equation_0 = const()[name = tensor<string, []>("op_43143_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_43143_cast_fp16 = einsum(equation = var_43143_equation_0, values = (var_42985_cast_fp16, var_42902_cast_fp16))[name = tensor<string, []>("op_43143_cast_fp16")];
            tensor<fp16, []> var_43144_to_fp16 = const()[name = tensor<string, []>("op_43144_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3651_cast_fp16 = mul(x = var_43143_cast_fp16, y = var_43144_to_fp16)[name = tensor<string, []>("aw_3651_cast_fp16")];
            tensor<string, []> var_43147_equation_0 = const()[name = tensor<string, []>("op_43147_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_43147_cast_fp16 = einsum(equation = var_43147_equation_0, values = (var_42989_cast_fp16, var_42906_cast_fp16))[name = tensor<string, []>("op_43147_cast_fp16")];
            tensor<fp16, []> var_43148_to_fp16 = const()[name = tensor<string, []>("op_43148_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3653_cast_fp16 = mul(x = var_43147_cast_fp16, y = var_43148_to_fp16)[name = tensor<string, []>("aw_3653_cast_fp16")];
            tensor<string, []> var_43151_equation_0 = const()[name = tensor<string, []>("op_43151_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_43151_cast_fp16 = einsum(equation = var_43151_equation_0, values = (var_42993_cast_fp16, var_42910_cast_fp16))[name = tensor<string, []>("op_43151_cast_fp16")];
            tensor<fp16, []> var_43152_to_fp16 = const()[name = tensor<string, []>("op_43152_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3655_cast_fp16 = mul(x = var_43151_cast_fp16, y = var_43152_to_fp16)[name = tensor<string, []>("aw_3655_cast_fp16")];
            tensor<string, []> var_43155_equation_0 = const()[name = tensor<string, []>("op_43155_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_43155_cast_fp16 = einsum(equation = var_43155_equation_0, values = (var_42997_cast_fp16, var_42914_cast_fp16))[name = tensor<string, []>("op_43155_cast_fp16")];
            tensor<fp16, []> var_43156_to_fp16 = const()[name = tensor<string, []>("op_43156_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3657_cast_fp16 = mul(x = var_43155_cast_fp16, y = var_43156_to_fp16)[name = tensor<string, []>("aw_3657_cast_fp16")];
            tensor<string, []> var_43159_equation_0 = const()[name = tensor<string, []>("op_43159_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_43159_cast_fp16 = einsum(equation = var_43159_equation_0, values = (var_43001_cast_fp16, var_42918_cast_fp16))[name = tensor<string, []>("op_43159_cast_fp16")];
            tensor<fp16, []> var_43160_to_fp16 = const()[name = tensor<string, []>("op_43160_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3659_cast_fp16 = mul(x = var_43159_cast_fp16, y = var_43160_to_fp16)[name = tensor<string, []>("aw_3659_cast_fp16")];
            tensor<string, []> var_43163_equation_0 = const()[name = tensor<string, []>("op_43163_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_43163_cast_fp16 = einsum(equation = var_43163_equation_0, values = (var_43005_cast_fp16, var_42922_cast_fp16))[name = tensor<string, []>("op_43163_cast_fp16")];
            tensor<fp16, []> var_43164_to_fp16 = const()[name = tensor<string, []>("op_43164_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3661_cast_fp16 = mul(x = var_43163_cast_fp16, y = var_43164_to_fp16)[name = tensor<string, []>("aw_3661_cast_fp16")];
            tensor<string, []> var_43167_equation_0 = const()[name = tensor<string, []>("op_43167_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_43167_cast_fp16 = einsum(equation = var_43167_equation_0, values = (var_43009_cast_fp16, var_42926_cast_fp16))[name = tensor<string, []>("op_43167_cast_fp16")];
            tensor<fp16, []> var_43168_to_fp16 = const()[name = tensor<string, []>("op_43168_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3663_cast_fp16 = mul(x = var_43167_cast_fp16, y = var_43168_to_fp16)[name = tensor<string, []>("aw_3663_cast_fp16")];
            tensor<string, []> var_43171_equation_0 = const()[name = tensor<string, []>("op_43171_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_43171_cast_fp16 = einsum(equation = var_43171_equation_0, values = (var_43013_cast_fp16, var_42930_cast_fp16))[name = tensor<string, []>("op_43171_cast_fp16")];
            tensor<fp16, []> var_43172_to_fp16 = const()[name = tensor<string, []>("op_43172_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3665_cast_fp16 = mul(x = var_43171_cast_fp16, y = var_43172_to_fp16)[name = tensor<string, []>("aw_3665_cast_fp16")];
            tensor<string, []> var_43175_equation_0 = const()[name = tensor<string, []>("op_43175_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_43175_cast_fp16 = einsum(equation = var_43175_equation_0, values = (var_43017_cast_fp16, var_42934_cast_fp16))[name = tensor<string, []>("op_43175_cast_fp16")];
            tensor<fp16, []> var_43176_to_fp16 = const()[name = tensor<string, []>("op_43176_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3667_cast_fp16 = mul(x = var_43175_cast_fp16, y = var_43176_to_fp16)[name = tensor<string, []>("aw_3667_cast_fp16")];
            tensor<string, []> var_43179_equation_0 = const()[name = tensor<string, []>("op_43179_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_43179_cast_fp16 = einsum(equation = var_43179_equation_0, values = (var_43021_cast_fp16, var_42938_cast_fp16))[name = tensor<string, []>("op_43179_cast_fp16")];
            tensor<fp16, []> var_43180_to_fp16 = const()[name = tensor<string, []>("op_43180_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3669_cast_fp16 = mul(x = var_43179_cast_fp16, y = var_43180_to_fp16)[name = tensor<string, []>("aw_3669_cast_fp16")];
            tensor<string, []> var_43183_equation_0 = const()[name = tensor<string, []>("op_43183_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_43183_cast_fp16 = einsum(equation = var_43183_equation_0, values = (var_43025_cast_fp16, var_42942_cast_fp16))[name = tensor<string, []>("op_43183_cast_fp16")];
            tensor<fp16, []> var_43184_to_fp16 = const()[name = tensor<string, []>("op_43184_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3671_cast_fp16 = mul(x = var_43183_cast_fp16, y = var_43184_to_fp16)[name = tensor<string, []>("aw_3671_cast_fp16")];
            tensor<string, []> var_43187_equation_0 = const()[name = tensor<string, []>("op_43187_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_43187_cast_fp16 = einsum(equation = var_43187_equation_0, values = (var_43029_cast_fp16, var_42946_cast_fp16))[name = tensor<string, []>("op_43187_cast_fp16")];
            tensor<fp16, []> var_43188_to_fp16 = const()[name = tensor<string, []>("op_43188_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3673_cast_fp16 = mul(x = var_43187_cast_fp16, y = var_43188_to_fp16)[name = tensor<string, []>("aw_3673_cast_fp16")];
            tensor<string, []> var_43191_equation_0 = const()[name = tensor<string, []>("op_43191_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_43191_cast_fp16 = einsum(equation = var_43191_equation_0, values = (var_43033_cast_fp16, var_42950_cast_fp16))[name = tensor<string, []>("op_43191_cast_fp16")];
            tensor<fp16, []> var_43192_to_fp16 = const()[name = tensor<string, []>("op_43192_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3675_cast_fp16 = mul(x = var_43191_cast_fp16, y = var_43192_to_fp16)[name = tensor<string, []>("aw_3675_cast_fp16")];
            tensor<string, []> var_43195_equation_0 = const()[name = tensor<string, []>("op_43195_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_43195_cast_fp16 = einsum(equation = var_43195_equation_0, values = (var_43037_cast_fp16, var_42954_cast_fp16))[name = tensor<string, []>("op_43195_cast_fp16")];
            tensor<fp16, []> var_43196_to_fp16 = const()[name = tensor<string, []>("op_43196_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3677_cast_fp16 = mul(x = var_43195_cast_fp16, y = var_43196_to_fp16)[name = tensor<string, []>("aw_3677_cast_fp16")];
            tensor<string, []> var_43199_equation_0 = const()[name = tensor<string, []>("op_43199_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_43199_cast_fp16 = einsum(equation = var_43199_equation_0, values = (var_43041_cast_fp16, var_42958_cast_fp16))[name = tensor<string, []>("op_43199_cast_fp16")];
            tensor<fp16, []> var_43200_to_fp16 = const()[name = tensor<string, []>("op_43200_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3679_cast_fp16 = mul(x = var_43199_cast_fp16, y = var_43200_to_fp16)[name = tensor<string, []>("aw_3679_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_43202_cast_fp16 = softmax(axis = var_30355, x = aw_3641_cast_fp16)[name = tensor<string, []>("op_43202_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_43203_cast_fp16 = softmax(axis = var_30355, x = aw_3643_cast_fp16)[name = tensor<string, []>("op_43203_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_43204_cast_fp16 = softmax(axis = var_30355, x = aw_3645_cast_fp16)[name = tensor<string, []>("op_43204_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_43205_cast_fp16 = softmax(axis = var_30355, x = aw_3647_cast_fp16)[name = tensor<string, []>("op_43205_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_43206_cast_fp16 = softmax(axis = var_30355, x = aw_3649_cast_fp16)[name = tensor<string, []>("op_43206_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_43207_cast_fp16 = softmax(axis = var_30355, x = aw_3651_cast_fp16)[name = tensor<string, []>("op_43207_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_43208_cast_fp16 = softmax(axis = var_30355, x = aw_3653_cast_fp16)[name = tensor<string, []>("op_43208_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_43209_cast_fp16 = softmax(axis = var_30355, x = aw_3655_cast_fp16)[name = tensor<string, []>("op_43209_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_43210_cast_fp16 = softmax(axis = var_30355, x = aw_3657_cast_fp16)[name = tensor<string, []>("op_43210_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_43211_cast_fp16 = softmax(axis = var_30355, x = aw_3659_cast_fp16)[name = tensor<string, []>("op_43211_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_43212_cast_fp16 = softmax(axis = var_30355, x = aw_3661_cast_fp16)[name = tensor<string, []>("op_43212_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_43213_cast_fp16 = softmax(axis = var_30355, x = aw_3663_cast_fp16)[name = tensor<string, []>("op_43213_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_43214_cast_fp16 = softmax(axis = var_30355, x = aw_3665_cast_fp16)[name = tensor<string, []>("op_43214_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_43215_cast_fp16 = softmax(axis = var_30355, x = aw_3667_cast_fp16)[name = tensor<string, []>("op_43215_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_43216_cast_fp16 = softmax(axis = var_30355, x = aw_3669_cast_fp16)[name = tensor<string, []>("op_43216_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_43217_cast_fp16 = softmax(axis = var_30355, x = aw_3671_cast_fp16)[name = tensor<string, []>("op_43217_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_43218_cast_fp16 = softmax(axis = var_30355, x = aw_3673_cast_fp16)[name = tensor<string, []>("op_43218_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_43219_cast_fp16 = softmax(axis = var_30355, x = aw_3675_cast_fp16)[name = tensor<string, []>("op_43219_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_43220_cast_fp16 = softmax(axis = var_30355, x = aw_3677_cast_fp16)[name = tensor<string, []>("op_43220_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_43221_cast_fp16 = softmax(axis = var_30355, x = aw_3679_cast_fp16)[name = tensor<string, []>("op_43221_cast_fp16")];
            tensor<string, []> var_43223_equation_0 = const()[name = tensor<string, []>("op_43223_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43223_cast_fp16 = einsum(equation = var_43223_equation_0, values = (var_43043_cast_fp16, var_43202_cast_fp16))[name = tensor<string, []>("op_43223_cast_fp16")];
            tensor<string, []> var_43225_equation_0 = const()[name = tensor<string, []>("op_43225_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43225_cast_fp16 = einsum(equation = var_43225_equation_0, values = (var_43047_cast_fp16, var_43203_cast_fp16))[name = tensor<string, []>("op_43225_cast_fp16")];
            tensor<string, []> var_43227_equation_0 = const()[name = tensor<string, []>("op_43227_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43227_cast_fp16 = einsum(equation = var_43227_equation_0, values = (var_43051_cast_fp16, var_43204_cast_fp16))[name = tensor<string, []>("op_43227_cast_fp16")];
            tensor<string, []> var_43229_equation_0 = const()[name = tensor<string, []>("op_43229_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43229_cast_fp16 = einsum(equation = var_43229_equation_0, values = (var_43055_cast_fp16, var_43205_cast_fp16))[name = tensor<string, []>("op_43229_cast_fp16")];
            tensor<string, []> var_43231_equation_0 = const()[name = tensor<string, []>("op_43231_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43231_cast_fp16 = einsum(equation = var_43231_equation_0, values = (var_43059_cast_fp16, var_43206_cast_fp16))[name = tensor<string, []>("op_43231_cast_fp16")];
            tensor<string, []> var_43233_equation_0 = const()[name = tensor<string, []>("op_43233_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43233_cast_fp16 = einsum(equation = var_43233_equation_0, values = (var_43063_cast_fp16, var_43207_cast_fp16))[name = tensor<string, []>("op_43233_cast_fp16")];
            tensor<string, []> var_43235_equation_0 = const()[name = tensor<string, []>("op_43235_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43235_cast_fp16 = einsum(equation = var_43235_equation_0, values = (var_43067_cast_fp16, var_43208_cast_fp16))[name = tensor<string, []>("op_43235_cast_fp16")];
            tensor<string, []> var_43237_equation_0 = const()[name = tensor<string, []>("op_43237_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43237_cast_fp16 = einsum(equation = var_43237_equation_0, values = (var_43071_cast_fp16, var_43209_cast_fp16))[name = tensor<string, []>("op_43237_cast_fp16")];
            tensor<string, []> var_43239_equation_0 = const()[name = tensor<string, []>("op_43239_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43239_cast_fp16 = einsum(equation = var_43239_equation_0, values = (var_43075_cast_fp16, var_43210_cast_fp16))[name = tensor<string, []>("op_43239_cast_fp16")];
            tensor<string, []> var_43241_equation_0 = const()[name = tensor<string, []>("op_43241_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43241_cast_fp16 = einsum(equation = var_43241_equation_0, values = (var_43079_cast_fp16, var_43211_cast_fp16))[name = tensor<string, []>("op_43241_cast_fp16")];
            tensor<string, []> var_43243_equation_0 = const()[name = tensor<string, []>("op_43243_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43243_cast_fp16 = einsum(equation = var_43243_equation_0, values = (var_43083_cast_fp16, var_43212_cast_fp16))[name = tensor<string, []>("op_43243_cast_fp16")];
            tensor<string, []> var_43245_equation_0 = const()[name = tensor<string, []>("op_43245_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43245_cast_fp16 = einsum(equation = var_43245_equation_0, values = (var_43087_cast_fp16, var_43213_cast_fp16))[name = tensor<string, []>("op_43245_cast_fp16")];
            tensor<string, []> var_43247_equation_0 = const()[name = tensor<string, []>("op_43247_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43247_cast_fp16 = einsum(equation = var_43247_equation_0, values = (var_43091_cast_fp16, var_43214_cast_fp16))[name = tensor<string, []>("op_43247_cast_fp16")];
            tensor<string, []> var_43249_equation_0 = const()[name = tensor<string, []>("op_43249_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43249_cast_fp16 = einsum(equation = var_43249_equation_0, values = (var_43095_cast_fp16, var_43215_cast_fp16))[name = tensor<string, []>("op_43249_cast_fp16")];
            tensor<string, []> var_43251_equation_0 = const()[name = tensor<string, []>("op_43251_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43251_cast_fp16 = einsum(equation = var_43251_equation_0, values = (var_43099_cast_fp16, var_43216_cast_fp16))[name = tensor<string, []>("op_43251_cast_fp16")];
            tensor<string, []> var_43253_equation_0 = const()[name = tensor<string, []>("op_43253_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43253_cast_fp16 = einsum(equation = var_43253_equation_0, values = (var_43103_cast_fp16, var_43217_cast_fp16))[name = tensor<string, []>("op_43253_cast_fp16")];
            tensor<string, []> var_43255_equation_0 = const()[name = tensor<string, []>("op_43255_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43255_cast_fp16 = einsum(equation = var_43255_equation_0, values = (var_43107_cast_fp16, var_43218_cast_fp16))[name = tensor<string, []>("op_43255_cast_fp16")];
            tensor<string, []> var_43257_equation_0 = const()[name = tensor<string, []>("op_43257_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43257_cast_fp16 = einsum(equation = var_43257_equation_0, values = (var_43111_cast_fp16, var_43219_cast_fp16))[name = tensor<string, []>("op_43257_cast_fp16")];
            tensor<string, []> var_43259_equation_0 = const()[name = tensor<string, []>("op_43259_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43259_cast_fp16 = einsum(equation = var_43259_equation_0, values = (var_43115_cast_fp16, var_43220_cast_fp16))[name = tensor<string, []>("op_43259_cast_fp16")];
            tensor<string, []> var_43261_equation_0 = const()[name = tensor<string, []>("op_43261_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43261_cast_fp16 = einsum(equation = var_43261_equation_0, values = (var_43119_cast_fp16, var_43221_cast_fp16))[name = tensor<string, []>("op_43261_cast_fp16")];
            tensor<bool, []> input_563_interleave_0 = const()[name = tensor<string, []>("input_563_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_563_cast_fp16 = concat(axis = var_30355, interleave = input_563_interleave_0, values = (var_43223_cast_fp16, var_43225_cast_fp16, var_43227_cast_fp16, var_43229_cast_fp16, var_43231_cast_fp16, var_43233_cast_fp16, var_43235_cast_fp16, var_43237_cast_fp16, var_43239_cast_fp16, var_43241_cast_fp16, var_43243_cast_fp16, var_43245_cast_fp16, var_43247_cast_fp16, var_43249_cast_fp16, var_43251_cast_fp16, var_43253_cast_fp16, var_43255_cast_fp16, var_43257_cast_fp16, var_43259_cast_fp16, var_43261_cast_fp16))[name = tensor<string, []>("input_563_cast_fp16")];
            tensor<string, []> var_43271_pad_type_0 = const()[name = tensor<string, []>("op_43271_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_43271_strides_0 = const()[name = tensor<string, []>("op_43271_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_43271_pad_0 = const()[name = tensor<string, []>("op_43271_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_43271_dilations_0 = const()[name = tensor<string, []>("op_43271_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_43271_groups_0 = const()[name = tensor<string, []>("op_43271_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_3_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(392432640))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(393661504))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_3_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_3_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_3_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(393661696)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_43271_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_3_attn2_to_out_0_bias_to_fp16, dilations = var_43271_dilations_0, groups = var_43271_groups_0, pad = var_43271_pad_0, pad_type = var_43271_pad_type_0, strides = var_43271_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_3_attn2_to_out_0_weight_to_fp16_palettized, x = input_563_cast_fp16)[name = tensor<string, []>("op_43271_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_287_cast_fp16 = add(x = var_43271_cast_fp16, y = inputs_285_cast_fp16)[name = tensor<string, []>("inputs_287_cast_fp16")];
            tensor<int32, [1]> input_565_axes_0 = const()[name = tensor<string, []>("input_565_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> input_565_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_565_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(393664320)))];
            tensor<fp16, [1280]> input_565_beta_0_to_fp16 = const()[name = tensor<string, []>("input_565_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(393666944)))];
            tensor<fp16, []> var_43281_to_fp16 = const()[name = tensor<string, []>("op_43281_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> input_565_cast_fp16 = layer_norm(axes = input_565_axes_0, beta = input_565_beta_0_to_fp16, epsilon = var_43281_to_fp16, gamma = input_565_gamma_0_to_fp16, x = inputs_287_cast_fp16)[name = tensor<string, []>("input_565_cast_fp16")];
            tensor<string, []> var_43301_pad_type_0 = const()[name = tensor<string, []>("op_43301_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_43301_strides_0 = const()[name = tensor<string, []>("op_43301_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_43301_pad_0 = const()[name = tensor<string, []>("op_43301_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_43301_dilations_0 = const()[name = tensor<string, []>("op_43301_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_43301_groups_0 = const()[name = tensor<string, []>("op_43301_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_3_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [9830400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(393669568))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(403500032))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_3_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([10240, 1280, 1, 1])];
            tensor<fp16, [10240]> up_blocks_0_attentions_1_transformer_blocks_3_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_3_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(403500224)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_43301_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_3_ff_net_0_proj_bias_to_fp16, dilations = var_43301_dilations_0, groups = var_43301_groups_0, pad = var_43301_pad_0, pad_type = var_43301_pad_type_0, strides = var_43301_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_3_ff_net_0_proj_weight_to_fp16_palettized, x = input_565_cast_fp16)[name = tensor<string, []>("op_43301_cast_fp16")];
            tensor<int32, [2]> var_43302_split_sizes_0 = const()[name = tensor<string, []>("op_43302_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_43302_axis_0 = const()[name = tensor<string, []>("op_43302_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_43302_cast_fp16_0, tensor<fp16, [2, 5120, 1, 1024]> var_43302_cast_fp16_1 = split(axis = var_43302_axis_0, split_sizes = var_43302_split_sizes_0, x = var_43301_cast_fp16)[name = tensor<string, []>("op_43302_cast_fp16")];
            tensor<string, []> var_43304_mode_0 = const()[name = tensor<string, []>("op_43304_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_43304_cast_fp16 = gelu(mode = var_43304_mode_0, x = var_43302_cast_fp16_1)[name = tensor<string, []>("op_43304_cast_fp16")];
            tensor<fp16, [2, 5120, 1, 1024]> input_567_cast_fp16 = mul(x = var_43302_cast_fp16_0, y = var_43304_cast_fp16)[name = tensor<string, []>("input_567_cast_fp16")];
            tensor<string, []> var_43312_pad_type_0 = const()[name = tensor<string, []>("op_43312_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_43312_strides_0 = const()[name = tensor<string, []>("op_43312_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_43312_pad_0 = const()[name = tensor<string, []>("op_43312_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_43312_dilations_0 = const()[name = tensor<string, []>("op_43312_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_43312_groups_0 = const()[name = tensor<string, []>("op_43312_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_3_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [4915200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(403520768))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(408436032))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_3_ff_net_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 5120, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_3_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_3_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(408436224)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_43312_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_3_ff_net_2_bias_to_fp16, dilations = var_43312_dilations_0, groups = var_43312_groups_0, pad = var_43312_pad_0, pad_type = var_43312_pad_type_0, strides = var_43312_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_3_ff_net_2_weight_to_fp16_palettized, x = input_567_cast_fp16)[name = tensor<string, []>("op_43312_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_289_cast_fp16 = add(x = var_43312_cast_fp16, y = inputs_287_cast_fp16)[name = tensor<string, []>("inputs_289_cast_fp16")];
            tensor<int32, [1]> hidden_states_385_axes_0 = const()[name = tensor<string, []>("hidden_states_385_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_385_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_385_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(408438848)))];
            tensor<fp16, [1280]> hidden_states_385_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_385_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(408441472)))];
            tensor<fp16, []> var_43328_to_fp16 = const()[name = tensor<string, []>("op_43328_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_385_cast_fp16 = layer_norm(axes = hidden_states_385_axes_0, beta = hidden_states_385_beta_0_to_fp16, epsilon = var_43328_to_fp16, gamma = hidden_states_385_gamma_0_to_fp16, x = inputs_289_cast_fp16)[name = tensor<string, []>("hidden_states_385_cast_fp16")];
            tensor<string, []> q_193_pad_type_0 = const()[name = tensor<string, []>("q_193_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_193_strides_0 = const()[name = tensor<string, []>("q_193_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_193_pad_0 = const()[name = tensor<string, []>("q_193_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_193_dilations_0 = const()[name = tensor<string, []>("q_193_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_193_groups_0 = const()[name = tensor<string, []>("q_193_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_4_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(408444096))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(409672960))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_4_attn1_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_193_cast_fp16 = conv(dilations = q_193_dilations_0, groups = q_193_groups_0, pad = q_193_pad_0, pad_type = q_193_pad_type_0, strides = q_193_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_4_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_385_cast_fp16)[name = tensor<string, []>("q_193_cast_fp16")];
            tensor<string, []> k_385_pad_type_0 = const()[name = tensor<string, []>("k_385_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_385_strides_0 = const()[name = tensor<string, []>("k_385_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_385_pad_0 = const()[name = tensor<string, []>("k_385_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_385_dilations_0 = const()[name = tensor<string, []>("k_385_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_385_groups_0 = const()[name = tensor<string, []>("k_385_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_4_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(409673152))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(410902016))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_4_attn1_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> k_385_cast_fp16 = conv(dilations = k_385_dilations_0, groups = k_385_groups_0, pad = k_385_pad_0, pad_type = k_385_pad_type_0, strides = k_385_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_4_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_385_cast_fp16)[name = tensor<string, []>("k_385_cast_fp16")];
            tensor<string, []> v_193_pad_type_0 = const()[name = tensor<string, []>("v_193_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_193_strides_0 = const()[name = tensor<string, []>("v_193_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_193_pad_0 = const()[name = tensor<string, []>("v_193_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_193_dilations_0 = const()[name = tensor<string, []>("v_193_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_193_groups_0 = const()[name = tensor<string, []>("v_193_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_4_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(410902208))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(412131072))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_4_attn1_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> v_193_cast_fp16 = conv(dilations = v_193_dilations_0, groups = v_193_groups_0, pad = v_193_pad_0, pad_type = v_193_pad_type_0, strides = v_193_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_4_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_385_cast_fp16)[name = tensor<string, []>("v_193_cast_fp16")];
            tensor<int32, [4]> var_43361_begin_0 = const()[name = tensor<string, []>("op_43361_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_43361_end_0 = const()[name = tensor<string, []>("op_43361_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_43361_end_mask_0 = const()[name = tensor<string, []>("op_43361_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43361_cast_fp16 = slice_by_index(begin = var_43361_begin_0, end = var_43361_end_0, end_mask = var_43361_end_mask_0, x = q_193_cast_fp16)[name = tensor<string, []>("op_43361_cast_fp16")];
            tensor<int32, [4]> var_43365_begin_0 = const()[name = tensor<string, []>("op_43365_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_43365_end_0 = const()[name = tensor<string, []>("op_43365_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_43365_end_mask_0 = const()[name = tensor<string, []>("op_43365_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43365_cast_fp16 = slice_by_index(begin = var_43365_begin_0, end = var_43365_end_0, end_mask = var_43365_end_mask_0, x = q_193_cast_fp16)[name = tensor<string, []>("op_43365_cast_fp16")];
            tensor<int32, [4]> var_43369_begin_0 = const()[name = tensor<string, []>("op_43369_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_43369_end_0 = const()[name = tensor<string, []>("op_43369_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_43369_end_mask_0 = const()[name = tensor<string, []>("op_43369_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43369_cast_fp16 = slice_by_index(begin = var_43369_begin_0, end = var_43369_end_0, end_mask = var_43369_end_mask_0, x = q_193_cast_fp16)[name = tensor<string, []>("op_43369_cast_fp16")];
            tensor<int32, [4]> var_43373_begin_0 = const()[name = tensor<string, []>("op_43373_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_43373_end_0 = const()[name = tensor<string, []>("op_43373_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_43373_end_mask_0 = const()[name = tensor<string, []>("op_43373_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43373_cast_fp16 = slice_by_index(begin = var_43373_begin_0, end = var_43373_end_0, end_mask = var_43373_end_mask_0, x = q_193_cast_fp16)[name = tensor<string, []>("op_43373_cast_fp16")];
            tensor<int32, [4]> var_43377_begin_0 = const()[name = tensor<string, []>("op_43377_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_43377_end_0 = const()[name = tensor<string, []>("op_43377_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_43377_end_mask_0 = const()[name = tensor<string, []>("op_43377_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43377_cast_fp16 = slice_by_index(begin = var_43377_begin_0, end = var_43377_end_0, end_mask = var_43377_end_mask_0, x = q_193_cast_fp16)[name = tensor<string, []>("op_43377_cast_fp16")];
            tensor<int32, [4]> var_43381_begin_0 = const()[name = tensor<string, []>("op_43381_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_43381_end_0 = const()[name = tensor<string, []>("op_43381_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_43381_end_mask_0 = const()[name = tensor<string, []>("op_43381_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43381_cast_fp16 = slice_by_index(begin = var_43381_begin_0, end = var_43381_end_0, end_mask = var_43381_end_mask_0, x = q_193_cast_fp16)[name = tensor<string, []>("op_43381_cast_fp16")];
            tensor<int32, [4]> var_43385_begin_0 = const()[name = tensor<string, []>("op_43385_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_43385_end_0 = const()[name = tensor<string, []>("op_43385_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_43385_end_mask_0 = const()[name = tensor<string, []>("op_43385_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43385_cast_fp16 = slice_by_index(begin = var_43385_begin_0, end = var_43385_end_0, end_mask = var_43385_end_mask_0, x = q_193_cast_fp16)[name = tensor<string, []>("op_43385_cast_fp16")];
            tensor<int32, [4]> var_43389_begin_0 = const()[name = tensor<string, []>("op_43389_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_43389_end_0 = const()[name = tensor<string, []>("op_43389_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_43389_end_mask_0 = const()[name = tensor<string, []>("op_43389_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43389_cast_fp16 = slice_by_index(begin = var_43389_begin_0, end = var_43389_end_0, end_mask = var_43389_end_mask_0, x = q_193_cast_fp16)[name = tensor<string, []>("op_43389_cast_fp16")];
            tensor<int32, [4]> var_43393_begin_0 = const()[name = tensor<string, []>("op_43393_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_43393_end_0 = const()[name = tensor<string, []>("op_43393_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_43393_end_mask_0 = const()[name = tensor<string, []>("op_43393_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43393_cast_fp16 = slice_by_index(begin = var_43393_begin_0, end = var_43393_end_0, end_mask = var_43393_end_mask_0, x = q_193_cast_fp16)[name = tensor<string, []>("op_43393_cast_fp16")];
            tensor<int32, [4]> var_43397_begin_0 = const()[name = tensor<string, []>("op_43397_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_43397_end_0 = const()[name = tensor<string, []>("op_43397_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_43397_end_mask_0 = const()[name = tensor<string, []>("op_43397_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43397_cast_fp16 = slice_by_index(begin = var_43397_begin_0, end = var_43397_end_0, end_mask = var_43397_end_mask_0, x = q_193_cast_fp16)[name = tensor<string, []>("op_43397_cast_fp16")];
            tensor<int32, [4]> var_43401_begin_0 = const()[name = tensor<string, []>("op_43401_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_43401_end_0 = const()[name = tensor<string, []>("op_43401_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_43401_end_mask_0 = const()[name = tensor<string, []>("op_43401_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43401_cast_fp16 = slice_by_index(begin = var_43401_begin_0, end = var_43401_end_0, end_mask = var_43401_end_mask_0, x = q_193_cast_fp16)[name = tensor<string, []>("op_43401_cast_fp16")];
            tensor<int32, [4]> var_43405_begin_0 = const()[name = tensor<string, []>("op_43405_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_43405_end_0 = const()[name = tensor<string, []>("op_43405_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_43405_end_mask_0 = const()[name = tensor<string, []>("op_43405_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43405_cast_fp16 = slice_by_index(begin = var_43405_begin_0, end = var_43405_end_0, end_mask = var_43405_end_mask_0, x = q_193_cast_fp16)[name = tensor<string, []>("op_43405_cast_fp16")];
            tensor<int32, [4]> var_43409_begin_0 = const()[name = tensor<string, []>("op_43409_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_43409_end_0 = const()[name = tensor<string, []>("op_43409_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_43409_end_mask_0 = const()[name = tensor<string, []>("op_43409_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43409_cast_fp16 = slice_by_index(begin = var_43409_begin_0, end = var_43409_end_0, end_mask = var_43409_end_mask_0, x = q_193_cast_fp16)[name = tensor<string, []>("op_43409_cast_fp16")];
            tensor<int32, [4]> var_43413_begin_0 = const()[name = tensor<string, []>("op_43413_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_43413_end_0 = const()[name = tensor<string, []>("op_43413_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_43413_end_mask_0 = const()[name = tensor<string, []>("op_43413_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43413_cast_fp16 = slice_by_index(begin = var_43413_begin_0, end = var_43413_end_0, end_mask = var_43413_end_mask_0, x = q_193_cast_fp16)[name = tensor<string, []>("op_43413_cast_fp16")];
            tensor<int32, [4]> var_43417_begin_0 = const()[name = tensor<string, []>("op_43417_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_43417_end_0 = const()[name = tensor<string, []>("op_43417_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_43417_end_mask_0 = const()[name = tensor<string, []>("op_43417_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43417_cast_fp16 = slice_by_index(begin = var_43417_begin_0, end = var_43417_end_0, end_mask = var_43417_end_mask_0, x = q_193_cast_fp16)[name = tensor<string, []>("op_43417_cast_fp16")];
            tensor<int32, [4]> var_43421_begin_0 = const()[name = tensor<string, []>("op_43421_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_43421_end_0 = const()[name = tensor<string, []>("op_43421_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_43421_end_mask_0 = const()[name = tensor<string, []>("op_43421_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43421_cast_fp16 = slice_by_index(begin = var_43421_begin_0, end = var_43421_end_0, end_mask = var_43421_end_mask_0, x = q_193_cast_fp16)[name = tensor<string, []>("op_43421_cast_fp16")];
            tensor<int32, [4]> var_43425_begin_0 = const()[name = tensor<string, []>("op_43425_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_43425_end_0 = const()[name = tensor<string, []>("op_43425_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_43425_end_mask_0 = const()[name = tensor<string, []>("op_43425_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43425_cast_fp16 = slice_by_index(begin = var_43425_begin_0, end = var_43425_end_0, end_mask = var_43425_end_mask_0, x = q_193_cast_fp16)[name = tensor<string, []>("op_43425_cast_fp16")];
            tensor<int32, [4]> var_43429_begin_0 = const()[name = tensor<string, []>("op_43429_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_43429_end_0 = const()[name = tensor<string, []>("op_43429_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_43429_end_mask_0 = const()[name = tensor<string, []>("op_43429_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43429_cast_fp16 = slice_by_index(begin = var_43429_begin_0, end = var_43429_end_0, end_mask = var_43429_end_mask_0, x = q_193_cast_fp16)[name = tensor<string, []>("op_43429_cast_fp16")];
            tensor<int32, [4]> var_43433_begin_0 = const()[name = tensor<string, []>("op_43433_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_43433_end_0 = const()[name = tensor<string, []>("op_43433_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_43433_end_mask_0 = const()[name = tensor<string, []>("op_43433_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43433_cast_fp16 = slice_by_index(begin = var_43433_begin_0, end = var_43433_end_0, end_mask = var_43433_end_mask_0, x = q_193_cast_fp16)[name = tensor<string, []>("op_43433_cast_fp16")];
            tensor<int32, [4]> var_43437_begin_0 = const()[name = tensor<string, []>("op_43437_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_43437_end_0 = const()[name = tensor<string, []>("op_43437_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_43437_end_mask_0 = const()[name = tensor<string, []>("op_43437_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43437_cast_fp16 = slice_by_index(begin = var_43437_begin_0, end = var_43437_end_0, end_mask = var_43437_end_mask_0, x = q_193_cast_fp16)[name = tensor<string, []>("op_43437_cast_fp16")];
            tensor<int32, [4]> k_387_perm_0 = const()[name = tensor<string, []>("k_387_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_43444_begin_0 = const()[name = tensor<string, []>("op_43444_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_43444_end_0 = const()[name = tensor<string, []>("op_43444_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_43444_end_mask_0 = const()[name = tensor<string, []>("op_43444_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> k_387_cast_fp16 = transpose(perm = k_387_perm_0, x = k_385_cast_fp16)[name = tensor<string, []>("transpose_43")];
            tensor<fp16, [2, 1024, 1, 64]> var_43444_cast_fp16 = slice_by_index(begin = var_43444_begin_0, end = var_43444_end_0, end_mask = var_43444_end_mask_0, x = k_387_cast_fp16)[name = tensor<string, []>("op_43444_cast_fp16")];
            tensor<int32, [4]> var_43448_begin_0 = const()[name = tensor<string, []>("op_43448_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_43448_end_0 = const()[name = tensor<string, []>("op_43448_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_43448_end_mask_0 = const()[name = tensor<string, []>("op_43448_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_43448_cast_fp16 = slice_by_index(begin = var_43448_begin_0, end = var_43448_end_0, end_mask = var_43448_end_mask_0, x = k_387_cast_fp16)[name = tensor<string, []>("op_43448_cast_fp16")];
            tensor<int32, [4]> var_43452_begin_0 = const()[name = tensor<string, []>("op_43452_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_43452_end_0 = const()[name = tensor<string, []>("op_43452_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_43452_end_mask_0 = const()[name = tensor<string, []>("op_43452_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_43452_cast_fp16 = slice_by_index(begin = var_43452_begin_0, end = var_43452_end_0, end_mask = var_43452_end_mask_0, x = k_387_cast_fp16)[name = tensor<string, []>("op_43452_cast_fp16")];
            tensor<int32, [4]> var_43456_begin_0 = const()[name = tensor<string, []>("op_43456_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_43456_end_0 = const()[name = tensor<string, []>("op_43456_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_43456_end_mask_0 = const()[name = tensor<string, []>("op_43456_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_43456_cast_fp16 = slice_by_index(begin = var_43456_begin_0, end = var_43456_end_0, end_mask = var_43456_end_mask_0, x = k_387_cast_fp16)[name = tensor<string, []>("op_43456_cast_fp16")];
            tensor<int32, [4]> var_43460_begin_0 = const()[name = tensor<string, []>("op_43460_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_43460_end_0 = const()[name = tensor<string, []>("op_43460_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_43460_end_mask_0 = const()[name = tensor<string, []>("op_43460_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_43460_cast_fp16 = slice_by_index(begin = var_43460_begin_0, end = var_43460_end_0, end_mask = var_43460_end_mask_0, x = k_387_cast_fp16)[name = tensor<string, []>("op_43460_cast_fp16")];
            tensor<int32, [4]> var_43464_begin_0 = const()[name = tensor<string, []>("op_43464_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_43464_end_0 = const()[name = tensor<string, []>("op_43464_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_43464_end_mask_0 = const()[name = tensor<string, []>("op_43464_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_43464_cast_fp16 = slice_by_index(begin = var_43464_begin_0, end = var_43464_end_0, end_mask = var_43464_end_mask_0, x = k_387_cast_fp16)[name = tensor<string, []>("op_43464_cast_fp16")];
            tensor<int32, [4]> var_43468_begin_0 = const()[name = tensor<string, []>("op_43468_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_43468_end_0 = const()[name = tensor<string, []>("op_43468_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_43468_end_mask_0 = const()[name = tensor<string, []>("op_43468_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_43468_cast_fp16 = slice_by_index(begin = var_43468_begin_0, end = var_43468_end_0, end_mask = var_43468_end_mask_0, x = k_387_cast_fp16)[name = tensor<string, []>("op_43468_cast_fp16")];
            tensor<int32, [4]> var_43472_begin_0 = const()[name = tensor<string, []>("op_43472_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_43472_end_0 = const()[name = tensor<string, []>("op_43472_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_43472_end_mask_0 = const()[name = tensor<string, []>("op_43472_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_43472_cast_fp16 = slice_by_index(begin = var_43472_begin_0, end = var_43472_end_0, end_mask = var_43472_end_mask_0, x = k_387_cast_fp16)[name = tensor<string, []>("op_43472_cast_fp16")];
            tensor<int32, [4]> var_43476_begin_0 = const()[name = tensor<string, []>("op_43476_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_43476_end_0 = const()[name = tensor<string, []>("op_43476_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_43476_end_mask_0 = const()[name = tensor<string, []>("op_43476_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_43476_cast_fp16 = slice_by_index(begin = var_43476_begin_0, end = var_43476_end_0, end_mask = var_43476_end_mask_0, x = k_387_cast_fp16)[name = tensor<string, []>("op_43476_cast_fp16")];
            tensor<int32, [4]> var_43480_begin_0 = const()[name = tensor<string, []>("op_43480_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_43480_end_0 = const()[name = tensor<string, []>("op_43480_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_43480_end_mask_0 = const()[name = tensor<string, []>("op_43480_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_43480_cast_fp16 = slice_by_index(begin = var_43480_begin_0, end = var_43480_end_0, end_mask = var_43480_end_mask_0, x = k_387_cast_fp16)[name = tensor<string, []>("op_43480_cast_fp16")];
            tensor<int32, [4]> var_43484_begin_0 = const()[name = tensor<string, []>("op_43484_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_43484_end_0 = const()[name = tensor<string, []>("op_43484_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_43484_end_mask_0 = const()[name = tensor<string, []>("op_43484_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_43484_cast_fp16 = slice_by_index(begin = var_43484_begin_0, end = var_43484_end_0, end_mask = var_43484_end_mask_0, x = k_387_cast_fp16)[name = tensor<string, []>("op_43484_cast_fp16")];
            tensor<int32, [4]> var_43488_begin_0 = const()[name = tensor<string, []>("op_43488_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_43488_end_0 = const()[name = tensor<string, []>("op_43488_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_43488_end_mask_0 = const()[name = tensor<string, []>("op_43488_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_43488_cast_fp16 = slice_by_index(begin = var_43488_begin_0, end = var_43488_end_0, end_mask = var_43488_end_mask_0, x = k_387_cast_fp16)[name = tensor<string, []>("op_43488_cast_fp16")];
            tensor<int32, [4]> var_43492_begin_0 = const()[name = tensor<string, []>("op_43492_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_43492_end_0 = const()[name = tensor<string, []>("op_43492_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_43492_end_mask_0 = const()[name = tensor<string, []>("op_43492_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_43492_cast_fp16 = slice_by_index(begin = var_43492_begin_0, end = var_43492_end_0, end_mask = var_43492_end_mask_0, x = k_387_cast_fp16)[name = tensor<string, []>("op_43492_cast_fp16")];
            tensor<int32, [4]> var_43496_begin_0 = const()[name = tensor<string, []>("op_43496_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_43496_end_0 = const()[name = tensor<string, []>("op_43496_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_43496_end_mask_0 = const()[name = tensor<string, []>("op_43496_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_43496_cast_fp16 = slice_by_index(begin = var_43496_begin_0, end = var_43496_end_0, end_mask = var_43496_end_mask_0, x = k_387_cast_fp16)[name = tensor<string, []>("op_43496_cast_fp16")];
            tensor<int32, [4]> var_43500_begin_0 = const()[name = tensor<string, []>("op_43500_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_43500_end_0 = const()[name = tensor<string, []>("op_43500_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_43500_end_mask_0 = const()[name = tensor<string, []>("op_43500_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_43500_cast_fp16 = slice_by_index(begin = var_43500_begin_0, end = var_43500_end_0, end_mask = var_43500_end_mask_0, x = k_387_cast_fp16)[name = tensor<string, []>("op_43500_cast_fp16")];
            tensor<int32, [4]> var_43504_begin_0 = const()[name = tensor<string, []>("op_43504_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_43504_end_0 = const()[name = tensor<string, []>("op_43504_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_43504_end_mask_0 = const()[name = tensor<string, []>("op_43504_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_43504_cast_fp16 = slice_by_index(begin = var_43504_begin_0, end = var_43504_end_0, end_mask = var_43504_end_mask_0, x = k_387_cast_fp16)[name = tensor<string, []>("op_43504_cast_fp16")];
            tensor<int32, [4]> var_43508_begin_0 = const()[name = tensor<string, []>("op_43508_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_43508_end_0 = const()[name = tensor<string, []>("op_43508_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_43508_end_mask_0 = const()[name = tensor<string, []>("op_43508_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_43508_cast_fp16 = slice_by_index(begin = var_43508_begin_0, end = var_43508_end_0, end_mask = var_43508_end_mask_0, x = k_387_cast_fp16)[name = tensor<string, []>("op_43508_cast_fp16")];
            tensor<int32, [4]> var_43512_begin_0 = const()[name = tensor<string, []>("op_43512_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_43512_end_0 = const()[name = tensor<string, []>("op_43512_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_43512_end_mask_0 = const()[name = tensor<string, []>("op_43512_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_43512_cast_fp16 = slice_by_index(begin = var_43512_begin_0, end = var_43512_end_0, end_mask = var_43512_end_mask_0, x = k_387_cast_fp16)[name = tensor<string, []>("op_43512_cast_fp16")];
            tensor<int32, [4]> var_43516_begin_0 = const()[name = tensor<string, []>("op_43516_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_43516_end_0 = const()[name = tensor<string, []>("op_43516_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_43516_end_mask_0 = const()[name = tensor<string, []>("op_43516_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_43516_cast_fp16 = slice_by_index(begin = var_43516_begin_0, end = var_43516_end_0, end_mask = var_43516_end_mask_0, x = k_387_cast_fp16)[name = tensor<string, []>("op_43516_cast_fp16")];
            tensor<int32, [4]> var_43520_begin_0 = const()[name = tensor<string, []>("op_43520_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_43520_end_0 = const()[name = tensor<string, []>("op_43520_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_43520_end_mask_0 = const()[name = tensor<string, []>("op_43520_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_43520_cast_fp16 = slice_by_index(begin = var_43520_begin_0, end = var_43520_end_0, end_mask = var_43520_end_mask_0, x = k_387_cast_fp16)[name = tensor<string, []>("op_43520_cast_fp16")];
            tensor<int32, [4]> var_43522_begin_0 = const()[name = tensor<string, []>("op_43522_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_43522_end_0 = const()[name = tensor<string, []>("op_43522_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_43522_end_mask_0 = const()[name = tensor<string, []>("op_43522_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43522_cast_fp16 = slice_by_index(begin = var_43522_begin_0, end = var_43522_end_0, end_mask = var_43522_end_mask_0, x = v_193_cast_fp16)[name = tensor<string, []>("op_43522_cast_fp16")];
            tensor<int32, [4]> var_43526_begin_0 = const()[name = tensor<string, []>("op_43526_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_43526_end_0 = const()[name = tensor<string, []>("op_43526_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_43526_end_mask_0 = const()[name = tensor<string, []>("op_43526_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43526_cast_fp16 = slice_by_index(begin = var_43526_begin_0, end = var_43526_end_0, end_mask = var_43526_end_mask_0, x = v_193_cast_fp16)[name = tensor<string, []>("op_43526_cast_fp16")];
            tensor<int32, [4]> var_43530_begin_0 = const()[name = tensor<string, []>("op_43530_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_43530_end_0 = const()[name = tensor<string, []>("op_43530_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_43530_end_mask_0 = const()[name = tensor<string, []>("op_43530_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43530_cast_fp16 = slice_by_index(begin = var_43530_begin_0, end = var_43530_end_0, end_mask = var_43530_end_mask_0, x = v_193_cast_fp16)[name = tensor<string, []>("op_43530_cast_fp16")];
            tensor<int32, [4]> var_43534_begin_0 = const()[name = tensor<string, []>("op_43534_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_43534_end_0 = const()[name = tensor<string, []>("op_43534_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_43534_end_mask_0 = const()[name = tensor<string, []>("op_43534_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43534_cast_fp16 = slice_by_index(begin = var_43534_begin_0, end = var_43534_end_0, end_mask = var_43534_end_mask_0, x = v_193_cast_fp16)[name = tensor<string, []>("op_43534_cast_fp16")];
            tensor<int32, [4]> var_43538_begin_0 = const()[name = tensor<string, []>("op_43538_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_43538_end_0 = const()[name = tensor<string, []>("op_43538_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_43538_end_mask_0 = const()[name = tensor<string, []>("op_43538_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43538_cast_fp16 = slice_by_index(begin = var_43538_begin_0, end = var_43538_end_0, end_mask = var_43538_end_mask_0, x = v_193_cast_fp16)[name = tensor<string, []>("op_43538_cast_fp16")];
            tensor<int32, [4]> var_43542_begin_0 = const()[name = tensor<string, []>("op_43542_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_43542_end_0 = const()[name = tensor<string, []>("op_43542_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_43542_end_mask_0 = const()[name = tensor<string, []>("op_43542_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43542_cast_fp16 = slice_by_index(begin = var_43542_begin_0, end = var_43542_end_0, end_mask = var_43542_end_mask_0, x = v_193_cast_fp16)[name = tensor<string, []>("op_43542_cast_fp16")];
            tensor<int32, [4]> var_43546_begin_0 = const()[name = tensor<string, []>("op_43546_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_43546_end_0 = const()[name = tensor<string, []>("op_43546_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_43546_end_mask_0 = const()[name = tensor<string, []>("op_43546_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43546_cast_fp16 = slice_by_index(begin = var_43546_begin_0, end = var_43546_end_0, end_mask = var_43546_end_mask_0, x = v_193_cast_fp16)[name = tensor<string, []>("op_43546_cast_fp16")];
            tensor<int32, [4]> var_43550_begin_0 = const()[name = tensor<string, []>("op_43550_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_43550_end_0 = const()[name = tensor<string, []>("op_43550_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_43550_end_mask_0 = const()[name = tensor<string, []>("op_43550_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43550_cast_fp16 = slice_by_index(begin = var_43550_begin_0, end = var_43550_end_0, end_mask = var_43550_end_mask_0, x = v_193_cast_fp16)[name = tensor<string, []>("op_43550_cast_fp16")];
            tensor<int32, [4]> var_43554_begin_0 = const()[name = tensor<string, []>("op_43554_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_43554_end_0 = const()[name = tensor<string, []>("op_43554_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_43554_end_mask_0 = const()[name = tensor<string, []>("op_43554_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43554_cast_fp16 = slice_by_index(begin = var_43554_begin_0, end = var_43554_end_0, end_mask = var_43554_end_mask_0, x = v_193_cast_fp16)[name = tensor<string, []>("op_43554_cast_fp16")];
            tensor<int32, [4]> var_43558_begin_0 = const()[name = tensor<string, []>("op_43558_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_43558_end_0 = const()[name = tensor<string, []>("op_43558_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_43558_end_mask_0 = const()[name = tensor<string, []>("op_43558_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43558_cast_fp16 = slice_by_index(begin = var_43558_begin_0, end = var_43558_end_0, end_mask = var_43558_end_mask_0, x = v_193_cast_fp16)[name = tensor<string, []>("op_43558_cast_fp16")];
            tensor<int32, [4]> var_43562_begin_0 = const()[name = tensor<string, []>("op_43562_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_43562_end_0 = const()[name = tensor<string, []>("op_43562_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_43562_end_mask_0 = const()[name = tensor<string, []>("op_43562_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43562_cast_fp16 = slice_by_index(begin = var_43562_begin_0, end = var_43562_end_0, end_mask = var_43562_end_mask_0, x = v_193_cast_fp16)[name = tensor<string, []>("op_43562_cast_fp16")];
            tensor<int32, [4]> var_43566_begin_0 = const()[name = tensor<string, []>("op_43566_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_43566_end_0 = const()[name = tensor<string, []>("op_43566_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_43566_end_mask_0 = const()[name = tensor<string, []>("op_43566_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43566_cast_fp16 = slice_by_index(begin = var_43566_begin_0, end = var_43566_end_0, end_mask = var_43566_end_mask_0, x = v_193_cast_fp16)[name = tensor<string, []>("op_43566_cast_fp16")];
            tensor<int32, [4]> var_43570_begin_0 = const()[name = tensor<string, []>("op_43570_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_43570_end_0 = const()[name = tensor<string, []>("op_43570_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_43570_end_mask_0 = const()[name = tensor<string, []>("op_43570_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43570_cast_fp16 = slice_by_index(begin = var_43570_begin_0, end = var_43570_end_0, end_mask = var_43570_end_mask_0, x = v_193_cast_fp16)[name = tensor<string, []>("op_43570_cast_fp16")];
            tensor<int32, [4]> var_43574_begin_0 = const()[name = tensor<string, []>("op_43574_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_43574_end_0 = const()[name = tensor<string, []>("op_43574_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_43574_end_mask_0 = const()[name = tensor<string, []>("op_43574_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43574_cast_fp16 = slice_by_index(begin = var_43574_begin_0, end = var_43574_end_0, end_mask = var_43574_end_mask_0, x = v_193_cast_fp16)[name = tensor<string, []>("op_43574_cast_fp16")];
            tensor<int32, [4]> var_43578_begin_0 = const()[name = tensor<string, []>("op_43578_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_43578_end_0 = const()[name = tensor<string, []>("op_43578_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_43578_end_mask_0 = const()[name = tensor<string, []>("op_43578_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43578_cast_fp16 = slice_by_index(begin = var_43578_begin_0, end = var_43578_end_0, end_mask = var_43578_end_mask_0, x = v_193_cast_fp16)[name = tensor<string, []>("op_43578_cast_fp16")];
            tensor<int32, [4]> var_43582_begin_0 = const()[name = tensor<string, []>("op_43582_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_43582_end_0 = const()[name = tensor<string, []>("op_43582_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_43582_end_mask_0 = const()[name = tensor<string, []>("op_43582_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43582_cast_fp16 = slice_by_index(begin = var_43582_begin_0, end = var_43582_end_0, end_mask = var_43582_end_mask_0, x = v_193_cast_fp16)[name = tensor<string, []>("op_43582_cast_fp16")];
            tensor<int32, [4]> var_43586_begin_0 = const()[name = tensor<string, []>("op_43586_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_43586_end_0 = const()[name = tensor<string, []>("op_43586_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_43586_end_mask_0 = const()[name = tensor<string, []>("op_43586_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43586_cast_fp16 = slice_by_index(begin = var_43586_begin_0, end = var_43586_end_0, end_mask = var_43586_end_mask_0, x = v_193_cast_fp16)[name = tensor<string, []>("op_43586_cast_fp16")];
            tensor<int32, [4]> var_43590_begin_0 = const()[name = tensor<string, []>("op_43590_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_43590_end_0 = const()[name = tensor<string, []>("op_43590_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_43590_end_mask_0 = const()[name = tensor<string, []>("op_43590_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43590_cast_fp16 = slice_by_index(begin = var_43590_begin_0, end = var_43590_end_0, end_mask = var_43590_end_mask_0, x = v_193_cast_fp16)[name = tensor<string, []>("op_43590_cast_fp16")];
            tensor<int32, [4]> var_43594_begin_0 = const()[name = tensor<string, []>("op_43594_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_43594_end_0 = const()[name = tensor<string, []>("op_43594_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_43594_end_mask_0 = const()[name = tensor<string, []>("op_43594_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43594_cast_fp16 = slice_by_index(begin = var_43594_begin_0, end = var_43594_end_0, end_mask = var_43594_end_mask_0, x = v_193_cast_fp16)[name = tensor<string, []>("op_43594_cast_fp16")];
            tensor<int32, [4]> var_43598_begin_0 = const()[name = tensor<string, []>("op_43598_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_43598_end_0 = const()[name = tensor<string, []>("op_43598_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_43598_end_mask_0 = const()[name = tensor<string, []>("op_43598_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43598_cast_fp16 = slice_by_index(begin = var_43598_begin_0, end = var_43598_end_0, end_mask = var_43598_end_mask_0, x = v_193_cast_fp16)[name = tensor<string, []>("op_43598_cast_fp16")];
            tensor<string, []> var_43602_equation_0 = const()[name = tensor<string, []>("op_43602_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43602_cast_fp16 = einsum(equation = var_43602_equation_0, values = (var_43444_cast_fp16, var_43361_cast_fp16))[name = tensor<string, []>("op_43602_cast_fp16")];
            tensor<fp16, []> var_43603_to_fp16 = const()[name = tensor<string, []>("op_43603_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3681_cast_fp16 = mul(x = var_43602_cast_fp16, y = var_43603_to_fp16)[name = tensor<string, []>("aw_3681_cast_fp16")];
            tensor<string, []> var_43606_equation_0 = const()[name = tensor<string, []>("op_43606_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43606_cast_fp16 = einsum(equation = var_43606_equation_0, values = (var_43448_cast_fp16, var_43365_cast_fp16))[name = tensor<string, []>("op_43606_cast_fp16")];
            tensor<fp16, []> var_43607_to_fp16 = const()[name = tensor<string, []>("op_43607_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3683_cast_fp16 = mul(x = var_43606_cast_fp16, y = var_43607_to_fp16)[name = tensor<string, []>("aw_3683_cast_fp16")];
            tensor<string, []> var_43610_equation_0 = const()[name = tensor<string, []>("op_43610_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43610_cast_fp16 = einsum(equation = var_43610_equation_0, values = (var_43452_cast_fp16, var_43369_cast_fp16))[name = tensor<string, []>("op_43610_cast_fp16")];
            tensor<fp16, []> var_43611_to_fp16 = const()[name = tensor<string, []>("op_43611_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3685_cast_fp16 = mul(x = var_43610_cast_fp16, y = var_43611_to_fp16)[name = tensor<string, []>("aw_3685_cast_fp16")];
            tensor<string, []> var_43614_equation_0 = const()[name = tensor<string, []>("op_43614_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43614_cast_fp16 = einsum(equation = var_43614_equation_0, values = (var_43456_cast_fp16, var_43373_cast_fp16))[name = tensor<string, []>("op_43614_cast_fp16")];
            tensor<fp16, []> var_43615_to_fp16 = const()[name = tensor<string, []>("op_43615_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3687_cast_fp16 = mul(x = var_43614_cast_fp16, y = var_43615_to_fp16)[name = tensor<string, []>("aw_3687_cast_fp16")];
            tensor<string, []> var_43618_equation_0 = const()[name = tensor<string, []>("op_43618_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43618_cast_fp16 = einsum(equation = var_43618_equation_0, values = (var_43460_cast_fp16, var_43377_cast_fp16))[name = tensor<string, []>("op_43618_cast_fp16")];
            tensor<fp16, []> var_43619_to_fp16 = const()[name = tensor<string, []>("op_43619_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3689_cast_fp16 = mul(x = var_43618_cast_fp16, y = var_43619_to_fp16)[name = tensor<string, []>("aw_3689_cast_fp16")];
            tensor<string, []> var_43622_equation_0 = const()[name = tensor<string, []>("op_43622_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43622_cast_fp16 = einsum(equation = var_43622_equation_0, values = (var_43464_cast_fp16, var_43381_cast_fp16))[name = tensor<string, []>("op_43622_cast_fp16")];
            tensor<fp16, []> var_43623_to_fp16 = const()[name = tensor<string, []>("op_43623_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3691_cast_fp16 = mul(x = var_43622_cast_fp16, y = var_43623_to_fp16)[name = tensor<string, []>("aw_3691_cast_fp16")];
            tensor<string, []> var_43626_equation_0 = const()[name = tensor<string, []>("op_43626_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43626_cast_fp16 = einsum(equation = var_43626_equation_0, values = (var_43468_cast_fp16, var_43385_cast_fp16))[name = tensor<string, []>("op_43626_cast_fp16")];
            tensor<fp16, []> var_43627_to_fp16 = const()[name = tensor<string, []>("op_43627_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3693_cast_fp16 = mul(x = var_43626_cast_fp16, y = var_43627_to_fp16)[name = tensor<string, []>("aw_3693_cast_fp16")];
            tensor<string, []> var_43630_equation_0 = const()[name = tensor<string, []>("op_43630_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43630_cast_fp16 = einsum(equation = var_43630_equation_0, values = (var_43472_cast_fp16, var_43389_cast_fp16))[name = tensor<string, []>("op_43630_cast_fp16")];
            tensor<fp16, []> var_43631_to_fp16 = const()[name = tensor<string, []>("op_43631_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3695_cast_fp16 = mul(x = var_43630_cast_fp16, y = var_43631_to_fp16)[name = tensor<string, []>("aw_3695_cast_fp16")];
            tensor<string, []> var_43634_equation_0 = const()[name = tensor<string, []>("op_43634_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43634_cast_fp16 = einsum(equation = var_43634_equation_0, values = (var_43476_cast_fp16, var_43393_cast_fp16))[name = tensor<string, []>("op_43634_cast_fp16")];
            tensor<fp16, []> var_43635_to_fp16 = const()[name = tensor<string, []>("op_43635_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3697_cast_fp16 = mul(x = var_43634_cast_fp16, y = var_43635_to_fp16)[name = tensor<string, []>("aw_3697_cast_fp16")];
            tensor<string, []> var_43638_equation_0 = const()[name = tensor<string, []>("op_43638_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43638_cast_fp16 = einsum(equation = var_43638_equation_0, values = (var_43480_cast_fp16, var_43397_cast_fp16))[name = tensor<string, []>("op_43638_cast_fp16")];
            tensor<fp16, []> var_43639_to_fp16 = const()[name = tensor<string, []>("op_43639_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3699_cast_fp16 = mul(x = var_43638_cast_fp16, y = var_43639_to_fp16)[name = tensor<string, []>("aw_3699_cast_fp16")];
            tensor<string, []> var_43642_equation_0 = const()[name = tensor<string, []>("op_43642_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43642_cast_fp16 = einsum(equation = var_43642_equation_0, values = (var_43484_cast_fp16, var_43401_cast_fp16))[name = tensor<string, []>("op_43642_cast_fp16")];
            tensor<fp16, []> var_43643_to_fp16 = const()[name = tensor<string, []>("op_43643_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3701_cast_fp16 = mul(x = var_43642_cast_fp16, y = var_43643_to_fp16)[name = tensor<string, []>("aw_3701_cast_fp16")];
            tensor<string, []> var_43646_equation_0 = const()[name = tensor<string, []>("op_43646_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43646_cast_fp16 = einsum(equation = var_43646_equation_0, values = (var_43488_cast_fp16, var_43405_cast_fp16))[name = tensor<string, []>("op_43646_cast_fp16")];
            tensor<fp16, []> var_43647_to_fp16 = const()[name = tensor<string, []>("op_43647_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3703_cast_fp16 = mul(x = var_43646_cast_fp16, y = var_43647_to_fp16)[name = tensor<string, []>("aw_3703_cast_fp16")];
            tensor<string, []> var_43650_equation_0 = const()[name = tensor<string, []>("op_43650_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43650_cast_fp16 = einsum(equation = var_43650_equation_0, values = (var_43492_cast_fp16, var_43409_cast_fp16))[name = tensor<string, []>("op_43650_cast_fp16")];
            tensor<fp16, []> var_43651_to_fp16 = const()[name = tensor<string, []>("op_43651_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3705_cast_fp16 = mul(x = var_43650_cast_fp16, y = var_43651_to_fp16)[name = tensor<string, []>("aw_3705_cast_fp16")];
            tensor<string, []> var_43654_equation_0 = const()[name = tensor<string, []>("op_43654_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43654_cast_fp16 = einsum(equation = var_43654_equation_0, values = (var_43496_cast_fp16, var_43413_cast_fp16))[name = tensor<string, []>("op_43654_cast_fp16")];
            tensor<fp16, []> var_43655_to_fp16 = const()[name = tensor<string, []>("op_43655_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3707_cast_fp16 = mul(x = var_43654_cast_fp16, y = var_43655_to_fp16)[name = tensor<string, []>("aw_3707_cast_fp16")];
            tensor<string, []> var_43658_equation_0 = const()[name = tensor<string, []>("op_43658_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43658_cast_fp16 = einsum(equation = var_43658_equation_0, values = (var_43500_cast_fp16, var_43417_cast_fp16))[name = tensor<string, []>("op_43658_cast_fp16")];
            tensor<fp16, []> var_43659_to_fp16 = const()[name = tensor<string, []>("op_43659_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3709_cast_fp16 = mul(x = var_43658_cast_fp16, y = var_43659_to_fp16)[name = tensor<string, []>("aw_3709_cast_fp16")];
            tensor<string, []> var_43662_equation_0 = const()[name = tensor<string, []>("op_43662_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43662_cast_fp16 = einsum(equation = var_43662_equation_0, values = (var_43504_cast_fp16, var_43421_cast_fp16))[name = tensor<string, []>("op_43662_cast_fp16")];
            tensor<fp16, []> var_43663_to_fp16 = const()[name = tensor<string, []>("op_43663_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3711_cast_fp16 = mul(x = var_43662_cast_fp16, y = var_43663_to_fp16)[name = tensor<string, []>("aw_3711_cast_fp16")];
            tensor<string, []> var_43666_equation_0 = const()[name = tensor<string, []>("op_43666_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43666_cast_fp16 = einsum(equation = var_43666_equation_0, values = (var_43508_cast_fp16, var_43425_cast_fp16))[name = tensor<string, []>("op_43666_cast_fp16")];
            tensor<fp16, []> var_43667_to_fp16 = const()[name = tensor<string, []>("op_43667_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3713_cast_fp16 = mul(x = var_43666_cast_fp16, y = var_43667_to_fp16)[name = tensor<string, []>("aw_3713_cast_fp16")];
            tensor<string, []> var_43670_equation_0 = const()[name = tensor<string, []>("op_43670_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43670_cast_fp16 = einsum(equation = var_43670_equation_0, values = (var_43512_cast_fp16, var_43429_cast_fp16))[name = tensor<string, []>("op_43670_cast_fp16")];
            tensor<fp16, []> var_43671_to_fp16 = const()[name = tensor<string, []>("op_43671_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3715_cast_fp16 = mul(x = var_43670_cast_fp16, y = var_43671_to_fp16)[name = tensor<string, []>("aw_3715_cast_fp16")];
            tensor<string, []> var_43674_equation_0 = const()[name = tensor<string, []>("op_43674_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43674_cast_fp16 = einsum(equation = var_43674_equation_0, values = (var_43516_cast_fp16, var_43433_cast_fp16))[name = tensor<string, []>("op_43674_cast_fp16")];
            tensor<fp16, []> var_43675_to_fp16 = const()[name = tensor<string, []>("op_43675_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3717_cast_fp16 = mul(x = var_43674_cast_fp16, y = var_43675_to_fp16)[name = tensor<string, []>("aw_3717_cast_fp16")];
            tensor<string, []> var_43678_equation_0 = const()[name = tensor<string, []>("op_43678_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43678_cast_fp16 = einsum(equation = var_43678_equation_0, values = (var_43520_cast_fp16, var_43437_cast_fp16))[name = tensor<string, []>("op_43678_cast_fp16")];
            tensor<fp16, []> var_43679_to_fp16 = const()[name = tensor<string, []>("op_43679_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3719_cast_fp16 = mul(x = var_43678_cast_fp16, y = var_43679_to_fp16)[name = tensor<string, []>("aw_3719_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43681_cast_fp16 = softmax(axis = var_30355, x = aw_3681_cast_fp16)[name = tensor<string, []>("op_43681_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43682_cast_fp16 = softmax(axis = var_30355, x = aw_3683_cast_fp16)[name = tensor<string, []>("op_43682_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43683_cast_fp16 = softmax(axis = var_30355, x = aw_3685_cast_fp16)[name = tensor<string, []>("op_43683_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43684_cast_fp16 = softmax(axis = var_30355, x = aw_3687_cast_fp16)[name = tensor<string, []>("op_43684_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43685_cast_fp16 = softmax(axis = var_30355, x = aw_3689_cast_fp16)[name = tensor<string, []>("op_43685_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43686_cast_fp16 = softmax(axis = var_30355, x = aw_3691_cast_fp16)[name = tensor<string, []>("op_43686_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43687_cast_fp16 = softmax(axis = var_30355, x = aw_3693_cast_fp16)[name = tensor<string, []>("op_43687_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43688_cast_fp16 = softmax(axis = var_30355, x = aw_3695_cast_fp16)[name = tensor<string, []>("op_43688_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43689_cast_fp16 = softmax(axis = var_30355, x = aw_3697_cast_fp16)[name = tensor<string, []>("op_43689_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43690_cast_fp16 = softmax(axis = var_30355, x = aw_3699_cast_fp16)[name = tensor<string, []>("op_43690_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43691_cast_fp16 = softmax(axis = var_30355, x = aw_3701_cast_fp16)[name = tensor<string, []>("op_43691_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43692_cast_fp16 = softmax(axis = var_30355, x = aw_3703_cast_fp16)[name = tensor<string, []>("op_43692_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43693_cast_fp16 = softmax(axis = var_30355, x = aw_3705_cast_fp16)[name = tensor<string, []>("op_43693_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43694_cast_fp16 = softmax(axis = var_30355, x = aw_3707_cast_fp16)[name = tensor<string, []>("op_43694_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43695_cast_fp16 = softmax(axis = var_30355, x = aw_3709_cast_fp16)[name = tensor<string, []>("op_43695_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43696_cast_fp16 = softmax(axis = var_30355, x = aw_3711_cast_fp16)[name = tensor<string, []>("op_43696_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43697_cast_fp16 = softmax(axis = var_30355, x = aw_3713_cast_fp16)[name = tensor<string, []>("op_43697_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43698_cast_fp16 = softmax(axis = var_30355, x = aw_3715_cast_fp16)[name = tensor<string, []>("op_43698_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43699_cast_fp16 = softmax(axis = var_30355, x = aw_3717_cast_fp16)[name = tensor<string, []>("op_43699_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43700_cast_fp16 = softmax(axis = var_30355, x = aw_3719_cast_fp16)[name = tensor<string, []>("op_43700_cast_fp16")];
            tensor<string, []> var_43702_equation_0 = const()[name = tensor<string, []>("op_43702_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43702_cast_fp16 = einsum(equation = var_43702_equation_0, values = (var_43522_cast_fp16, var_43681_cast_fp16))[name = tensor<string, []>("op_43702_cast_fp16")];
            tensor<string, []> var_43704_equation_0 = const()[name = tensor<string, []>("op_43704_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43704_cast_fp16 = einsum(equation = var_43704_equation_0, values = (var_43526_cast_fp16, var_43682_cast_fp16))[name = tensor<string, []>("op_43704_cast_fp16")];
            tensor<string, []> var_43706_equation_0 = const()[name = tensor<string, []>("op_43706_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43706_cast_fp16 = einsum(equation = var_43706_equation_0, values = (var_43530_cast_fp16, var_43683_cast_fp16))[name = tensor<string, []>("op_43706_cast_fp16")];
            tensor<string, []> var_43708_equation_0 = const()[name = tensor<string, []>("op_43708_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43708_cast_fp16 = einsum(equation = var_43708_equation_0, values = (var_43534_cast_fp16, var_43684_cast_fp16))[name = tensor<string, []>("op_43708_cast_fp16")];
            tensor<string, []> var_43710_equation_0 = const()[name = tensor<string, []>("op_43710_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43710_cast_fp16 = einsum(equation = var_43710_equation_0, values = (var_43538_cast_fp16, var_43685_cast_fp16))[name = tensor<string, []>("op_43710_cast_fp16")];
            tensor<string, []> var_43712_equation_0 = const()[name = tensor<string, []>("op_43712_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43712_cast_fp16 = einsum(equation = var_43712_equation_0, values = (var_43542_cast_fp16, var_43686_cast_fp16))[name = tensor<string, []>("op_43712_cast_fp16")];
            tensor<string, []> var_43714_equation_0 = const()[name = tensor<string, []>("op_43714_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43714_cast_fp16 = einsum(equation = var_43714_equation_0, values = (var_43546_cast_fp16, var_43687_cast_fp16))[name = tensor<string, []>("op_43714_cast_fp16")];
            tensor<string, []> var_43716_equation_0 = const()[name = tensor<string, []>("op_43716_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43716_cast_fp16 = einsum(equation = var_43716_equation_0, values = (var_43550_cast_fp16, var_43688_cast_fp16))[name = tensor<string, []>("op_43716_cast_fp16")];
            tensor<string, []> var_43718_equation_0 = const()[name = tensor<string, []>("op_43718_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43718_cast_fp16 = einsum(equation = var_43718_equation_0, values = (var_43554_cast_fp16, var_43689_cast_fp16))[name = tensor<string, []>("op_43718_cast_fp16")];
            tensor<string, []> var_43720_equation_0 = const()[name = tensor<string, []>("op_43720_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43720_cast_fp16 = einsum(equation = var_43720_equation_0, values = (var_43558_cast_fp16, var_43690_cast_fp16))[name = tensor<string, []>("op_43720_cast_fp16")];
            tensor<string, []> var_43722_equation_0 = const()[name = tensor<string, []>("op_43722_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43722_cast_fp16 = einsum(equation = var_43722_equation_0, values = (var_43562_cast_fp16, var_43691_cast_fp16))[name = tensor<string, []>("op_43722_cast_fp16")];
            tensor<string, []> var_43724_equation_0 = const()[name = tensor<string, []>("op_43724_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43724_cast_fp16 = einsum(equation = var_43724_equation_0, values = (var_43566_cast_fp16, var_43692_cast_fp16))[name = tensor<string, []>("op_43724_cast_fp16")];
            tensor<string, []> var_43726_equation_0 = const()[name = tensor<string, []>("op_43726_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43726_cast_fp16 = einsum(equation = var_43726_equation_0, values = (var_43570_cast_fp16, var_43693_cast_fp16))[name = tensor<string, []>("op_43726_cast_fp16")];
            tensor<string, []> var_43728_equation_0 = const()[name = tensor<string, []>("op_43728_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43728_cast_fp16 = einsum(equation = var_43728_equation_0, values = (var_43574_cast_fp16, var_43694_cast_fp16))[name = tensor<string, []>("op_43728_cast_fp16")];
            tensor<string, []> var_43730_equation_0 = const()[name = tensor<string, []>("op_43730_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43730_cast_fp16 = einsum(equation = var_43730_equation_0, values = (var_43578_cast_fp16, var_43695_cast_fp16))[name = tensor<string, []>("op_43730_cast_fp16")];
            tensor<string, []> var_43732_equation_0 = const()[name = tensor<string, []>("op_43732_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43732_cast_fp16 = einsum(equation = var_43732_equation_0, values = (var_43582_cast_fp16, var_43696_cast_fp16))[name = tensor<string, []>("op_43732_cast_fp16")];
            tensor<string, []> var_43734_equation_0 = const()[name = tensor<string, []>("op_43734_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43734_cast_fp16 = einsum(equation = var_43734_equation_0, values = (var_43586_cast_fp16, var_43697_cast_fp16))[name = tensor<string, []>("op_43734_cast_fp16")];
            tensor<string, []> var_43736_equation_0 = const()[name = tensor<string, []>("op_43736_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43736_cast_fp16 = einsum(equation = var_43736_equation_0, values = (var_43590_cast_fp16, var_43698_cast_fp16))[name = tensor<string, []>("op_43736_cast_fp16")];
            tensor<string, []> var_43738_equation_0 = const()[name = tensor<string, []>("op_43738_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43738_cast_fp16 = einsum(equation = var_43738_equation_0, values = (var_43594_cast_fp16, var_43699_cast_fp16))[name = tensor<string, []>("op_43738_cast_fp16")];
            tensor<string, []> var_43740_equation_0 = const()[name = tensor<string, []>("op_43740_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43740_cast_fp16 = einsum(equation = var_43740_equation_0, values = (var_43598_cast_fp16, var_43700_cast_fp16))[name = tensor<string, []>("op_43740_cast_fp16")];
            tensor<bool, []> input_569_interleave_0 = const()[name = tensor<string, []>("input_569_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_569_cast_fp16 = concat(axis = var_30355, interleave = input_569_interleave_0, values = (var_43702_cast_fp16, var_43704_cast_fp16, var_43706_cast_fp16, var_43708_cast_fp16, var_43710_cast_fp16, var_43712_cast_fp16, var_43714_cast_fp16, var_43716_cast_fp16, var_43718_cast_fp16, var_43720_cast_fp16, var_43722_cast_fp16, var_43724_cast_fp16, var_43726_cast_fp16, var_43728_cast_fp16, var_43730_cast_fp16, var_43732_cast_fp16, var_43734_cast_fp16, var_43736_cast_fp16, var_43738_cast_fp16, var_43740_cast_fp16))[name = tensor<string, []>("input_569_cast_fp16")];
            tensor<string, []> var_43750_pad_type_0 = const()[name = tensor<string, []>("op_43750_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_43750_strides_0 = const()[name = tensor<string, []>("op_43750_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_43750_pad_0 = const()[name = tensor<string, []>("op_43750_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_43750_dilations_0 = const()[name = tensor<string, []>("op_43750_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_43750_groups_0 = const()[name = tensor<string, []>("op_43750_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_4_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(412131264))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(413360128))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_4_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_4_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_4_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(413360320)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_43750_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_4_attn1_to_out_0_bias_to_fp16, dilations = var_43750_dilations_0, groups = var_43750_groups_0, pad = var_43750_pad_0, pad_type = var_43750_pad_type_0, strides = var_43750_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_4_attn1_to_out_0_weight_to_fp16_palettized, x = input_569_cast_fp16)[name = tensor<string, []>("op_43750_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_291_cast_fp16 = add(x = var_43750_cast_fp16, y = inputs_289_cast_fp16)[name = tensor<string, []>("inputs_291_cast_fp16")];
            tensor<int32, [1]> hidden_states_387_axes_0 = const()[name = tensor<string, []>("hidden_states_387_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_387_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_387_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(413362944)))];
            tensor<fp16, [1280]> hidden_states_387_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_387_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(413365568)))];
            tensor<fp16, []> var_43760_to_fp16 = const()[name = tensor<string, []>("op_43760_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_387_cast_fp16 = layer_norm(axes = hidden_states_387_axes_0, beta = hidden_states_387_beta_0_to_fp16, epsilon = var_43760_to_fp16, gamma = hidden_states_387_gamma_0_to_fp16, x = inputs_291_cast_fp16)[name = tensor<string, []>("hidden_states_387_cast_fp16")];
            tensor<string, []> q_195_pad_type_0 = const()[name = tensor<string, []>("q_195_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_195_strides_0 = const()[name = tensor<string, []>("q_195_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_195_pad_0 = const()[name = tensor<string, []>("q_195_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_195_dilations_0 = const()[name = tensor<string, []>("q_195_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_195_groups_0 = const()[name = tensor<string, []>("q_195_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_4_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(413368192))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(414597056))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_4_attn2_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_195_cast_fp16 = conv(dilations = q_195_dilations_0, groups = q_195_groups_0, pad = q_195_pad_0, pad_type = q_195_pad_type_0, strides = q_195_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_4_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_387_cast_fp16)[name = tensor<string, []>("q_195_cast_fp16")];
            tensor<string, []> k_389_pad_type_0 = const()[name = tensor<string, []>("k_389_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_389_strides_0 = const()[name = tensor<string, []>("k_389_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_389_pad_0 = const()[name = tensor<string, []>("k_389_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_389_dilations_0 = const()[name = tensor<string, []>("k_389_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_389_groups_0 = const()[name = tensor<string, []>("k_389_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_4_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(414597248))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(416563392))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_4_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> k_389_cast_fp16 = conv(dilations = k_389_dilations_0, groups = k_389_groups_0, pad = k_389_pad_0, pad_type = k_389_pad_type_0, strides = k_389_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_4_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("k_389_cast_fp16")];
            tensor<string, []> v_195_pad_type_0 = const()[name = tensor<string, []>("v_195_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_195_strides_0 = const()[name = tensor<string, []>("v_195_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_195_pad_0 = const()[name = tensor<string, []>("v_195_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_195_dilations_0 = const()[name = tensor<string, []>("v_195_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_195_groups_0 = const()[name = tensor<string, []>("v_195_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_4_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(416563584))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(418529728))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_4_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> v_195_cast_fp16 = conv(dilations = v_195_dilations_0, groups = v_195_groups_0, pad = v_195_pad_0, pad_type = v_195_pad_type_0, strides = v_195_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_4_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("v_195_cast_fp16")];
            tensor<int32, [4]> var_43793_begin_0 = const()[name = tensor<string, []>("op_43793_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_43793_end_0 = const()[name = tensor<string, []>("op_43793_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_43793_end_mask_0 = const()[name = tensor<string, []>("op_43793_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43793_cast_fp16 = slice_by_index(begin = var_43793_begin_0, end = var_43793_end_0, end_mask = var_43793_end_mask_0, x = q_195_cast_fp16)[name = tensor<string, []>("op_43793_cast_fp16")];
            tensor<int32, [4]> var_43797_begin_0 = const()[name = tensor<string, []>("op_43797_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_43797_end_0 = const()[name = tensor<string, []>("op_43797_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_43797_end_mask_0 = const()[name = tensor<string, []>("op_43797_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43797_cast_fp16 = slice_by_index(begin = var_43797_begin_0, end = var_43797_end_0, end_mask = var_43797_end_mask_0, x = q_195_cast_fp16)[name = tensor<string, []>("op_43797_cast_fp16")];
            tensor<int32, [4]> var_43801_begin_0 = const()[name = tensor<string, []>("op_43801_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_43801_end_0 = const()[name = tensor<string, []>("op_43801_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_43801_end_mask_0 = const()[name = tensor<string, []>("op_43801_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43801_cast_fp16 = slice_by_index(begin = var_43801_begin_0, end = var_43801_end_0, end_mask = var_43801_end_mask_0, x = q_195_cast_fp16)[name = tensor<string, []>("op_43801_cast_fp16")];
            tensor<int32, [4]> var_43805_begin_0 = const()[name = tensor<string, []>("op_43805_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_43805_end_0 = const()[name = tensor<string, []>("op_43805_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_43805_end_mask_0 = const()[name = tensor<string, []>("op_43805_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43805_cast_fp16 = slice_by_index(begin = var_43805_begin_0, end = var_43805_end_0, end_mask = var_43805_end_mask_0, x = q_195_cast_fp16)[name = tensor<string, []>("op_43805_cast_fp16")];
            tensor<int32, [4]> var_43809_begin_0 = const()[name = tensor<string, []>("op_43809_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_43809_end_0 = const()[name = tensor<string, []>("op_43809_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_43809_end_mask_0 = const()[name = tensor<string, []>("op_43809_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43809_cast_fp16 = slice_by_index(begin = var_43809_begin_0, end = var_43809_end_0, end_mask = var_43809_end_mask_0, x = q_195_cast_fp16)[name = tensor<string, []>("op_43809_cast_fp16")];
            tensor<int32, [4]> var_43813_begin_0 = const()[name = tensor<string, []>("op_43813_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_43813_end_0 = const()[name = tensor<string, []>("op_43813_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_43813_end_mask_0 = const()[name = tensor<string, []>("op_43813_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43813_cast_fp16 = slice_by_index(begin = var_43813_begin_0, end = var_43813_end_0, end_mask = var_43813_end_mask_0, x = q_195_cast_fp16)[name = tensor<string, []>("op_43813_cast_fp16")];
            tensor<int32, [4]> var_43817_begin_0 = const()[name = tensor<string, []>("op_43817_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_43817_end_0 = const()[name = tensor<string, []>("op_43817_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_43817_end_mask_0 = const()[name = tensor<string, []>("op_43817_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43817_cast_fp16 = slice_by_index(begin = var_43817_begin_0, end = var_43817_end_0, end_mask = var_43817_end_mask_0, x = q_195_cast_fp16)[name = tensor<string, []>("op_43817_cast_fp16")];
            tensor<int32, [4]> var_43821_begin_0 = const()[name = tensor<string, []>("op_43821_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_43821_end_0 = const()[name = tensor<string, []>("op_43821_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_43821_end_mask_0 = const()[name = tensor<string, []>("op_43821_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43821_cast_fp16 = slice_by_index(begin = var_43821_begin_0, end = var_43821_end_0, end_mask = var_43821_end_mask_0, x = q_195_cast_fp16)[name = tensor<string, []>("op_43821_cast_fp16")];
            tensor<int32, [4]> var_43825_begin_0 = const()[name = tensor<string, []>("op_43825_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_43825_end_0 = const()[name = tensor<string, []>("op_43825_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_43825_end_mask_0 = const()[name = tensor<string, []>("op_43825_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43825_cast_fp16 = slice_by_index(begin = var_43825_begin_0, end = var_43825_end_0, end_mask = var_43825_end_mask_0, x = q_195_cast_fp16)[name = tensor<string, []>("op_43825_cast_fp16")];
            tensor<int32, [4]> var_43829_begin_0 = const()[name = tensor<string, []>("op_43829_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_43829_end_0 = const()[name = tensor<string, []>("op_43829_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_43829_end_mask_0 = const()[name = tensor<string, []>("op_43829_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43829_cast_fp16 = slice_by_index(begin = var_43829_begin_0, end = var_43829_end_0, end_mask = var_43829_end_mask_0, x = q_195_cast_fp16)[name = tensor<string, []>("op_43829_cast_fp16")];
            tensor<int32, [4]> var_43833_begin_0 = const()[name = tensor<string, []>("op_43833_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_43833_end_0 = const()[name = tensor<string, []>("op_43833_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_43833_end_mask_0 = const()[name = tensor<string, []>("op_43833_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43833_cast_fp16 = slice_by_index(begin = var_43833_begin_0, end = var_43833_end_0, end_mask = var_43833_end_mask_0, x = q_195_cast_fp16)[name = tensor<string, []>("op_43833_cast_fp16")];
            tensor<int32, [4]> var_43837_begin_0 = const()[name = tensor<string, []>("op_43837_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_43837_end_0 = const()[name = tensor<string, []>("op_43837_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_43837_end_mask_0 = const()[name = tensor<string, []>("op_43837_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43837_cast_fp16 = slice_by_index(begin = var_43837_begin_0, end = var_43837_end_0, end_mask = var_43837_end_mask_0, x = q_195_cast_fp16)[name = tensor<string, []>("op_43837_cast_fp16")];
            tensor<int32, [4]> var_43841_begin_0 = const()[name = tensor<string, []>("op_43841_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_43841_end_0 = const()[name = tensor<string, []>("op_43841_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_43841_end_mask_0 = const()[name = tensor<string, []>("op_43841_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43841_cast_fp16 = slice_by_index(begin = var_43841_begin_0, end = var_43841_end_0, end_mask = var_43841_end_mask_0, x = q_195_cast_fp16)[name = tensor<string, []>("op_43841_cast_fp16")];
            tensor<int32, [4]> var_43845_begin_0 = const()[name = tensor<string, []>("op_43845_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_43845_end_0 = const()[name = tensor<string, []>("op_43845_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_43845_end_mask_0 = const()[name = tensor<string, []>("op_43845_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43845_cast_fp16 = slice_by_index(begin = var_43845_begin_0, end = var_43845_end_0, end_mask = var_43845_end_mask_0, x = q_195_cast_fp16)[name = tensor<string, []>("op_43845_cast_fp16")];
            tensor<int32, [4]> var_43849_begin_0 = const()[name = tensor<string, []>("op_43849_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_43849_end_0 = const()[name = tensor<string, []>("op_43849_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_43849_end_mask_0 = const()[name = tensor<string, []>("op_43849_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43849_cast_fp16 = slice_by_index(begin = var_43849_begin_0, end = var_43849_end_0, end_mask = var_43849_end_mask_0, x = q_195_cast_fp16)[name = tensor<string, []>("op_43849_cast_fp16")];
            tensor<int32, [4]> var_43853_begin_0 = const()[name = tensor<string, []>("op_43853_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_43853_end_0 = const()[name = tensor<string, []>("op_43853_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_43853_end_mask_0 = const()[name = tensor<string, []>("op_43853_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43853_cast_fp16 = slice_by_index(begin = var_43853_begin_0, end = var_43853_end_0, end_mask = var_43853_end_mask_0, x = q_195_cast_fp16)[name = tensor<string, []>("op_43853_cast_fp16")];
            tensor<int32, [4]> var_43857_begin_0 = const()[name = tensor<string, []>("op_43857_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_43857_end_0 = const()[name = tensor<string, []>("op_43857_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_43857_end_mask_0 = const()[name = tensor<string, []>("op_43857_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43857_cast_fp16 = slice_by_index(begin = var_43857_begin_0, end = var_43857_end_0, end_mask = var_43857_end_mask_0, x = q_195_cast_fp16)[name = tensor<string, []>("op_43857_cast_fp16")];
            tensor<int32, [4]> var_43861_begin_0 = const()[name = tensor<string, []>("op_43861_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_43861_end_0 = const()[name = tensor<string, []>("op_43861_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_43861_end_mask_0 = const()[name = tensor<string, []>("op_43861_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43861_cast_fp16 = slice_by_index(begin = var_43861_begin_0, end = var_43861_end_0, end_mask = var_43861_end_mask_0, x = q_195_cast_fp16)[name = tensor<string, []>("op_43861_cast_fp16")];
            tensor<int32, [4]> var_43865_begin_0 = const()[name = tensor<string, []>("op_43865_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_43865_end_0 = const()[name = tensor<string, []>("op_43865_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_43865_end_mask_0 = const()[name = tensor<string, []>("op_43865_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43865_cast_fp16 = slice_by_index(begin = var_43865_begin_0, end = var_43865_end_0, end_mask = var_43865_end_mask_0, x = q_195_cast_fp16)[name = tensor<string, []>("op_43865_cast_fp16")];
            tensor<int32, [4]> var_43869_begin_0 = const()[name = tensor<string, []>("op_43869_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_43869_end_0 = const()[name = tensor<string, []>("op_43869_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_43869_end_mask_0 = const()[name = tensor<string, []>("op_43869_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43869_cast_fp16 = slice_by_index(begin = var_43869_begin_0, end = var_43869_end_0, end_mask = var_43869_end_mask_0, x = q_195_cast_fp16)[name = tensor<string, []>("op_43869_cast_fp16")];
            tensor<int32, [4]> k_391_perm_0 = const()[name = tensor<string, []>("k_391_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_43876_begin_0 = const()[name = tensor<string, []>("op_43876_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_43876_end_0 = const()[name = tensor<string, []>("op_43876_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_43876_end_mask_0 = const()[name = tensor<string, []>("op_43876_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> k_391_cast_fp16 = transpose(perm = k_391_perm_0, x = k_389_cast_fp16)[name = tensor<string, []>("transpose_42")];
            tensor<fp16, [2, 77, 1, 64]> var_43876_cast_fp16 = slice_by_index(begin = var_43876_begin_0, end = var_43876_end_0, end_mask = var_43876_end_mask_0, x = k_391_cast_fp16)[name = tensor<string, []>("op_43876_cast_fp16")];
            tensor<int32, [4]> var_43880_begin_0 = const()[name = tensor<string, []>("op_43880_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_43880_end_0 = const()[name = tensor<string, []>("op_43880_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_43880_end_mask_0 = const()[name = tensor<string, []>("op_43880_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43880_cast_fp16 = slice_by_index(begin = var_43880_begin_0, end = var_43880_end_0, end_mask = var_43880_end_mask_0, x = k_391_cast_fp16)[name = tensor<string, []>("op_43880_cast_fp16")];
            tensor<int32, [4]> var_43884_begin_0 = const()[name = tensor<string, []>("op_43884_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_43884_end_0 = const()[name = tensor<string, []>("op_43884_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_43884_end_mask_0 = const()[name = tensor<string, []>("op_43884_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43884_cast_fp16 = slice_by_index(begin = var_43884_begin_0, end = var_43884_end_0, end_mask = var_43884_end_mask_0, x = k_391_cast_fp16)[name = tensor<string, []>("op_43884_cast_fp16")];
            tensor<int32, [4]> var_43888_begin_0 = const()[name = tensor<string, []>("op_43888_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_43888_end_0 = const()[name = tensor<string, []>("op_43888_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_43888_end_mask_0 = const()[name = tensor<string, []>("op_43888_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43888_cast_fp16 = slice_by_index(begin = var_43888_begin_0, end = var_43888_end_0, end_mask = var_43888_end_mask_0, x = k_391_cast_fp16)[name = tensor<string, []>("op_43888_cast_fp16")];
            tensor<int32, [4]> var_43892_begin_0 = const()[name = tensor<string, []>("op_43892_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_43892_end_0 = const()[name = tensor<string, []>("op_43892_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_43892_end_mask_0 = const()[name = tensor<string, []>("op_43892_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43892_cast_fp16 = slice_by_index(begin = var_43892_begin_0, end = var_43892_end_0, end_mask = var_43892_end_mask_0, x = k_391_cast_fp16)[name = tensor<string, []>("op_43892_cast_fp16")];
            tensor<int32, [4]> var_43896_begin_0 = const()[name = tensor<string, []>("op_43896_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_43896_end_0 = const()[name = tensor<string, []>("op_43896_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_43896_end_mask_0 = const()[name = tensor<string, []>("op_43896_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43896_cast_fp16 = slice_by_index(begin = var_43896_begin_0, end = var_43896_end_0, end_mask = var_43896_end_mask_0, x = k_391_cast_fp16)[name = tensor<string, []>("op_43896_cast_fp16")];
            tensor<int32, [4]> var_43900_begin_0 = const()[name = tensor<string, []>("op_43900_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_43900_end_0 = const()[name = tensor<string, []>("op_43900_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_43900_end_mask_0 = const()[name = tensor<string, []>("op_43900_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43900_cast_fp16 = slice_by_index(begin = var_43900_begin_0, end = var_43900_end_0, end_mask = var_43900_end_mask_0, x = k_391_cast_fp16)[name = tensor<string, []>("op_43900_cast_fp16")];
            tensor<int32, [4]> var_43904_begin_0 = const()[name = tensor<string, []>("op_43904_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_43904_end_0 = const()[name = tensor<string, []>("op_43904_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_43904_end_mask_0 = const()[name = tensor<string, []>("op_43904_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43904_cast_fp16 = slice_by_index(begin = var_43904_begin_0, end = var_43904_end_0, end_mask = var_43904_end_mask_0, x = k_391_cast_fp16)[name = tensor<string, []>("op_43904_cast_fp16")];
            tensor<int32, [4]> var_43908_begin_0 = const()[name = tensor<string, []>("op_43908_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_43908_end_0 = const()[name = tensor<string, []>("op_43908_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_43908_end_mask_0 = const()[name = tensor<string, []>("op_43908_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43908_cast_fp16 = slice_by_index(begin = var_43908_begin_0, end = var_43908_end_0, end_mask = var_43908_end_mask_0, x = k_391_cast_fp16)[name = tensor<string, []>("op_43908_cast_fp16")];
            tensor<int32, [4]> var_43912_begin_0 = const()[name = tensor<string, []>("op_43912_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_43912_end_0 = const()[name = tensor<string, []>("op_43912_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_43912_end_mask_0 = const()[name = tensor<string, []>("op_43912_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43912_cast_fp16 = slice_by_index(begin = var_43912_begin_0, end = var_43912_end_0, end_mask = var_43912_end_mask_0, x = k_391_cast_fp16)[name = tensor<string, []>("op_43912_cast_fp16")];
            tensor<int32, [4]> var_43916_begin_0 = const()[name = tensor<string, []>("op_43916_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_43916_end_0 = const()[name = tensor<string, []>("op_43916_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_43916_end_mask_0 = const()[name = tensor<string, []>("op_43916_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43916_cast_fp16 = slice_by_index(begin = var_43916_begin_0, end = var_43916_end_0, end_mask = var_43916_end_mask_0, x = k_391_cast_fp16)[name = tensor<string, []>("op_43916_cast_fp16")];
            tensor<int32, [4]> var_43920_begin_0 = const()[name = tensor<string, []>("op_43920_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_43920_end_0 = const()[name = tensor<string, []>("op_43920_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_43920_end_mask_0 = const()[name = tensor<string, []>("op_43920_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43920_cast_fp16 = slice_by_index(begin = var_43920_begin_0, end = var_43920_end_0, end_mask = var_43920_end_mask_0, x = k_391_cast_fp16)[name = tensor<string, []>("op_43920_cast_fp16")];
            tensor<int32, [4]> var_43924_begin_0 = const()[name = tensor<string, []>("op_43924_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_43924_end_0 = const()[name = tensor<string, []>("op_43924_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_43924_end_mask_0 = const()[name = tensor<string, []>("op_43924_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43924_cast_fp16 = slice_by_index(begin = var_43924_begin_0, end = var_43924_end_0, end_mask = var_43924_end_mask_0, x = k_391_cast_fp16)[name = tensor<string, []>("op_43924_cast_fp16")];
            tensor<int32, [4]> var_43928_begin_0 = const()[name = tensor<string, []>("op_43928_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_43928_end_0 = const()[name = tensor<string, []>("op_43928_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_43928_end_mask_0 = const()[name = tensor<string, []>("op_43928_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43928_cast_fp16 = slice_by_index(begin = var_43928_begin_0, end = var_43928_end_0, end_mask = var_43928_end_mask_0, x = k_391_cast_fp16)[name = tensor<string, []>("op_43928_cast_fp16")];
            tensor<int32, [4]> var_43932_begin_0 = const()[name = tensor<string, []>("op_43932_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_43932_end_0 = const()[name = tensor<string, []>("op_43932_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_43932_end_mask_0 = const()[name = tensor<string, []>("op_43932_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43932_cast_fp16 = slice_by_index(begin = var_43932_begin_0, end = var_43932_end_0, end_mask = var_43932_end_mask_0, x = k_391_cast_fp16)[name = tensor<string, []>("op_43932_cast_fp16")];
            tensor<int32, [4]> var_43936_begin_0 = const()[name = tensor<string, []>("op_43936_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_43936_end_0 = const()[name = tensor<string, []>("op_43936_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_43936_end_mask_0 = const()[name = tensor<string, []>("op_43936_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43936_cast_fp16 = slice_by_index(begin = var_43936_begin_0, end = var_43936_end_0, end_mask = var_43936_end_mask_0, x = k_391_cast_fp16)[name = tensor<string, []>("op_43936_cast_fp16")];
            tensor<int32, [4]> var_43940_begin_0 = const()[name = tensor<string, []>("op_43940_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_43940_end_0 = const()[name = tensor<string, []>("op_43940_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_43940_end_mask_0 = const()[name = tensor<string, []>("op_43940_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43940_cast_fp16 = slice_by_index(begin = var_43940_begin_0, end = var_43940_end_0, end_mask = var_43940_end_mask_0, x = k_391_cast_fp16)[name = tensor<string, []>("op_43940_cast_fp16")];
            tensor<int32, [4]> var_43944_begin_0 = const()[name = tensor<string, []>("op_43944_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_43944_end_0 = const()[name = tensor<string, []>("op_43944_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_43944_end_mask_0 = const()[name = tensor<string, []>("op_43944_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43944_cast_fp16 = slice_by_index(begin = var_43944_begin_0, end = var_43944_end_0, end_mask = var_43944_end_mask_0, x = k_391_cast_fp16)[name = tensor<string, []>("op_43944_cast_fp16")];
            tensor<int32, [4]> var_43948_begin_0 = const()[name = tensor<string, []>("op_43948_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_43948_end_0 = const()[name = tensor<string, []>("op_43948_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_43948_end_mask_0 = const()[name = tensor<string, []>("op_43948_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43948_cast_fp16 = slice_by_index(begin = var_43948_begin_0, end = var_43948_end_0, end_mask = var_43948_end_mask_0, x = k_391_cast_fp16)[name = tensor<string, []>("op_43948_cast_fp16")];
            tensor<int32, [4]> var_43952_begin_0 = const()[name = tensor<string, []>("op_43952_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_43952_end_0 = const()[name = tensor<string, []>("op_43952_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_43952_end_mask_0 = const()[name = tensor<string, []>("op_43952_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43952_cast_fp16 = slice_by_index(begin = var_43952_begin_0, end = var_43952_end_0, end_mask = var_43952_end_mask_0, x = k_391_cast_fp16)[name = tensor<string, []>("op_43952_cast_fp16")];
            tensor<int32, [4]> var_43954_begin_0 = const()[name = tensor<string, []>("op_43954_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_43954_end_0 = const()[name = tensor<string, []>("op_43954_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_43954_end_mask_0 = const()[name = tensor<string, []>("op_43954_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43954_cast_fp16 = slice_by_index(begin = var_43954_begin_0, end = var_43954_end_0, end_mask = var_43954_end_mask_0, x = v_195_cast_fp16)[name = tensor<string, []>("op_43954_cast_fp16")];
            tensor<int32, [4]> var_43958_begin_0 = const()[name = tensor<string, []>("op_43958_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_43958_end_0 = const()[name = tensor<string, []>("op_43958_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_43958_end_mask_0 = const()[name = tensor<string, []>("op_43958_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43958_cast_fp16 = slice_by_index(begin = var_43958_begin_0, end = var_43958_end_0, end_mask = var_43958_end_mask_0, x = v_195_cast_fp16)[name = tensor<string, []>("op_43958_cast_fp16")];
            tensor<int32, [4]> var_43962_begin_0 = const()[name = tensor<string, []>("op_43962_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_43962_end_0 = const()[name = tensor<string, []>("op_43962_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_43962_end_mask_0 = const()[name = tensor<string, []>("op_43962_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43962_cast_fp16 = slice_by_index(begin = var_43962_begin_0, end = var_43962_end_0, end_mask = var_43962_end_mask_0, x = v_195_cast_fp16)[name = tensor<string, []>("op_43962_cast_fp16")];
            tensor<int32, [4]> var_43966_begin_0 = const()[name = tensor<string, []>("op_43966_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_43966_end_0 = const()[name = tensor<string, []>("op_43966_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_43966_end_mask_0 = const()[name = tensor<string, []>("op_43966_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43966_cast_fp16 = slice_by_index(begin = var_43966_begin_0, end = var_43966_end_0, end_mask = var_43966_end_mask_0, x = v_195_cast_fp16)[name = tensor<string, []>("op_43966_cast_fp16")];
            tensor<int32, [4]> var_43970_begin_0 = const()[name = tensor<string, []>("op_43970_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_43970_end_0 = const()[name = tensor<string, []>("op_43970_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_43970_end_mask_0 = const()[name = tensor<string, []>("op_43970_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43970_cast_fp16 = slice_by_index(begin = var_43970_begin_0, end = var_43970_end_0, end_mask = var_43970_end_mask_0, x = v_195_cast_fp16)[name = tensor<string, []>("op_43970_cast_fp16")];
            tensor<int32, [4]> var_43974_begin_0 = const()[name = tensor<string, []>("op_43974_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_43974_end_0 = const()[name = tensor<string, []>("op_43974_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_43974_end_mask_0 = const()[name = tensor<string, []>("op_43974_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43974_cast_fp16 = slice_by_index(begin = var_43974_begin_0, end = var_43974_end_0, end_mask = var_43974_end_mask_0, x = v_195_cast_fp16)[name = tensor<string, []>("op_43974_cast_fp16")];
            tensor<int32, [4]> var_43978_begin_0 = const()[name = tensor<string, []>("op_43978_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_43978_end_0 = const()[name = tensor<string, []>("op_43978_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_43978_end_mask_0 = const()[name = tensor<string, []>("op_43978_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43978_cast_fp16 = slice_by_index(begin = var_43978_begin_0, end = var_43978_end_0, end_mask = var_43978_end_mask_0, x = v_195_cast_fp16)[name = tensor<string, []>("op_43978_cast_fp16")];
            tensor<int32, [4]> var_43982_begin_0 = const()[name = tensor<string, []>("op_43982_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_43982_end_0 = const()[name = tensor<string, []>("op_43982_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_43982_end_mask_0 = const()[name = tensor<string, []>("op_43982_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43982_cast_fp16 = slice_by_index(begin = var_43982_begin_0, end = var_43982_end_0, end_mask = var_43982_end_mask_0, x = v_195_cast_fp16)[name = tensor<string, []>("op_43982_cast_fp16")];
            tensor<int32, [4]> var_43986_begin_0 = const()[name = tensor<string, []>("op_43986_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_43986_end_0 = const()[name = tensor<string, []>("op_43986_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_43986_end_mask_0 = const()[name = tensor<string, []>("op_43986_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43986_cast_fp16 = slice_by_index(begin = var_43986_begin_0, end = var_43986_end_0, end_mask = var_43986_end_mask_0, x = v_195_cast_fp16)[name = tensor<string, []>("op_43986_cast_fp16")];
            tensor<int32, [4]> var_43990_begin_0 = const()[name = tensor<string, []>("op_43990_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_43990_end_0 = const()[name = tensor<string, []>("op_43990_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_43990_end_mask_0 = const()[name = tensor<string, []>("op_43990_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43990_cast_fp16 = slice_by_index(begin = var_43990_begin_0, end = var_43990_end_0, end_mask = var_43990_end_mask_0, x = v_195_cast_fp16)[name = tensor<string, []>("op_43990_cast_fp16")];
            tensor<int32, [4]> var_43994_begin_0 = const()[name = tensor<string, []>("op_43994_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_43994_end_0 = const()[name = tensor<string, []>("op_43994_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_43994_end_mask_0 = const()[name = tensor<string, []>("op_43994_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43994_cast_fp16 = slice_by_index(begin = var_43994_begin_0, end = var_43994_end_0, end_mask = var_43994_end_mask_0, x = v_195_cast_fp16)[name = tensor<string, []>("op_43994_cast_fp16")];
            tensor<int32, [4]> var_43998_begin_0 = const()[name = tensor<string, []>("op_43998_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_43998_end_0 = const()[name = tensor<string, []>("op_43998_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_43998_end_mask_0 = const()[name = tensor<string, []>("op_43998_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43998_cast_fp16 = slice_by_index(begin = var_43998_begin_0, end = var_43998_end_0, end_mask = var_43998_end_mask_0, x = v_195_cast_fp16)[name = tensor<string, []>("op_43998_cast_fp16")];
            tensor<int32, [4]> var_44002_begin_0 = const()[name = tensor<string, []>("op_44002_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_44002_end_0 = const()[name = tensor<string, []>("op_44002_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_44002_end_mask_0 = const()[name = tensor<string, []>("op_44002_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44002_cast_fp16 = slice_by_index(begin = var_44002_begin_0, end = var_44002_end_0, end_mask = var_44002_end_mask_0, x = v_195_cast_fp16)[name = tensor<string, []>("op_44002_cast_fp16")];
            tensor<int32, [4]> var_44006_begin_0 = const()[name = tensor<string, []>("op_44006_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_44006_end_0 = const()[name = tensor<string, []>("op_44006_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_44006_end_mask_0 = const()[name = tensor<string, []>("op_44006_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44006_cast_fp16 = slice_by_index(begin = var_44006_begin_0, end = var_44006_end_0, end_mask = var_44006_end_mask_0, x = v_195_cast_fp16)[name = tensor<string, []>("op_44006_cast_fp16")];
            tensor<int32, [4]> var_44010_begin_0 = const()[name = tensor<string, []>("op_44010_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_44010_end_0 = const()[name = tensor<string, []>("op_44010_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_44010_end_mask_0 = const()[name = tensor<string, []>("op_44010_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44010_cast_fp16 = slice_by_index(begin = var_44010_begin_0, end = var_44010_end_0, end_mask = var_44010_end_mask_0, x = v_195_cast_fp16)[name = tensor<string, []>("op_44010_cast_fp16")];
            tensor<int32, [4]> var_44014_begin_0 = const()[name = tensor<string, []>("op_44014_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_44014_end_0 = const()[name = tensor<string, []>("op_44014_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_44014_end_mask_0 = const()[name = tensor<string, []>("op_44014_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44014_cast_fp16 = slice_by_index(begin = var_44014_begin_0, end = var_44014_end_0, end_mask = var_44014_end_mask_0, x = v_195_cast_fp16)[name = tensor<string, []>("op_44014_cast_fp16")];
            tensor<int32, [4]> var_44018_begin_0 = const()[name = tensor<string, []>("op_44018_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_44018_end_0 = const()[name = tensor<string, []>("op_44018_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_44018_end_mask_0 = const()[name = tensor<string, []>("op_44018_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44018_cast_fp16 = slice_by_index(begin = var_44018_begin_0, end = var_44018_end_0, end_mask = var_44018_end_mask_0, x = v_195_cast_fp16)[name = tensor<string, []>("op_44018_cast_fp16")];
            tensor<int32, [4]> var_44022_begin_0 = const()[name = tensor<string, []>("op_44022_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_44022_end_0 = const()[name = tensor<string, []>("op_44022_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_44022_end_mask_0 = const()[name = tensor<string, []>("op_44022_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44022_cast_fp16 = slice_by_index(begin = var_44022_begin_0, end = var_44022_end_0, end_mask = var_44022_end_mask_0, x = v_195_cast_fp16)[name = tensor<string, []>("op_44022_cast_fp16")];
            tensor<int32, [4]> var_44026_begin_0 = const()[name = tensor<string, []>("op_44026_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_44026_end_0 = const()[name = tensor<string, []>("op_44026_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_44026_end_mask_0 = const()[name = tensor<string, []>("op_44026_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44026_cast_fp16 = slice_by_index(begin = var_44026_begin_0, end = var_44026_end_0, end_mask = var_44026_end_mask_0, x = v_195_cast_fp16)[name = tensor<string, []>("op_44026_cast_fp16")];
            tensor<int32, [4]> var_44030_begin_0 = const()[name = tensor<string, []>("op_44030_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_44030_end_0 = const()[name = tensor<string, []>("op_44030_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_44030_end_mask_0 = const()[name = tensor<string, []>("op_44030_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44030_cast_fp16 = slice_by_index(begin = var_44030_begin_0, end = var_44030_end_0, end_mask = var_44030_end_mask_0, x = v_195_cast_fp16)[name = tensor<string, []>("op_44030_cast_fp16")];
            tensor<string, []> var_44034_equation_0 = const()[name = tensor<string, []>("op_44034_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44034_cast_fp16 = einsum(equation = var_44034_equation_0, values = (var_43876_cast_fp16, var_43793_cast_fp16))[name = tensor<string, []>("op_44034_cast_fp16")];
            tensor<fp16, []> var_44035_to_fp16 = const()[name = tensor<string, []>("op_44035_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3721_cast_fp16 = mul(x = var_44034_cast_fp16, y = var_44035_to_fp16)[name = tensor<string, []>("aw_3721_cast_fp16")];
            tensor<string, []> var_44038_equation_0 = const()[name = tensor<string, []>("op_44038_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44038_cast_fp16 = einsum(equation = var_44038_equation_0, values = (var_43880_cast_fp16, var_43797_cast_fp16))[name = tensor<string, []>("op_44038_cast_fp16")];
            tensor<fp16, []> var_44039_to_fp16 = const()[name = tensor<string, []>("op_44039_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3723_cast_fp16 = mul(x = var_44038_cast_fp16, y = var_44039_to_fp16)[name = tensor<string, []>("aw_3723_cast_fp16")];
            tensor<string, []> var_44042_equation_0 = const()[name = tensor<string, []>("op_44042_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44042_cast_fp16 = einsum(equation = var_44042_equation_0, values = (var_43884_cast_fp16, var_43801_cast_fp16))[name = tensor<string, []>("op_44042_cast_fp16")];
            tensor<fp16, []> var_44043_to_fp16 = const()[name = tensor<string, []>("op_44043_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3725_cast_fp16 = mul(x = var_44042_cast_fp16, y = var_44043_to_fp16)[name = tensor<string, []>("aw_3725_cast_fp16")];
            tensor<string, []> var_44046_equation_0 = const()[name = tensor<string, []>("op_44046_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44046_cast_fp16 = einsum(equation = var_44046_equation_0, values = (var_43888_cast_fp16, var_43805_cast_fp16))[name = tensor<string, []>("op_44046_cast_fp16")];
            tensor<fp16, []> var_44047_to_fp16 = const()[name = tensor<string, []>("op_44047_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3727_cast_fp16 = mul(x = var_44046_cast_fp16, y = var_44047_to_fp16)[name = tensor<string, []>("aw_3727_cast_fp16")];
            tensor<string, []> var_44050_equation_0 = const()[name = tensor<string, []>("op_44050_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44050_cast_fp16 = einsum(equation = var_44050_equation_0, values = (var_43892_cast_fp16, var_43809_cast_fp16))[name = tensor<string, []>("op_44050_cast_fp16")];
            tensor<fp16, []> var_44051_to_fp16 = const()[name = tensor<string, []>("op_44051_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3729_cast_fp16 = mul(x = var_44050_cast_fp16, y = var_44051_to_fp16)[name = tensor<string, []>("aw_3729_cast_fp16")];
            tensor<string, []> var_44054_equation_0 = const()[name = tensor<string, []>("op_44054_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44054_cast_fp16 = einsum(equation = var_44054_equation_0, values = (var_43896_cast_fp16, var_43813_cast_fp16))[name = tensor<string, []>("op_44054_cast_fp16")];
            tensor<fp16, []> var_44055_to_fp16 = const()[name = tensor<string, []>("op_44055_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3731_cast_fp16 = mul(x = var_44054_cast_fp16, y = var_44055_to_fp16)[name = tensor<string, []>("aw_3731_cast_fp16")];
            tensor<string, []> var_44058_equation_0 = const()[name = tensor<string, []>("op_44058_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44058_cast_fp16 = einsum(equation = var_44058_equation_0, values = (var_43900_cast_fp16, var_43817_cast_fp16))[name = tensor<string, []>("op_44058_cast_fp16")];
            tensor<fp16, []> var_44059_to_fp16 = const()[name = tensor<string, []>("op_44059_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3733_cast_fp16 = mul(x = var_44058_cast_fp16, y = var_44059_to_fp16)[name = tensor<string, []>("aw_3733_cast_fp16")];
            tensor<string, []> var_44062_equation_0 = const()[name = tensor<string, []>("op_44062_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44062_cast_fp16 = einsum(equation = var_44062_equation_0, values = (var_43904_cast_fp16, var_43821_cast_fp16))[name = tensor<string, []>("op_44062_cast_fp16")];
            tensor<fp16, []> var_44063_to_fp16 = const()[name = tensor<string, []>("op_44063_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3735_cast_fp16 = mul(x = var_44062_cast_fp16, y = var_44063_to_fp16)[name = tensor<string, []>("aw_3735_cast_fp16")];
            tensor<string, []> var_44066_equation_0 = const()[name = tensor<string, []>("op_44066_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44066_cast_fp16 = einsum(equation = var_44066_equation_0, values = (var_43908_cast_fp16, var_43825_cast_fp16))[name = tensor<string, []>("op_44066_cast_fp16")];
            tensor<fp16, []> var_44067_to_fp16 = const()[name = tensor<string, []>("op_44067_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3737_cast_fp16 = mul(x = var_44066_cast_fp16, y = var_44067_to_fp16)[name = tensor<string, []>("aw_3737_cast_fp16")];
            tensor<string, []> var_44070_equation_0 = const()[name = tensor<string, []>("op_44070_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44070_cast_fp16 = einsum(equation = var_44070_equation_0, values = (var_43912_cast_fp16, var_43829_cast_fp16))[name = tensor<string, []>("op_44070_cast_fp16")];
            tensor<fp16, []> var_44071_to_fp16 = const()[name = tensor<string, []>("op_44071_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3739_cast_fp16 = mul(x = var_44070_cast_fp16, y = var_44071_to_fp16)[name = tensor<string, []>("aw_3739_cast_fp16")];
            tensor<string, []> var_44074_equation_0 = const()[name = tensor<string, []>("op_44074_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44074_cast_fp16 = einsum(equation = var_44074_equation_0, values = (var_43916_cast_fp16, var_43833_cast_fp16))[name = tensor<string, []>("op_44074_cast_fp16")];
            tensor<fp16, []> var_44075_to_fp16 = const()[name = tensor<string, []>("op_44075_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3741_cast_fp16 = mul(x = var_44074_cast_fp16, y = var_44075_to_fp16)[name = tensor<string, []>("aw_3741_cast_fp16")];
            tensor<string, []> var_44078_equation_0 = const()[name = tensor<string, []>("op_44078_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44078_cast_fp16 = einsum(equation = var_44078_equation_0, values = (var_43920_cast_fp16, var_43837_cast_fp16))[name = tensor<string, []>("op_44078_cast_fp16")];
            tensor<fp16, []> var_44079_to_fp16 = const()[name = tensor<string, []>("op_44079_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3743_cast_fp16 = mul(x = var_44078_cast_fp16, y = var_44079_to_fp16)[name = tensor<string, []>("aw_3743_cast_fp16")];
            tensor<string, []> var_44082_equation_0 = const()[name = tensor<string, []>("op_44082_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44082_cast_fp16 = einsum(equation = var_44082_equation_0, values = (var_43924_cast_fp16, var_43841_cast_fp16))[name = tensor<string, []>("op_44082_cast_fp16")];
            tensor<fp16, []> var_44083_to_fp16 = const()[name = tensor<string, []>("op_44083_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3745_cast_fp16 = mul(x = var_44082_cast_fp16, y = var_44083_to_fp16)[name = tensor<string, []>("aw_3745_cast_fp16")];
            tensor<string, []> var_44086_equation_0 = const()[name = tensor<string, []>("op_44086_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44086_cast_fp16 = einsum(equation = var_44086_equation_0, values = (var_43928_cast_fp16, var_43845_cast_fp16))[name = tensor<string, []>("op_44086_cast_fp16")];
            tensor<fp16, []> var_44087_to_fp16 = const()[name = tensor<string, []>("op_44087_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3747_cast_fp16 = mul(x = var_44086_cast_fp16, y = var_44087_to_fp16)[name = tensor<string, []>("aw_3747_cast_fp16")];
            tensor<string, []> var_44090_equation_0 = const()[name = tensor<string, []>("op_44090_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44090_cast_fp16 = einsum(equation = var_44090_equation_0, values = (var_43932_cast_fp16, var_43849_cast_fp16))[name = tensor<string, []>("op_44090_cast_fp16")];
            tensor<fp16, []> var_44091_to_fp16 = const()[name = tensor<string, []>("op_44091_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3749_cast_fp16 = mul(x = var_44090_cast_fp16, y = var_44091_to_fp16)[name = tensor<string, []>("aw_3749_cast_fp16")];
            tensor<string, []> var_44094_equation_0 = const()[name = tensor<string, []>("op_44094_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44094_cast_fp16 = einsum(equation = var_44094_equation_0, values = (var_43936_cast_fp16, var_43853_cast_fp16))[name = tensor<string, []>("op_44094_cast_fp16")];
            tensor<fp16, []> var_44095_to_fp16 = const()[name = tensor<string, []>("op_44095_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3751_cast_fp16 = mul(x = var_44094_cast_fp16, y = var_44095_to_fp16)[name = tensor<string, []>("aw_3751_cast_fp16")];
            tensor<string, []> var_44098_equation_0 = const()[name = tensor<string, []>("op_44098_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44098_cast_fp16 = einsum(equation = var_44098_equation_0, values = (var_43940_cast_fp16, var_43857_cast_fp16))[name = tensor<string, []>("op_44098_cast_fp16")];
            tensor<fp16, []> var_44099_to_fp16 = const()[name = tensor<string, []>("op_44099_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3753_cast_fp16 = mul(x = var_44098_cast_fp16, y = var_44099_to_fp16)[name = tensor<string, []>("aw_3753_cast_fp16")];
            tensor<string, []> var_44102_equation_0 = const()[name = tensor<string, []>("op_44102_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44102_cast_fp16 = einsum(equation = var_44102_equation_0, values = (var_43944_cast_fp16, var_43861_cast_fp16))[name = tensor<string, []>("op_44102_cast_fp16")];
            tensor<fp16, []> var_44103_to_fp16 = const()[name = tensor<string, []>("op_44103_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3755_cast_fp16 = mul(x = var_44102_cast_fp16, y = var_44103_to_fp16)[name = tensor<string, []>("aw_3755_cast_fp16")];
            tensor<string, []> var_44106_equation_0 = const()[name = tensor<string, []>("op_44106_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44106_cast_fp16 = einsum(equation = var_44106_equation_0, values = (var_43948_cast_fp16, var_43865_cast_fp16))[name = tensor<string, []>("op_44106_cast_fp16")];
            tensor<fp16, []> var_44107_to_fp16 = const()[name = tensor<string, []>("op_44107_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3757_cast_fp16 = mul(x = var_44106_cast_fp16, y = var_44107_to_fp16)[name = tensor<string, []>("aw_3757_cast_fp16")];
            tensor<string, []> var_44110_equation_0 = const()[name = tensor<string, []>("op_44110_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44110_cast_fp16 = einsum(equation = var_44110_equation_0, values = (var_43952_cast_fp16, var_43869_cast_fp16))[name = tensor<string, []>("op_44110_cast_fp16")];
            tensor<fp16, []> var_44111_to_fp16 = const()[name = tensor<string, []>("op_44111_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3759_cast_fp16 = mul(x = var_44110_cast_fp16, y = var_44111_to_fp16)[name = tensor<string, []>("aw_3759_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_44113_cast_fp16 = softmax(axis = var_30355, x = aw_3721_cast_fp16)[name = tensor<string, []>("op_44113_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_44114_cast_fp16 = softmax(axis = var_30355, x = aw_3723_cast_fp16)[name = tensor<string, []>("op_44114_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_44115_cast_fp16 = softmax(axis = var_30355, x = aw_3725_cast_fp16)[name = tensor<string, []>("op_44115_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_44116_cast_fp16 = softmax(axis = var_30355, x = aw_3727_cast_fp16)[name = tensor<string, []>("op_44116_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_44117_cast_fp16 = softmax(axis = var_30355, x = aw_3729_cast_fp16)[name = tensor<string, []>("op_44117_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_44118_cast_fp16 = softmax(axis = var_30355, x = aw_3731_cast_fp16)[name = tensor<string, []>("op_44118_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_44119_cast_fp16 = softmax(axis = var_30355, x = aw_3733_cast_fp16)[name = tensor<string, []>("op_44119_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_44120_cast_fp16 = softmax(axis = var_30355, x = aw_3735_cast_fp16)[name = tensor<string, []>("op_44120_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_44121_cast_fp16 = softmax(axis = var_30355, x = aw_3737_cast_fp16)[name = tensor<string, []>("op_44121_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_44122_cast_fp16 = softmax(axis = var_30355, x = aw_3739_cast_fp16)[name = tensor<string, []>("op_44122_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_44123_cast_fp16 = softmax(axis = var_30355, x = aw_3741_cast_fp16)[name = tensor<string, []>("op_44123_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_44124_cast_fp16 = softmax(axis = var_30355, x = aw_3743_cast_fp16)[name = tensor<string, []>("op_44124_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_44125_cast_fp16 = softmax(axis = var_30355, x = aw_3745_cast_fp16)[name = tensor<string, []>("op_44125_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_44126_cast_fp16 = softmax(axis = var_30355, x = aw_3747_cast_fp16)[name = tensor<string, []>("op_44126_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_44127_cast_fp16 = softmax(axis = var_30355, x = aw_3749_cast_fp16)[name = tensor<string, []>("op_44127_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_44128_cast_fp16 = softmax(axis = var_30355, x = aw_3751_cast_fp16)[name = tensor<string, []>("op_44128_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_44129_cast_fp16 = softmax(axis = var_30355, x = aw_3753_cast_fp16)[name = tensor<string, []>("op_44129_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_44130_cast_fp16 = softmax(axis = var_30355, x = aw_3755_cast_fp16)[name = tensor<string, []>("op_44130_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_44131_cast_fp16 = softmax(axis = var_30355, x = aw_3757_cast_fp16)[name = tensor<string, []>("op_44131_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_44132_cast_fp16 = softmax(axis = var_30355, x = aw_3759_cast_fp16)[name = tensor<string, []>("op_44132_cast_fp16")];
            tensor<string, []> var_44134_equation_0 = const()[name = tensor<string, []>("op_44134_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44134_cast_fp16 = einsum(equation = var_44134_equation_0, values = (var_43954_cast_fp16, var_44113_cast_fp16))[name = tensor<string, []>("op_44134_cast_fp16")];
            tensor<string, []> var_44136_equation_0 = const()[name = tensor<string, []>("op_44136_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44136_cast_fp16 = einsum(equation = var_44136_equation_0, values = (var_43958_cast_fp16, var_44114_cast_fp16))[name = tensor<string, []>("op_44136_cast_fp16")];
            tensor<string, []> var_44138_equation_0 = const()[name = tensor<string, []>("op_44138_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44138_cast_fp16 = einsum(equation = var_44138_equation_0, values = (var_43962_cast_fp16, var_44115_cast_fp16))[name = tensor<string, []>("op_44138_cast_fp16")];
            tensor<string, []> var_44140_equation_0 = const()[name = tensor<string, []>("op_44140_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44140_cast_fp16 = einsum(equation = var_44140_equation_0, values = (var_43966_cast_fp16, var_44116_cast_fp16))[name = tensor<string, []>("op_44140_cast_fp16")];
            tensor<string, []> var_44142_equation_0 = const()[name = tensor<string, []>("op_44142_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44142_cast_fp16 = einsum(equation = var_44142_equation_0, values = (var_43970_cast_fp16, var_44117_cast_fp16))[name = tensor<string, []>("op_44142_cast_fp16")];
            tensor<string, []> var_44144_equation_0 = const()[name = tensor<string, []>("op_44144_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44144_cast_fp16 = einsum(equation = var_44144_equation_0, values = (var_43974_cast_fp16, var_44118_cast_fp16))[name = tensor<string, []>("op_44144_cast_fp16")];
            tensor<string, []> var_44146_equation_0 = const()[name = tensor<string, []>("op_44146_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44146_cast_fp16 = einsum(equation = var_44146_equation_0, values = (var_43978_cast_fp16, var_44119_cast_fp16))[name = tensor<string, []>("op_44146_cast_fp16")];
            tensor<string, []> var_44148_equation_0 = const()[name = tensor<string, []>("op_44148_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44148_cast_fp16 = einsum(equation = var_44148_equation_0, values = (var_43982_cast_fp16, var_44120_cast_fp16))[name = tensor<string, []>("op_44148_cast_fp16")];
            tensor<string, []> var_44150_equation_0 = const()[name = tensor<string, []>("op_44150_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44150_cast_fp16 = einsum(equation = var_44150_equation_0, values = (var_43986_cast_fp16, var_44121_cast_fp16))[name = tensor<string, []>("op_44150_cast_fp16")];
            tensor<string, []> var_44152_equation_0 = const()[name = tensor<string, []>("op_44152_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44152_cast_fp16 = einsum(equation = var_44152_equation_0, values = (var_43990_cast_fp16, var_44122_cast_fp16))[name = tensor<string, []>("op_44152_cast_fp16")];
            tensor<string, []> var_44154_equation_0 = const()[name = tensor<string, []>("op_44154_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44154_cast_fp16 = einsum(equation = var_44154_equation_0, values = (var_43994_cast_fp16, var_44123_cast_fp16))[name = tensor<string, []>("op_44154_cast_fp16")];
            tensor<string, []> var_44156_equation_0 = const()[name = tensor<string, []>("op_44156_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44156_cast_fp16 = einsum(equation = var_44156_equation_0, values = (var_43998_cast_fp16, var_44124_cast_fp16))[name = tensor<string, []>("op_44156_cast_fp16")];
            tensor<string, []> var_44158_equation_0 = const()[name = tensor<string, []>("op_44158_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44158_cast_fp16 = einsum(equation = var_44158_equation_0, values = (var_44002_cast_fp16, var_44125_cast_fp16))[name = tensor<string, []>("op_44158_cast_fp16")];
            tensor<string, []> var_44160_equation_0 = const()[name = tensor<string, []>("op_44160_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44160_cast_fp16 = einsum(equation = var_44160_equation_0, values = (var_44006_cast_fp16, var_44126_cast_fp16))[name = tensor<string, []>("op_44160_cast_fp16")];
            tensor<string, []> var_44162_equation_0 = const()[name = tensor<string, []>("op_44162_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44162_cast_fp16 = einsum(equation = var_44162_equation_0, values = (var_44010_cast_fp16, var_44127_cast_fp16))[name = tensor<string, []>("op_44162_cast_fp16")];
            tensor<string, []> var_44164_equation_0 = const()[name = tensor<string, []>("op_44164_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44164_cast_fp16 = einsum(equation = var_44164_equation_0, values = (var_44014_cast_fp16, var_44128_cast_fp16))[name = tensor<string, []>("op_44164_cast_fp16")];
            tensor<string, []> var_44166_equation_0 = const()[name = tensor<string, []>("op_44166_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44166_cast_fp16 = einsum(equation = var_44166_equation_0, values = (var_44018_cast_fp16, var_44129_cast_fp16))[name = tensor<string, []>("op_44166_cast_fp16")];
            tensor<string, []> var_44168_equation_0 = const()[name = tensor<string, []>("op_44168_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44168_cast_fp16 = einsum(equation = var_44168_equation_0, values = (var_44022_cast_fp16, var_44130_cast_fp16))[name = tensor<string, []>("op_44168_cast_fp16")];
            tensor<string, []> var_44170_equation_0 = const()[name = tensor<string, []>("op_44170_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44170_cast_fp16 = einsum(equation = var_44170_equation_0, values = (var_44026_cast_fp16, var_44131_cast_fp16))[name = tensor<string, []>("op_44170_cast_fp16")];
            tensor<string, []> var_44172_equation_0 = const()[name = tensor<string, []>("op_44172_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44172_cast_fp16 = einsum(equation = var_44172_equation_0, values = (var_44030_cast_fp16, var_44132_cast_fp16))[name = tensor<string, []>("op_44172_cast_fp16")];
            tensor<bool, []> input_571_interleave_0 = const()[name = tensor<string, []>("input_571_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_571_cast_fp16 = concat(axis = var_30355, interleave = input_571_interleave_0, values = (var_44134_cast_fp16, var_44136_cast_fp16, var_44138_cast_fp16, var_44140_cast_fp16, var_44142_cast_fp16, var_44144_cast_fp16, var_44146_cast_fp16, var_44148_cast_fp16, var_44150_cast_fp16, var_44152_cast_fp16, var_44154_cast_fp16, var_44156_cast_fp16, var_44158_cast_fp16, var_44160_cast_fp16, var_44162_cast_fp16, var_44164_cast_fp16, var_44166_cast_fp16, var_44168_cast_fp16, var_44170_cast_fp16, var_44172_cast_fp16))[name = tensor<string, []>("input_571_cast_fp16")];
            tensor<string, []> var_44182_pad_type_0 = const()[name = tensor<string, []>("op_44182_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_44182_strides_0 = const()[name = tensor<string, []>("op_44182_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_44182_pad_0 = const()[name = tensor<string, []>("op_44182_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_44182_dilations_0 = const()[name = tensor<string, []>("op_44182_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_44182_groups_0 = const()[name = tensor<string, []>("op_44182_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_4_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(418529920))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(419758784))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_4_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_4_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_4_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(419758976)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_44182_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_4_attn2_to_out_0_bias_to_fp16, dilations = var_44182_dilations_0, groups = var_44182_groups_0, pad = var_44182_pad_0, pad_type = var_44182_pad_type_0, strides = var_44182_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_4_attn2_to_out_0_weight_to_fp16_palettized, x = input_571_cast_fp16)[name = tensor<string, []>("op_44182_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_293_cast_fp16 = add(x = var_44182_cast_fp16, y = inputs_291_cast_fp16)[name = tensor<string, []>("inputs_293_cast_fp16")];
            tensor<int32, [1]> input_573_axes_0 = const()[name = tensor<string, []>("input_573_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> input_573_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_573_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(419761600)))];
            tensor<fp16, [1280]> input_573_beta_0_to_fp16 = const()[name = tensor<string, []>("input_573_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(419764224)))];
            tensor<fp16, []> var_44192_to_fp16 = const()[name = tensor<string, []>("op_44192_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> input_573_cast_fp16 = layer_norm(axes = input_573_axes_0, beta = input_573_beta_0_to_fp16, epsilon = var_44192_to_fp16, gamma = input_573_gamma_0_to_fp16, x = inputs_293_cast_fp16)[name = tensor<string, []>("input_573_cast_fp16")];
            tensor<string, []> var_44212_pad_type_0 = const()[name = tensor<string, []>("op_44212_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_44212_strides_0 = const()[name = tensor<string, []>("op_44212_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_44212_pad_0 = const()[name = tensor<string, []>("op_44212_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_44212_dilations_0 = const()[name = tensor<string, []>("op_44212_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_44212_groups_0 = const()[name = tensor<string, []>("op_44212_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_4_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [9830400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(419766848))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(429597312))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_4_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([10240, 1280, 1, 1])];
            tensor<fp16, [10240]> up_blocks_0_attentions_1_transformer_blocks_4_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_4_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(429597504)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_44212_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_4_ff_net_0_proj_bias_to_fp16, dilations = var_44212_dilations_0, groups = var_44212_groups_0, pad = var_44212_pad_0, pad_type = var_44212_pad_type_0, strides = var_44212_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_4_ff_net_0_proj_weight_to_fp16_palettized, x = input_573_cast_fp16)[name = tensor<string, []>("op_44212_cast_fp16")];
            tensor<int32, [2]> var_44213_split_sizes_0 = const()[name = tensor<string, []>("op_44213_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_44213_axis_0 = const()[name = tensor<string, []>("op_44213_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_44213_cast_fp16_0, tensor<fp16, [2, 5120, 1, 1024]> var_44213_cast_fp16_1 = split(axis = var_44213_axis_0, split_sizes = var_44213_split_sizes_0, x = var_44212_cast_fp16)[name = tensor<string, []>("op_44213_cast_fp16")];
            tensor<string, []> var_44215_mode_0 = const()[name = tensor<string, []>("op_44215_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_44215_cast_fp16 = gelu(mode = var_44215_mode_0, x = var_44213_cast_fp16_1)[name = tensor<string, []>("op_44215_cast_fp16")];
            tensor<fp16, [2, 5120, 1, 1024]> input_575_cast_fp16 = mul(x = var_44213_cast_fp16_0, y = var_44215_cast_fp16)[name = tensor<string, []>("input_575_cast_fp16")];
            tensor<string, []> var_44223_pad_type_0 = const()[name = tensor<string, []>("op_44223_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_44223_strides_0 = const()[name = tensor<string, []>("op_44223_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_44223_pad_0 = const()[name = tensor<string, []>("op_44223_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_44223_dilations_0 = const()[name = tensor<string, []>("op_44223_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_44223_groups_0 = const()[name = tensor<string, []>("op_44223_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_4_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [4915200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(429618048))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(434533312))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_4_ff_net_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 5120, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_4_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_4_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(434533504)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_44223_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_4_ff_net_2_bias_to_fp16, dilations = var_44223_dilations_0, groups = var_44223_groups_0, pad = var_44223_pad_0, pad_type = var_44223_pad_type_0, strides = var_44223_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_4_ff_net_2_weight_to_fp16_palettized, x = input_575_cast_fp16)[name = tensor<string, []>("op_44223_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_295_cast_fp16 = add(x = var_44223_cast_fp16, y = inputs_293_cast_fp16)[name = tensor<string, []>("inputs_295_cast_fp16")];
            tensor<int32, [1]> hidden_states_391_axes_0 = const()[name = tensor<string, []>("hidden_states_391_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_391_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_391_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(434536128)))];
            tensor<fp16, [1280]> hidden_states_391_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_391_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(434538752)))];
            tensor<fp16, []> var_44239_to_fp16 = const()[name = tensor<string, []>("op_44239_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_391_cast_fp16 = layer_norm(axes = hidden_states_391_axes_0, beta = hidden_states_391_beta_0_to_fp16, epsilon = var_44239_to_fp16, gamma = hidden_states_391_gamma_0_to_fp16, x = inputs_295_cast_fp16)[name = tensor<string, []>("hidden_states_391_cast_fp16")];
            tensor<string, []> q_197_pad_type_0 = const()[name = tensor<string, []>("q_197_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_197_strides_0 = const()[name = tensor<string, []>("q_197_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_197_pad_0 = const()[name = tensor<string, []>("q_197_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_197_dilations_0 = const()[name = tensor<string, []>("q_197_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_197_groups_0 = const()[name = tensor<string, []>("q_197_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_5_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(434541376))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(435770240))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_5_attn1_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_197_cast_fp16 = conv(dilations = q_197_dilations_0, groups = q_197_groups_0, pad = q_197_pad_0, pad_type = q_197_pad_type_0, strides = q_197_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_5_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_391_cast_fp16)[name = tensor<string, []>("q_197_cast_fp16")];
            tensor<string, []> k_393_pad_type_0 = const()[name = tensor<string, []>("k_393_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_393_strides_0 = const()[name = tensor<string, []>("k_393_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_393_pad_0 = const()[name = tensor<string, []>("k_393_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_393_dilations_0 = const()[name = tensor<string, []>("k_393_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_393_groups_0 = const()[name = tensor<string, []>("k_393_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_5_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(435770432))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(436999296))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_5_attn1_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> k_393_cast_fp16 = conv(dilations = k_393_dilations_0, groups = k_393_groups_0, pad = k_393_pad_0, pad_type = k_393_pad_type_0, strides = k_393_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_5_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_391_cast_fp16)[name = tensor<string, []>("k_393_cast_fp16")];
            tensor<string, []> v_197_pad_type_0 = const()[name = tensor<string, []>("v_197_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_197_strides_0 = const()[name = tensor<string, []>("v_197_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_197_pad_0 = const()[name = tensor<string, []>("v_197_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_197_dilations_0 = const()[name = tensor<string, []>("v_197_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_197_groups_0 = const()[name = tensor<string, []>("v_197_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_5_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(436999488))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(438228352))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_5_attn1_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> v_197_cast_fp16 = conv(dilations = v_197_dilations_0, groups = v_197_groups_0, pad = v_197_pad_0, pad_type = v_197_pad_type_0, strides = v_197_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_5_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_391_cast_fp16)[name = tensor<string, []>("v_197_cast_fp16")];
            tensor<int32, [4]> var_44272_begin_0 = const()[name = tensor<string, []>("op_44272_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_44272_end_0 = const()[name = tensor<string, []>("op_44272_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_44272_end_mask_0 = const()[name = tensor<string, []>("op_44272_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44272_cast_fp16 = slice_by_index(begin = var_44272_begin_0, end = var_44272_end_0, end_mask = var_44272_end_mask_0, x = q_197_cast_fp16)[name = tensor<string, []>("op_44272_cast_fp16")];
            tensor<int32, [4]> var_44276_begin_0 = const()[name = tensor<string, []>("op_44276_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_44276_end_0 = const()[name = tensor<string, []>("op_44276_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_44276_end_mask_0 = const()[name = tensor<string, []>("op_44276_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44276_cast_fp16 = slice_by_index(begin = var_44276_begin_0, end = var_44276_end_0, end_mask = var_44276_end_mask_0, x = q_197_cast_fp16)[name = tensor<string, []>("op_44276_cast_fp16")];
            tensor<int32, [4]> var_44280_begin_0 = const()[name = tensor<string, []>("op_44280_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_44280_end_0 = const()[name = tensor<string, []>("op_44280_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_44280_end_mask_0 = const()[name = tensor<string, []>("op_44280_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44280_cast_fp16 = slice_by_index(begin = var_44280_begin_0, end = var_44280_end_0, end_mask = var_44280_end_mask_0, x = q_197_cast_fp16)[name = tensor<string, []>("op_44280_cast_fp16")];
            tensor<int32, [4]> var_44284_begin_0 = const()[name = tensor<string, []>("op_44284_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_44284_end_0 = const()[name = tensor<string, []>("op_44284_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_44284_end_mask_0 = const()[name = tensor<string, []>("op_44284_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44284_cast_fp16 = slice_by_index(begin = var_44284_begin_0, end = var_44284_end_0, end_mask = var_44284_end_mask_0, x = q_197_cast_fp16)[name = tensor<string, []>("op_44284_cast_fp16")];
            tensor<int32, [4]> var_44288_begin_0 = const()[name = tensor<string, []>("op_44288_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_44288_end_0 = const()[name = tensor<string, []>("op_44288_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_44288_end_mask_0 = const()[name = tensor<string, []>("op_44288_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44288_cast_fp16 = slice_by_index(begin = var_44288_begin_0, end = var_44288_end_0, end_mask = var_44288_end_mask_0, x = q_197_cast_fp16)[name = tensor<string, []>("op_44288_cast_fp16")];
            tensor<int32, [4]> var_44292_begin_0 = const()[name = tensor<string, []>("op_44292_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_44292_end_0 = const()[name = tensor<string, []>("op_44292_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_44292_end_mask_0 = const()[name = tensor<string, []>("op_44292_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44292_cast_fp16 = slice_by_index(begin = var_44292_begin_0, end = var_44292_end_0, end_mask = var_44292_end_mask_0, x = q_197_cast_fp16)[name = tensor<string, []>("op_44292_cast_fp16")];
            tensor<int32, [4]> var_44296_begin_0 = const()[name = tensor<string, []>("op_44296_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_44296_end_0 = const()[name = tensor<string, []>("op_44296_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_44296_end_mask_0 = const()[name = tensor<string, []>("op_44296_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44296_cast_fp16 = slice_by_index(begin = var_44296_begin_0, end = var_44296_end_0, end_mask = var_44296_end_mask_0, x = q_197_cast_fp16)[name = tensor<string, []>("op_44296_cast_fp16")];
            tensor<int32, [4]> var_44300_begin_0 = const()[name = tensor<string, []>("op_44300_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_44300_end_0 = const()[name = tensor<string, []>("op_44300_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_44300_end_mask_0 = const()[name = tensor<string, []>("op_44300_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44300_cast_fp16 = slice_by_index(begin = var_44300_begin_0, end = var_44300_end_0, end_mask = var_44300_end_mask_0, x = q_197_cast_fp16)[name = tensor<string, []>("op_44300_cast_fp16")];
            tensor<int32, [4]> var_44304_begin_0 = const()[name = tensor<string, []>("op_44304_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_44304_end_0 = const()[name = tensor<string, []>("op_44304_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_44304_end_mask_0 = const()[name = tensor<string, []>("op_44304_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44304_cast_fp16 = slice_by_index(begin = var_44304_begin_0, end = var_44304_end_0, end_mask = var_44304_end_mask_0, x = q_197_cast_fp16)[name = tensor<string, []>("op_44304_cast_fp16")];
            tensor<int32, [4]> var_44308_begin_0 = const()[name = tensor<string, []>("op_44308_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_44308_end_0 = const()[name = tensor<string, []>("op_44308_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_44308_end_mask_0 = const()[name = tensor<string, []>("op_44308_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44308_cast_fp16 = slice_by_index(begin = var_44308_begin_0, end = var_44308_end_0, end_mask = var_44308_end_mask_0, x = q_197_cast_fp16)[name = tensor<string, []>("op_44308_cast_fp16")];
            tensor<int32, [4]> var_44312_begin_0 = const()[name = tensor<string, []>("op_44312_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_44312_end_0 = const()[name = tensor<string, []>("op_44312_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_44312_end_mask_0 = const()[name = tensor<string, []>("op_44312_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44312_cast_fp16 = slice_by_index(begin = var_44312_begin_0, end = var_44312_end_0, end_mask = var_44312_end_mask_0, x = q_197_cast_fp16)[name = tensor<string, []>("op_44312_cast_fp16")];
            tensor<int32, [4]> var_44316_begin_0 = const()[name = tensor<string, []>("op_44316_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_44316_end_0 = const()[name = tensor<string, []>("op_44316_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_44316_end_mask_0 = const()[name = tensor<string, []>("op_44316_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44316_cast_fp16 = slice_by_index(begin = var_44316_begin_0, end = var_44316_end_0, end_mask = var_44316_end_mask_0, x = q_197_cast_fp16)[name = tensor<string, []>("op_44316_cast_fp16")];
            tensor<int32, [4]> var_44320_begin_0 = const()[name = tensor<string, []>("op_44320_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_44320_end_0 = const()[name = tensor<string, []>("op_44320_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_44320_end_mask_0 = const()[name = tensor<string, []>("op_44320_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44320_cast_fp16 = slice_by_index(begin = var_44320_begin_0, end = var_44320_end_0, end_mask = var_44320_end_mask_0, x = q_197_cast_fp16)[name = tensor<string, []>("op_44320_cast_fp16")];
            tensor<int32, [4]> var_44324_begin_0 = const()[name = tensor<string, []>("op_44324_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_44324_end_0 = const()[name = tensor<string, []>("op_44324_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_44324_end_mask_0 = const()[name = tensor<string, []>("op_44324_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44324_cast_fp16 = slice_by_index(begin = var_44324_begin_0, end = var_44324_end_0, end_mask = var_44324_end_mask_0, x = q_197_cast_fp16)[name = tensor<string, []>("op_44324_cast_fp16")];
            tensor<int32, [4]> var_44328_begin_0 = const()[name = tensor<string, []>("op_44328_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_44328_end_0 = const()[name = tensor<string, []>("op_44328_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_44328_end_mask_0 = const()[name = tensor<string, []>("op_44328_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44328_cast_fp16 = slice_by_index(begin = var_44328_begin_0, end = var_44328_end_0, end_mask = var_44328_end_mask_0, x = q_197_cast_fp16)[name = tensor<string, []>("op_44328_cast_fp16")];
            tensor<int32, [4]> var_44332_begin_0 = const()[name = tensor<string, []>("op_44332_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_44332_end_0 = const()[name = tensor<string, []>("op_44332_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_44332_end_mask_0 = const()[name = tensor<string, []>("op_44332_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44332_cast_fp16 = slice_by_index(begin = var_44332_begin_0, end = var_44332_end_0, end_mask = var_44332_end_mask_0, x = q_197_cast_fp16)[name = tensor<string, []>("op_44332_cast_fp16")];
            tensor<int32, [4]> var_44336_begin_0 = const()[name = tensor<string, []>("op_44336_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_44336_end_0 = const()[name = tensor<string, []>("op_44336_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_44336_end_mask_0 = const()[name = tensor<string, []>("op_44336_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44336_cast_fp16 = slice_by_index(begin = var_44336_begin_0, end = var_44336_end_0, end_mask = var_44336_end_mask_0, x = q_197_cast_fp16)[name = tensor<string, []>("op_44336_cast_fp16")];
            tensor<int32, [4]> var_44340_begin_0 = const()[name = tensor<string, []>("op_44340_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_44340_end_0 = const()[name = tensor<string, []>("op_44340_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_44340_end_mask_0 = const()[name = tensor<string, []>("op_44340_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44340_cast_fp16 = slice_by_index(begin = var_44340_begin_0, end = var_44340_end_0, end_mask = var_44340_end_mask_0, x = q_197_cast_fp16)[name = tensor<string, []>("op_44340_cast_fp16")];
            tensor<int32, [4]> var_44344_begin_0 = const()[name = tensor<string, []>("op_44344_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_44344_end_0 = const()[name = tensor<string, []>("op_44344_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_44344_end_mask_0 = const()[name = tensor<string, []>("op_44344_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44344_cast_fp16 = slice_by_index(begin = var_44344_begin_0, end = var_44344_end_0, end_mask = var_44344_end_mask_0, x = q_197_cast_fp16)[name = tensor<string, []>("op_44344_cast_fp16")];
            tensor<int32, [4]> var_44348_begin_0 = const()[name = tensor<string, []>("op_44348_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_44348_end_0 = const()[name = tensor<string, []>("op_44348_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_44348_end_mask_0 = const()[name = tensor<string, []>("op_44348_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44348_cast_fp16 = slice_by_index(begin = var_44348_begin_0, end = var_44348_end_0, end_mask = var_44348_end_mask_0, x = q_197_cast_fp16)[name = tensor<string, []>("op_44348_cast_fp16")];
            tensor<int32, [4]> k_395_perm_0 = const()[name = tensor<string, []>("k_395_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_44355_begin_0 = const()[name = tensor<string, []>("op_44355_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_44355_end_0 = const()[name = tensor<string, []>("op_44355_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_44355_end_mask_0 = const()[name = tensor<string, []>("op_44355_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> k_395_cast_fp16 = transpose(perm = k_395_perm_0, x = k_393_cast_fp16)[name = tensor<string, []>("transpose_41")];
            tensor<fp16, [2, 1024, 1, 64]> var_44355_cast_fp16 = slice_by_index(begin = var_44355_begin_0, end = var_44355_end_0, end_mask = var_44355_end_mask_0, x = k_395_cast_fp16)[name = tensor<string, []>("op_44355_cast_fp16")];
            tensor<int32, [4]> var_44359_begin_0 = const()[name = tensor<string, []>("op_44359_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_44359_end_0 = const()[name = tensor<string, []>("op_44359_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_44359_end_mask_0 = const()[name = tensor<string, []>("op_44359_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_44359_cast_fp16 = slice_by_index(begin = var_44359_begin_0, end = var_44359_end_0, end_mask = var_44359_end_mask_0, x = k_395_cast_fp16)[name = tensor<string, []>("op_44359_cast_fp16")];
            tensor<int32, [4]> var_44363_begin_0 = const()[name = tensor<string, []>("op_44363_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_44363_end_0 = const()[name = tensor<string, []>("op_44363_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_44363_end_mask_0 = const()[name = tensor<string, []>("op_44363_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_44363_cast_fp16 = slice_by_index(begin = var_44363_begin_0, end = var_44363_end_0, end_mask = var_44363_end_mask_0, x = k_395_cast_fp16)[name = tensor<string, []>("op_44363_cast_fp16")];
            tensor<int32, [4]> var_44367_begin_0 = const()[name = tensor<string, []>("op_44367_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_44367_end_0 = const()[name = tensor<string, []>("op_44367_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_44367_end_mask_0 = const()[name = tensor<string, []>("op_44367_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_44367_cast_fp16 = slice_by_index(begin = var_44367_begin_0, end = var_44367_end_0, end_mask = var_44367_end_mask_0, x = k_395_cast_fp16)[name = tensor<string, []>("op_44367_cast_fp16")];
            tensor<int32, [4]> var_44371_begin_0 = const()[name = tensor<string, []>("op_44371_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_44371_end_0 = const()[name = tensor<string, []>("op_44371_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_44371_end_mask_0 = const()[name = tensor<string, []>("op_44371_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_44371_cast_fp16 = slice_by_index(begin = var_44371_begin_0, end = var_44371_end_0, end_mask = var_44371_end_mask_0, x = k_395_cast_fp16)[name = tensor<string, []>("op_44371_cast_fp16")];
            tensor<int32, [4]> var_44375_begin_0 = const()[name = tensor<string, []>("op_44375_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_44375_end_0 = const()[name = tensor<string, []>("op_44375_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_44375_end_mask_0 = const()[name = tensor<string, []>("op_44375_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_44375_cast_fp16 = slice_by_index(begin = var_44375_begin_0, end = var_44375_end_0, end_mask = var_44375_end_mask_0, x = k_395_cast_fp16)[name = tensor<string, []>("op_44375_cast_fp16")];
            tensor<int32, [4]> var_44379_begin_0 = const()[name = tensor<string, []>("op_44379_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_44379_end_0 = const()[name = tensor<string, []>("op_44379_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_44379_end_mask_0 = const()[name = tensor<string, []>("op_44379_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_44379_cast_fp16 = slice_by_index(begin = var_44379_begin_0, end = var_44379_end_0, end_mask = var_44379_end_mask_0, x = k_395_cast_fp16)[name = tensor<string, []>("op_44379_cast_fp16")];
            tensor<int32, [4]> var_44383_begin_0 = const()[name = tensor<string, []>("op_44383_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_44383_end_0 = const()[name = tensor<string, []>("op_44383_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_44383_end_mask_0 = const()[name = tensor<string, []>("op_44383_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_44383_cast_fp16 = slice_by_index(begin = var_44383_begin_0, end = var_44383_end_0, end_mask = var_44383_end_mask_0, x = k_395_cast_fp16)[name = tensor<string, []>("op_44383_cast_fp16")];
            tensor<int32, [4]> var_44387_begin_0 = const()[name = tensor<string, []>("op_44387_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_44387_end_0 = const()[name = tensor<string, []>("op_44387_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_44387_end_mask_0 = const()[name = tensor<string, []>("op_44387_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_44387_cast_fp16 = slice_by_index(begin = var_44387_begin_0, end = var_44387_end_0, end_mask = var_44387_end_mask_0, x = k_395_cast_fp16)[name = tensor<string, []>("op_44387_cast_fp16")];
            tensor<int32, [4]> var_44391_begin_0 = const()[name = tensor<string, []>("op_44391_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_44391_end_0 = const()[name = tensor<string, []>("op_44391_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_44391_end_mask_0 = const()[name = tensor<string, []>("op_44391_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_44391_cast_fp16 = slice_by_index(begin = var_44391_begin_0, end = var_44391_end_0, end_mask = var_44391_end_mask_0, x = k_395_cast_fp16)[name = tensor<string, []>("op_44391_cast_fp16")];
            tensor<int32, [4]> var_44395_begin_0 = const()[name = tensor<string, []>("op_44395_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_44395_end_0 = const()[name = tensor<string, []>("op_44395_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_44395_end_mask_0 = const()[name = tensor<string, []>("op_44395_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_44395_cast_fp16 = slice_by_index(begin = var_44395_begin_0, end = var_44395_end_0, end_mask = var_44395_end_mask_0, x = k_395_cast_fp16)[name = tensor<string, []>("op_44395_cast_fp16")];
            tensor<int32, [4]> var_44399_begin_0 = const()[name = tensor<string, []>("op_44399_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_44399_end_0 = const()[name = tensor<string, []>("op_44399_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_44399_end_mask_0 = const()[name = tensor<string, []>("op_44399_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_44399_cast_fp16 = slice_by_index(begin = var_44399_begin_0, end = var_44399_end_0, end_mask = var_44399_end_mask_0, x = k_395_cast_fp16)[name = tensor<string, []>("op_44399_cast_fp16")];
            tensor<int32, [4]> var_44403_begin_0 = const()[name = tensor<string, []>("op_44403_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_44403_end_0 = const()[name = tensor<string, []>("op_44403_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_44403_end_mask_0 = const()[name = tensor<string, []>("op_44403_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_44403_cast_fp16 = slice_by_index(begin = var_44403_begin_0, end = var_44403_end_0, end_mask = var_44403_end_mask_0, x = k_395_cast_fp16)[name = tensor<string, []>("op_44403_cast_fp16")];
            tensor<int32, [4]> var_44407_begin_0 = const()[name = tensor<string, []>("op_44407_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_44407_end_0 = const()[name = tensor<string, []>("op_44407_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_44407_end_mask_0 = const()[name = tensor<string, []>("op_44407_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_44407_cast_fp16 = slice_by_index(begin = var_44407_begin_0, end = var_44407_end_0, end_mask = var_44407_end_mask_0, x = k_395_cast_fp16)[name = tensor<string, []>("op_44407_cast_fp16")];
            tensor<int32, [4]> var_44411_begin_0 = const()[name = tensor<string, []>("op_44411_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_44411_end_0 = const()[name = tensor<string, []>("op_44411_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_44411_end_mask_0 = const()[name = tensor<string, []>("op_44411_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_44411_cast_fp16 = slice_by_index(begin = var_44411_begin_0, end = var_44411_end_0, end_mask = var_44411_end_mask_0, x = k_395_cast_fp16)[name = tensor<string, []>("op_44411_cast_fp16")];
            tensor<int32, [4]> var_44415_begin_0 = const()[name = tensor<string, []>("op_44415_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_44415_end_0 = const()[name = tensor<string, []>("op_44415_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_44415_end_mask_0 = const()[name = tensor<string, []>("op_44415_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_44415_cast_fp16 = slice_by_index(begin = var_44415_begin_0, end = var_44415_end_0, end_mask = var_44415_end_mask_0, x = k_395_cast_fp16)[name = tensor<string, []>("op_44415_cast_fp16")];
            tensor<int32, [4]> var_44419_begin_0 = const()[name = tensor<string, []>("op_44419_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_44419_end_0 = const()[name = tensor<string, []>("op_44419_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_44419_end_mask_0 = const()[name = tensor<string, []>("op_44419_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_44419_cast_fp16 = slice_by_index(begin = var_44419_begin_0, end = var_44419_end_0, end_mask = var_44419_end_mask_0, x = k_395_cast_fp16)[name = tensor<string, []>("op_44419_cast_fp16")];
            tensor<int32, [4]> var_44423_begin_0 = const()[name = tensor<string, []>("op_44423_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_44423_end_0 = const()[name = tensor<string, []>("op_44423_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_44423_end_mask_0 = const()[name = tensor<string, []>("op_44423_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_44423_cast_fp16 = slice_by_index(begin = var_44423_begin_0, end = var_44423_end_0, end_mask = var_44423_end_mask_0, x = k_395_cast_fp16)[name = tensor<string, []>("op_44423_cast_fp16")];
            tensor<int32, [4]> var_44427_begin_0 = const()[name = tensor<string, []>("op_44427_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_44427_end_0 = const()[name = tensor<string, []>("op_44427_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_44427_end_mask_0 = const()[name = tensor<string, []>("op_44427_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_44427_cast_fp16 = slice_by_index(begin = var_44427_begin_0, end = var_44427_end_0, end_mask = var_44427_end_mask_0, x = k_395_cast_fp16)[name = tensor<string, []>("op_44427_cast_fp16")];
            tensor<int32, [4]> var_44431_begin_0 = const()[name = tensor<string, []>("op_44431_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_44431_end_0 = const()[name = tensor<string, []>("op_44431_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_44431_end_mask_0 = const()[name = tensor<string, []>("op_44431_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_44431_cast_fp16 = slice_by_index(begin = var_44431_begin_0, end = var_44431_end_0, end_mask = var_44431_end_mask_0, x = k_395_cast_fp16)[name = tensor<string, []>("op_44431_cast_fp16")];
            tensor<int32, [4]> var_44433_begin_0 = const()[name = tensor<string, []>("op_44433_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_44433_end_0 = const()[name = tensor<string, []>("op_44433_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_44433_end_mask_0 = const()[name = tensor<string, []>("op_44433_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44433_cast_fp16 = slice_by_index(begin = var_44433_begin_0, end = var_44433_end_0, end_mask = var_44433_end_mask_0, x = v_197_cast_fp16)[name = tensor<string, []>("op_44433_cast_fp16")];
            tensor<int32, [4]> var_44437_begin_0 = const()[name = tensor<string, []>("op_44437_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_44437_end_0 = const()[name = tensor<string, []>("op_44437_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_44437_end_mask_0 = const()[name = tensor<string, []>("op_44437_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44437_cast_fp16 = slice_by_index(begin = var_44437_begin_0, end = var_44437_end_0, end_mask = var_44437_end_mask_0, x = v_197_cast_fp16)[name = tensor<string, []>("op_44437_cast_fp16")];
            tensor<int32, [4]> var_44441_begin_0 = const()[name = tensor<string, []>("op_44441_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_44441_end_0 = const()[name = tensor<string, []>("op_44441_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_44441_end_mask_0 = const()[name = tensor<string, []>("op_44441_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44441_cast_fp16 = slice_by_index(begin = var_44441_begin_0, end = var_44441_end_0, end_mask = var_44441_end_mask_0, x = v_197_cast_fp16)[name = tensor<string, []>("op_44441_cast_fp16")];
            tensor<int32, [4]> var_44445_begin_0 = const()[name = tensor<string, []>("op_44445_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_44445_end_0 = const()[name = tensor<string, []>("op_44445_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_44445_end_mask_0 = const()[name = tensor<string, []>("op_44445_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44445_cast_fp16 = slice_by_index(begin = var_44445_begin_0, end = var_44445_end_0, end_mask = var_44445_end_mask_0, x = v_197_cast_fp16)[name = tensor<string, []>("op_44445_cast_fp16")];
            tensor<int32, [4]> var_44449_begin_0 = const()[name = tensor<string, []>("op_44449_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_44449_end_0 = const()[name = tensor<string, []>("op_44449_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_44449_end_mask_0 = const()[name = tensor<string, []>("op_44449_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44449_cast_fp16 = slice_by_index(begin = var_44449_begin_0, end = var_44449_end_0, end_mask = var_44449_end_mask_0, x = v_197_cast_fp16)[name = tensor<string, []>("op_44449_cast_fp16")];
            tensor<int32, [4]> var_44453_begin_0 = const()[name = tensor<string, []>("op_44453_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_44453_end_0 = const()[name = tensor<string, []>("op_44453_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_44453_end_mask_0 = const()[name = tensor<string, []>("op_44453_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44453_cast_fp16 = slice_by_index(begin = var_44453_begin_0, end = var_44453_end_0, end_mask = var_44453_end_mask_0, x = v_197_cast_fp16)[name = tensor<string, []>("op_44453_cast_fp16")];
            tensor<int32, [4]> var_44457_begin_0 = const()[name = tensor<string, []>("op_44457_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_44457_end_0 = const()[name = tensor<string, []>("op_44457_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_44457_end_mask_0 = const()[name = tensor<string, []>("op_44457_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44457_cast_fp16 = slice_by_index(begin = var_44457_begin_0, end = var_44457_end_0, end_mask = var_44457_end_mask_0, x = v_197_cast_fp16)[name = tensor<string, []>("op_44457_cast_fp16")];
            tensor<int32, [4]> var_44461_begin_0 = const()[name = tensor<string, []>("op_44461_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_44461_end_0 = const()[name = tensor<string, []>("op_44461_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_44461_end_mask_0 = const()[name = tensor<string, []>("op_44461_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44461_cast_fp16 = slice_by_index(begin = var_44461_begin_0, end = var_44461_end_0, end_mask = var_44461_end_mask_0, x = v_197_cast_fp16)[name = tensor<string, []>("op_44461_cast_fp16")];
            tensor<int32, [4]> var_44465_begin_0 = const()[name = tensor<string, []>("op_44465_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_44465_end_0 = const()[name = tensor<string, []>("op_44465_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_44465_end_mask_0 = const()[name = tensor<string, []>("op_44465_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44465_cast_fp16 = slice_by_index(begin = var_44465_begin_0, end = var_44465_end_0, end_mask = var_44465_end_mask_0, x = v_197_cast_fp16)[name = tensor<string, []>("op_44465_cast_fp16")];
            tensor<int32, [4]> var_44469_begin_0 = const()[name = tensor<string, []>("op_44469_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_44469_end_0 = const()[name = tensor<string, []>("op_44469_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_44469_end_mask_0 = const()[name = tensor<string, []>("op_44469_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44469_cast_fp16 = slice_by_index(begin = var_44469_begin_0, end = var_44469_end_0, end_mask = var_44469_end_mask_0, x = v_197_cast_fp16)[name = tensor<string, []>("op_44469_cast_fp16")];
            tensor<int32, [4]> var_44473_begin_0 = const()[name = tensor<string, []>("op_44473_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_44473_end_0 = const()[name = tensor<string, []>("op_44473_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_44473_end_mask_0 = const()[name = tensor<string, []>("op_44473_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44473_cast_fp16 = slice_by_index(begin = var_44473_begin_0, end = var_44473_end_0, end_mask = var_44473_end_mask_0, x = v_197_cast_fp16)[name = tensor<string, []>("op_44473_cast_fp16")];
            tensor<int32, [4]> var_44477_begin_0 = const()[name = tensor<string, []>("op_44477_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_44477_end_0 = const()[name = tensor<string, []>("op_44477_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_44477_end_mask_0 = const()[name = tensor<string, []>("op_44477_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44477_cast_fp16 = slice_by_index(begin = var_44477_begin_0, end = var_44477_end_0, end_mask = var_44477_end_mask_0, x = v_197_cast_fp16)[name = tensor<string, []>("op_44477_cast_fp16")];
            tensor<int32, [4]> var_44481_begin_0 = const()[name = tensor<string, []>("op_44481_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_44481_end_0 = const()[name = tensor<string, []>("op_44481_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_44481_end_mask_0 = const()[name = tensor<string, []>("op_44481_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44481_cast_fp16 = slice_by_index(begin = var_44481_begin_0, end = var_44481_end_0, end_mask = var_44481_end_mask_0, x = v_197_cast_fp16)[name = tensor<string, []>("op_44481_cast_fp16")];
            tensor<int32, [4]> var_44485_begin_0 = const()[name = tensor<string, []>("op_44485_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_44485_end_0 = const()[name = tensor<string, []>("op_44485_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_44485_end_mask_0 = const()[name = tensor<string, []>("op_44485_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44485_cast_fp16 = slice_by_index(begin = var_44485_begin_0, end = var_44485_end_0, end_mask = var_44485_end_mask_0, x = v_197_cast_fp16)[name = tensor<string, []>("op_44485_cast_fp16")];
            tensor<int32, [4]> var_44489_begin_0 = const()[name = tensor<string, []>("op_44489_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_44489_end_0 = const()[name = tensor<string, []>("op_44489_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_44489_end_mask_0 = const()[name = tensor<string, []>("op_44489_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44489_cast_fp16 = slice_by_index(begin = var_44489_begin_0, end = var_44489_end_0, end_mask = var_44489_end_mask_0, x = v_197_cast_fp16)[name = tensor<string, []>("op_44489_cast_fp16")];
            tensor<int32, [4]> var_44493_begin_0 = const()[name = tensor<string, []>("op_44493_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_44493_end_0 = const()[name = tensor<string, []>("op_44493_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_44493_end_mask_0 = const()[name = tensor<string, []>("op_44493_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44493_cast_fp16 = slice_by_index(begin = var_44493_begin_0, end = var_44493_end_0, end_mask = var_44493_end_mask_0, x = v_197_cast_fp16)[name = tensor<string, []>("op_44493_cast_fp16")];
            tensor<int32, [4]> var_44497_begin_0 = const()[name = tensor<string, []>("op_44497_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_44497_end_0 = const()[name = tensor<string, []>("op_44497_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_44497_end_mask_0 = const()[name = tensor<string, []>("op_44497_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44497_cast_fp16 = slice_by_index(begin = var_44497_begin_0, end = var_44497_end_0, end_mask = var_44497_end_mask_0, x = v_197_cast_fp16)[name = tensor<string, []>("op_44497_cast_fp16")];
            tensor<int32, [4]> var_44501_begin_0 = const()[name = tensor<string, []>("op_44501_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_44501_end_0 = const()[name = tensor<string, []>("op_44501_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_44501_end_mask_0 = const()[name = tensor<string, []>("op_44501_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44501_cast_fp16 = slice_by_index(begin = var_44501_begin_0, end = var_44501_end_0, end_mask = var_44501_end_mask_0, x = v_197_cast_fp16)[name = tensor<string, []>("op_44501_cast_fp16")];
            tensor<int32, [4]> var_44505_begin_0 = const()[name = tensor<string, []>("op_44505_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_44505_end_0 = const()[name = tensor<string, []>("op_44505_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_44505_end_mask_0 = const()[name = tensor<string, []>("op_44505_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44505_cast_fp16 = slice_by_index(begin = var_44505_begin_0, end = var_44505_end_0, end_mask = var_44505_end_mask_0, x = v_197_cast_fp16)[name = tensor<string, []>("op_44505_cast_fp16")];
            tensor<int32, [4]> var_44509_begin_0 = const()[name = tensor<string, []>("op_44509_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_44509_end_0 = const()[name = tensor<string, []>("op_44509_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_44509_end_mask_0 = const()[name = tensor<string, []>("op_44509_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44509_cast_fp16 = slice_by_index(begin = var_44509_begin_0, end = var_44509_end_0, end_mask = var_44509_end_mask_0, x = v_197_cast_fp16)[name = tensor<string, []>("op_44509_cast_fp16")];
            tensor<string, []> var_44513_equation_0 = const()[name = tensor<string, []>("op_44513_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44513_cast_fp16 = einsum(equation = var_44513_equation_0, values = (var_44355_cast_fp16, var_44272_cast_fp16))[name = tensor<string, []>("op_44513_cast_fp16")];
            tensor<fp16, []> var_44514_to_fp16 = const()[name = tensor<string, []>("op_44514_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3761_cast_fp16 = mul(x = var_44513_cast_fp16, y = var_44514_to_fp16)[name = tensor<string, []>("aw_3761_cast_fp16")];
            tensor<string, []> var_44517_equation_0 = const()[name = tensor<string, []>("op_44517_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44517_cast_fp16 = einsum(equation = var_44517_equation_0, values = (var_44359_cast_fp16, var_44276_cast_fp16))[name = tensor<string, []>("op_44517_cast_fp16")];
            tensor<fp16, []> var_44518_to_fp16 = const()[name = tensor<string, []>("op_44518_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3763_cast_fp16 = mul(x = var_44517_cast_fp16, y = var_44518_to_fp16)[name = tensor<string, []>("aw_3763_cast_fp16")];
            tensor<string, []> var_44521_equation_0 = const()[name = tensor<string, []>("op_44521_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44521_cast_fp16 = einsum(equation = var_44521_equation_0, values = (var_44363_cast_fp16, var_44280_cast_fp16))[name = tensor<string, []>("op_44521_cast_fp16")];
            tensor<fp16, []> var_44522_to_fp16 = const()[name = tensor<string, []>("op_44522_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3765_cast_fp16 = mul(x = var_44521_cast_fp16, y = var_44522_to_fp16)[name = tensor<string, []>("aw_3765_cast_fp16")];
            tensor<string, []> var_44525_equation_0 = const()[name = tensor<string, []>("op_44525_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44525_cast_fp16 = einsum(equation = var_44525_equation_0, values = (var_44367_cast_fp16, var_44284_cast_fp16))[name = tensor<string, []>("op_44525_cast_fp16")];
            tensor<fp16, []> var_44526_to_fp16 = const()[name = tensor<string, []>("op_44526_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3767_cast_fp16 = mul(x = var_44525_cast_fp16, y = var_44526_to_fp16)[name = tensor<string, []>("aw_3767_cast_fp16")];
            tensor<string, []> var_44529_equation_0 = const()[name = tensor<string, []>("op_44529_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44529_cast_fp16 = einsum(equation = var_44529_equation_0, values = (var_44371_cast_fp16, var_44288_cast_fp16))[name = tensor<string, []>("op_44529_cast_fp16")];
            tensor<fp16, []> var_44530_to_fp16 = const()[name = tensor<string, []>("op_44530_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3769_cast_fp16 = mul(x = var_44529_cast_fp16, y = var_44530_to_fp16)[name = tensor<string, []>("aw_3769_cast_fp16")];
            tensor<string, []> var_44533_equation_0 = const()[name = tensor<string, []>("op_44533_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44533_cast_fp16 = einsum(equation = var_44533_equation_0, values = (var_44375_cast_fp16, var_44292_cast_fp16))[name = tensor<string, []>("op_44533_cast_fp16")];
            tensor<fp16, []> var_44534_to_fp16 = const()[name = tensor<string, []>("op_44534_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3771_cast_fp16 = mul(x = var_44533_cast_fp16, y = var_44534_to_fp16)[name = tensor<string, []>("aw_3771_cast_fp16")];
            tensor<string, []> var_44537_equation_0 = const()[name = tensor<string, []>("op_44537_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44537_cast_fp16 = einsum(equation = var_44537_equation_0, values = (var_44379_cast_fp16, var_44296_cast_fp16))[name = tensor<string, []>("op_44537_cast_fp16")];
            tensor<fp16, []> var_44538_to_fp16 = const()[name = tensor<string, []>("op_44538_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3773_cast_fp16 = mul(x = var_44537_cast_fp16, y = var_44538_to_fp16)[name = tensor<string, []>("aw_3773_cast_fp16")];
            tensor<string, []> var_44541_equation_0 = const()[name = tensor<string, []>("op_44541_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44541_cast_fp16 = einsum(equation = var_44541_equation_0, values = (var_44383_cast_fp16, var_44300_cast_fp16))[name = tensor<string, []>("op_44541_cast_fp16")];
            tensor<fp16, []> var_44542_to_fp16 = const()[name = tensor<string, []>("op_44542_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3775_cast_fp16 = mul(x = var_44541_cast_fp16, y = var_44542_to_fp16)[name = tensor<string, []>("aw_3775_cast_fp16")];
            tensor<string, []> var_44545_equation_0 = const()[name = tensor<string, []>("op_44545_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44545_cast_fp16 = einsum(equation = var_44545_equation_0, values = (var_44387_cast_fp16, var_44304_cast_fp16))[name = tensor<string, []>("op_44545_cast_fp16")];
            tensor<fp16, []> var_44546_to_fp16 = const()[name = tensor<string, []>("op_44546_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3777_cast_fp16 = mul(x = var_44545_cast_fp16, y = var_44546_to_fp16)[name = tensor<string, []>("aw_3777_cast_fp16")];
            tensor<string, []> var_44549_equation_0 = const()[name = tensor<string, []>("op_44549_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44549_cast_fp16 = einsum(equation = var_44549_equation_0, values = (var_44391_cast_fp16, var_44308_cast_fp16))[name = tensor<string, []>("op_44549_cast_fp16")];
            tensor<fp16, []> var_44550_to_fp16 = const()[name = tensor<string, []>("op_44550_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3779_cast_fp16 = mul(x = var_44549_cast_fp16, y = var_44550_to_fp16)[name = tensor<string, []>("aw_3779_cast_fp16")];
            tensor<string, []> var_44553_equation_0 = const()[name = tensor<string, []>("op_44553_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44553_cast_fp16 = einsum(equation = var_44553_equation_0, values = (var_44395_cast_fp16, var_44312_cast_fp16))[name = tensor<string, []>("op_44553_cast_fp16")];
            tensor<fp16, []> var_44554_to_fp16 = const()[name = tensor<string, []>("op_44554_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3781_cast_fp16 = mul(x = var_44553_cast_fp16, y = var_44554_to_fp16)[name = tensor<string, []>("aw_3781_cast_fp16")];
            tensor<string, []> var_44557_equation_0 = const()[name = tensor<string, []>("op_44557_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44557_cast_fp16 = einsum(equation = var_44557_equation_0, values = (var_44399_cast_fp16, var_44316_cast_fp16))[name = tensor<string, []>("op_44557_cast_fp16")];
            tensor<fp16, []> var_44558_to_fp16 = const()[name = tensor<string, []>("op_44558_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3783_cast_fp16 = mul(x = var_44557_cast_fp16, y = var_44558_to_fp16)[name = tensor<string, []>("aw_3783_cast_fp16")];
            tensor<string, []> var_44561_equation_0 = const()[name = tensor<string, []>("op_44561_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44561_cast_fp16 = einsum(equation = var_44561_equation_0, values = (var_44403_cast_fp16, var_44320_cast_fp16))[name = tensor<string, []>("op_44561_cast_fp16")];
            tensor<fp16, []> var_44562_to_fp16 = const()[name = tensor<string, []>("op_44562_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3785_cast_fp16 = mul(x = var_44561_cast_fp16, y = var_44562_to_fp16)[name = tensor<string, []>("aw_3785_cast_fp16")];
            tensor<string, []> var_44565_equation_0 = const()[name = tensor<string, []>("op_44565_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44565_cast_fp16 = einsum(equation = var_44565_equation_0, values = (var_44407_cast_fp16, var_44324_cast_fp16))[name = tensor<string, []>("op_44565_cast_fp16")];
            tensor<fp16, []> var_44566_to_fp16 = const()[name = tensor<string, []>("op_44566_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3787_cast_fp16 = mul(x = var_44565_cast_fp16, y = var_44566_to_fp16)[name = tensor<string, []>("aw_3787_cast_fp16")];
            tensor<string, []> var_44569_equation_0 = const()[name = tensor<string, []>("op_44569_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44569_cast_fp16 = einsum(equation = var_44569_equation_0, values = (var_44411_cast_fp16, var_44328_cast_fp16))[name = tensor<string, []>("op_44569_cast_fp16")];
            tensor<fp16, []> var_44570_to_fp16 = const()[name = tensor<string, []>("op_44570_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3789_cast_fp16 = mul(x = var_44569_cast_fp16, y = var_44570_to_fp16)[name = tensor<string, []>("aw_3789_cast_fp16")];
            tensor<string, []> var_44573_equation_0 = const()[name = tensor<string, []>("op_44573_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44573_cast_fp16 = einsum(equation = var_44573_equation_0, values = (var_44415_cast_fp16, var_44332_cast_fp16))[name = tensor<string, []>("op_44573_cast_fp16")];
            tensor<fp16, []> var_44574_to_fp16 = const()[name = tensor<string, []>("op_44574_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3791_cast_fp16 = mul(x = var_44573_cast_fp16, y = var_44574_to_fp16)[name = tensor<string, []>("aw_3791_cast_fp16")];
            tensor<string, []> var_44577_equation_0 = const()[name = tensor<string, []>("op_44577_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44577_cast_fp16 = einsum(equation = var_44577_equation_0, values = (var_44419_cast_fp16, var_44336_cast_fp16))[name = tensor<string, []>("op_44577_cast_fp16")];
            tensor<fp16, []> var_44578_to_fp16 = const()[name = tensor<string, []>("op_44578_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3793_cast_fp16 = mul(x = var_44577_cast_fp16, y = var_44578_to_fp16)[name = tensor<string, []>("aw_3793_cast_fp16")];
            tensor<string, []> var_44581_equation_0 = const()[name = tensor<string, []>("op_44581_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44581_cast_fp16 = einsum(equation = var_44581_equation_0, values = (var_44423_cast_fp16, var_44340_cast_fp16))[name = tensor<string, []>("op_44581_cast_fp16")];
            tensor<fp16, []> var_44582_to_fp16 = const()[name = tensor<string, []>("op_44582_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3795_cast_fp16 = mul(x = var_44581_cast_fp16, y = var_44582_to_fp16)[name = tensor<string, []>("aw_3795_cast_fp16")];
            tensor<string, []> var_44585_equation_0 = const()[name = tensor<string, []>("op_44585_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44585_cast_fp16 = einsum(equation = var_44585_equation_0, values = (var_44427_cast_fp16, var_44344_cast_fp16))[name = tensor<string, []>("op_44585_cast_fp16")];
            tensor<fp16, []> var_44586_to_fp16 = const()[name = tensor<string, []>("op_44586_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3797_cast_fp16 = mul(x = var_44585_cast_fp16, y = var_44586_to_fp16)[name = tensor<string, []>("aw_3797_cast_fp16")];
            tensor<string, []> var_44589_equation_0 = const()[name = tensor<string, []>("op_44589_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44589_cast_fp16 = einsum(equation = var_44589_equation_0, values = (var_44431_cast_fp16, var_44348_cast_fp16))[name = tensor<string, []>("op_44589_cast_fp16")];
            tensor<fp16, []> var_44590_to_fp16 = const()[name = tensor<string, []>("op_44590_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3799_cast_fp16 = mul(x = var_44589_cast_fp16, y = var_44590_to_fp16)[name = tensor<string, []>("aw_3799_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44592_cast_fp16 = softmax(axis = var_30355, x = aw_3761_cast_fp16)[name = tensor<string, []>("op_44592_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44593_cast_fp16 = softmax(axis = var_30355, x = aw_3763_cast_fp16)[name = tensor<string, []>("op_44593_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44594_cast_fp16 = softmax(axis = var_30355, x = aw_3765_cast_fp16)[name = tensor<string, []>("op_44594_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44595_cast_fp16 = softmax(axis = var_30355, x = aw_3767_cast_fp16)[name = tensor<string, []>("op_44595_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44596_cast_fp16 = softmax(axis = var_30355, x = aw_3769_cast_fp16)[name = tensor<string, []>("op_44596_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44597_cast_fp16 = softmax(axis = var_30355, x = aw_3771_cast_fp16)[name = tensor<string, []>("op_44597_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44598_cast_fp16 = softmax(axis = var_30355, x = aw_3773_cast_fp16)[name = tensor<string, []>("op_44598_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44599_cast_fp16 = softmax(axis = var_30355, x = aw_3775_cast_fp16)[name = tensor<string, []>("op_44599_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44600_cast_fp16 = softmax(axis = var_30355, x = aw_3777_cast_fp16)[name = tensor<string, []>("op_44600_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44601_cast_fp16 = softmax(axis = var_30355, x = aw_3779_cast_fp16)[name = tensor<string, []>("op_44601_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44602_cast_fp16 = softmax(axis = var_30355, x = aw_3781_cast_fp16)[name = tensor<string, []>("op_44602_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44603_cast_fp16 = softmax(axis = var_30355, x = aw_3783_cast_fp16)[name = tensor<string, []>("op_44603_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44604_cast_fp16 = softmax(axis = var_30355, x = aw_3785_cast_fp16)[name = tensor<string, []>("op_44604_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44605_cast_fp16 = softmax(axis = var_30355, x = aw_3787_cast_fp16)[name = tensor<string, []>("op_44605_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44606_cast_fp16 = softmax(axis = var_30355, x = aw_3789_cast_fp16)[name = tensor<string, []>("op_44606_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44607_cast_fp16 = softmax(axis = var_30355, x = aw_3791_cast_fp16)[name = tensor<string, []>("op_44607_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44608_cast_fp16 = softmax(axis = var_30355, x = aw_3793_cast_fp16)[name = tensor<string, []>("op_44608_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44609_cast_fp16 = softmax(axis = var_30355, x = aw_3795_cast_fp16)[name = tensor<string, []>("op_44609_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44610_cast_fp16 = softmax(axis = var_30355, x = aw_3797_cast_fp16)[name = tensor<string, []>("op_44610_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44611_cast_fp16 = softmax(axis = var_30355, x = aw_3799_cast_fp16)[name = tensor<string, []>("op_44611_cast_fp16")];
            tensor<string, []> var_44613_equation_0 = const()[name = tensor<string, []>("op_44613_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44613_cast_fp16 = einsum(equation = var_44613_equation_0, values = (var_44433_cast_fp16, var_44592_cast_fp16))[name = tensor<string, []>("op_44613_cast_fp16")];
            tensor<string, []> var_44615_equation_0 = const()[name = tensor<string, []>("op_44615_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44615_cast_fp16 = einsum(equation = var_44615_equation_0, values = (var_44437_cast_fp16, var_44593_cast_fp16))[name = tensor<string, []>("op_44615_cast_fp16")];
            tensor<string, []> var_44617_equation_0 = const()[name = tensor<string, []>("op_44617_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44617_cast_fp16 = einsum(equation = var_44617_equation_0, values = (var_44441_cast_fp16, var_44594_cast_fp16))[name = tensor<string, []>("op_44617_cast_fp16")];
            tensor<string, []> var_44619_equation_0 = const()[name = tensor<string, []>("op_44619_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44619_cast_fp16 = einsum(equation = var_44619_equation_0, values = (var_44445_cast_fp16, var_44595_cast_fp16))[name = tensor<string, []>("op_44619_cast_fp16")];
            tensor<string, []> var_44621_equation_0 = const()[name = tensor<string, []>("op_44621_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44621_cast_fp16 = einsum(equation = var_44621_equation_0, values = (var_44449_cast_fp16, var_44596_cast_fp16))[name = tensor<string, []>("op_44621_cast_fp16")];
            tensor<string, []> var_44623_equation_0 = const()[name = tensor<string, []>("op_44623_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44623_cast_fp16 = einsum(equation = var_44623_equation_0, values = (var_44453_cast_fp16, var_44597_cast_fp16))[name = tensor<string, []>("op_44623_cast_fp16")];
            tensor<string, []> var_44625_equation_0 = const()[name = tensor<string, []>("op_44625_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44625_cast_fp16 = einsum(equation = var_44625_equation_0, values = (var_44457_cast_fp16, var_44598_cast_fp16))[name = tensor<string, []>("op_44625_cast_fp16")];
            tensor<string, []> var_44627_equation_0 = const()[name = tensor<string, []>("op_44627_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44627_cast_fp16 = einsum(equation = var_44627_equation_0, values = (var_44461_cast_fp16, var_44599_cast_fp16))[name = tensor<string, []>("op_44627_cast_fp16")];
            tensor<string, []> var_44629_equation_0 = const()[name = tensor<string, []>("op_44629_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44629_cast_fp16 = einsum(equation = var_44629_equation_0, values = (var_44465_cast_fp16, var_44600_cast_fp16))[name = tensor<string, []>("op_44629_cast_fp16")];
            tensor<string, []> var_44631_equation_0 = const()[name = tensor<string, []>("op_44631_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44631_cast_fp16 = einsum(equation = var_44631_equation_0, values = (var_44469_cast_fp16, var_44601_cast_fp16))[name = tensor<string, []>("op_44631_cast_fp16")];
            tensor<string, []> var_44633_equation_0 = const()[name = tensor<string, []>("op_44633_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44633_cast_fp16 = einsum(equation = var_44633_equation_0, values = (var_44473_cast_fp16, var_44602_cast_fp16))[name = tensor<string, []>("op_44633_cast_fp16")];
            tensor<string, []> var_44635_equation_0 = const()[name = tensor<string, []>("op_44635_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44635_cast_fp16 = einsum(equation = var_44635_equation_0, values = (var_44477_cast_fp16, var_44603_cast_fp16))[name = tensor<string, []>("op_44635_cast_fp16")];
            tensor<string, []> var_44637_equation_0 = const()[name = tensor<string, []>("op_44637_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44637_cast_fp16 = einsum(equation = var_44637_equation_0, values = (var_44481_cast_fp16, var_44604_cast_fp16))[name = tensor<string, []>("op_44637_cast_fp16")];
            tensor<string, []> var_44639_equation_0 = const()[name = tensor<string, []>("op_44639_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44639_cast_fp16 = einsum(equation = var_44639_equation_0, values = (var_44485_cast_fp16, var_44605_cast_fp16))[name = tensor<string, []>("op_44639_cast_fp16")];
            tensor<string, []> var_44641_equation_0 = const()[name = tensor<string, []>("op_44641_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44641_cast_fp16 = einsum(equation = var_44641_equation_0, values = (var_44489_cast_fp16, var_44606_cast_fp16))[name = tensor<string, []>("op_44641_cast_fp16")];
            tensor<string, []> var_44643_equation_0 = const()[name = tensor<string, []>("op_44643_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44643_cast_fp16 = einsum(equation = var_44643_equation_0, values = (var_44493_cast_fp16, var_44607_cast_fp16))[name = tensor<string, []>("op_44643_cast_fp16")];
            tensor<string, []> var_44645_equation_0 = const()[name = tensor<string, []>("op_44645_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44645_cast_fp16 = einsum(equation = var_44645_equation_0, values = (var_44497_cast_fp16, var_44608_cast_fp16))[name = tensor<string, []>("op_44645_cast_fp16")];
            tensor<string, []> var_44647_equation_0 = const()[name = tensor<string, []>("op_44647_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44647_cast_fp16 = einsum(equation = var_44647_equation_0, values = (var_44501_cast_fp16, var_44609_cast_fp16))[name = tensor<string, []>("op_44647_cast_fp16")];
            tensor<string, []> var_44649_equation_0 = const()[name = tensor<string, []>("op_44649_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44649_cast_fp16 = einsum(equation = var_44649_equation_0, values = (var_44505_cast_fp16, var_44610_cast_fp16))[name = tensor<string, []>("op_44649_cast_fp16")];
            tensor<string, []> var_44651_equation_0 = const()[name = tensor<string, []>("op_44651_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44651_cast_fp16 = einsum(equation = var_44651_equation_0, values = (var_44509_cast_fp16, var_44611_cast_fp16))[name = tensor<string, []>("op_44651_cast_fp16")];
            tensor<bool, []> input_577_interleave_0 = const()[name = tensor<string, []>("input_577_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_577_cast_fp16 = concat(axis = var_30355, interleave = input_577_interleave_0, values = (var_44613_cast_fp16, var_44615_cast_fp16, var_44617_cast_fp16, var_44619_cast_fp16, var_44621_cast_fp16, var_44623_cast_fp16, var_44625_cast_fp16, var_44627_cast_fp16, var_44629_cast_fp16, var_44631_cast_fp16, var_44633_cast_fp16, var_44635_cast_fp16, var_44637_cast_fp16, var_44639_cast_fp16, var_44641_cast_fp16, var_44643_cast_fp16, var_44645_cast_fp16, var_44647_cast_fp16, var_44649_cast_fp16, var_44651_cast_fp16))[name = tensor<string, []>("input_577_cast_fp16")];
            tensor<string, []> var_44661_pad_type_0 = const()[name = tensor<string, []>("op_44661_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_44661_strides_0 = const()[name = tensor<string, []>("op_44661_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_44661_pad_0 = const()[name = tensor<string, []>("op_44661_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_44661_dilations_0 = const()[name = tensor<string, []>("op_44661_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_44661_groups_0 = const()[name = tensor<string, []>("op_44661_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_5_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(438228544))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(439457408))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_5_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_5_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_5_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(439457600)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_44661_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_5_attn1_to_out_0_bias_to_fp16, dilations = var_44661_dilations_0, groups = var_44661_groups_0, pad = var_44661_pad_0, pad_type = var_44661_pad_type_0, strides = var_44661_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_5_attn1_to_out_0_weight_to_fp16_palettized, x = input_577_cast_fp16)[name = tensor<string, []>("op_44661_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_297_cast_fp16 = add(x = var_44661_cast_fp16, y = inputs_295_cast_fp16)[name = tensor<string, []>("inputs_297_cast_fp16")];
            tensor<int32, [1]> hidden_states_393_axes_0 = const()[name = tensor<string, []>("hidden_states_393_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_393_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_393_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(439460224)))];
            tensor<fp16, [1280]> hidden_states_393_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_393_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(439462848)))];
            tensor<fp16, []> var_44671_to_fp16 = const()[name = tensor<string, []>("op_44671_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_393_cast_fp16 = layer_norm(axes = hidden_states_393_axes_0, beta = hidden_states_393_beta_0_to_fp16, epsilon = var_44671_to_fp16, gamma = hidden_states_393_gamma_0_to_fp16, x = inputs_297_cast_fp16)[name = tensor<string, []>("hidden_states_393_cast_fp16")];
            tensor<string, []> q_199_pad_type_0 = const()[name = tensor<string, []>("q_199_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_199_strides_0 = const()[name = tensor<string, []>("q_199_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_199_pad_0 = const()[name = tensor<string, []>("q_199_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_199_dilations_0 = const()[name = tensor<string, []>("q_199_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_199_groups_0 = const()[name = tensor<string, []>("q_199_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_5_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(439465472))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(440694336))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_5_attn2_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_199_cast_fp16 = conv(dilations = q_199_dilations_0, groups = q_199_groups_0, pad = q_199_pad_0, pad_type = q_199_pad_type_0, strides = q_199_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_5_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_393_cast_fp16)[name = tensor<string, []>("q_199_cast_fp16")];
            tensor<string, []> k_397_pad_type_0 = const()[name = tensor<string, []>("k_397_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_397_strides_0 = const()[name = tensor<string, []>("k_397_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_397_pad_0 = const()[name = tensor<string, []>("k_397_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_397_dilations_0 = const()[name = tensor<string, []>("k_397_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_397_groups_0 = const()[name = tensor<string, []>("k_397_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_5_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(440694528))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(442660672))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_5_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> k_397_cast_fp16 = conv(dilations = k_397_dilations_0, groups = k_397_groups_0, pad = k_397_pad_0, pad_type = k_397_pad_type_0, strides = k_397_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_5_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("k_397_cast_fp16")];
            tensor<string, []> v_199_pad_type_0 = const()[name = tensor<string, []>("v_199_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_199_strides_0 = const()[name = tensor<string, []>("v_199_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_199_pad_0 = const()[name = tensor<string, []>("v_199_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_199_dilations_0 = const()[name = tensor<string, []>("v_199_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_199_groups_0 = const()[name = tensor<string, []>("v_199_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_5_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(442660864))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(444627008))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_5_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> v_199_cast_fp16 = conv(dilations = v_199_dilations_0, groups = v_199_groups_0, pad = v_199_pad_0, pad_type = v_199_pad_type_0, strides = v_199_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_5_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("v_199_cast_fp16")];
            tensor<int32, [4]> var_44704_begin_0 = const()[name = tensor<string, []>("op_44704_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_44704_end_0 = const()[name = tensor<string, []>("op_44704_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_44704_end_mask_0 = const()[name = tensor<string, []>("op_44704_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44704_cast_fp16 = slice_by_index(begin = var_44704_begin_0, end = var_44704_end_0, end_mask = var_44704_end_mask_0, x = q_199_cast_fp16)[name = tensor<string, []>("op_44704_cast_fp16")];
            tensor<int32, [4]> var_44708_begin_0 = const()[name = tensor<string, []>("op_44708_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_44708_end_0 = const()[name = tensor<string, []>("op_44708_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_44708_end_mask_0 = const()[name = tensor<string, []>("op_44708_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44708_cast_fp16 = slice_by_index(begin = var_44708_begin_0, end = var_44708_end_0, end_mask = var_44708_end_mask_0, x = q_199_cast_fp16)[name = tensor<string, []>("op_44708_cast_fp16")];
            tensor<int32, [4]> var_44712_begin_0 = const()[name = tensor<string, []>("op_44712_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_44712_end_0 = const()[name = tensor<string, []>("op_44712_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_44712_end_mask_0 = const()[name = tensor<string, []>("op_44712_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44712_cast_fp16 = slice_by_index(begin = var_44712_begin_0, end = var_44712_end_0, end_mask = var_44712_end_mask_0, x = q_199_cast_fp16)[name = tensor<string, []>("op_44712_cast_fp16")];
            tensor<int32, [4]> var_44716_begin_0 = const()[name = tensor<string, []>("op_44716_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_44716_end_0 = const()[name = tensor<string, []>("op_44716_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_44716_end_mask_0 = const()[name = tensor<string, []>("op_44716_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44716_cast_fp16 = slice_by_index(begin = var_44716_begin_0, end = var_44716_end_0, end_mask = var_44716_end_mask_0, x = q_199_cast_fp16)[name = tensor<string, []>("op_44716_cast_fp16")];
            tensor<int32, [4]> var_44720_begin_0 = const()[name = tensor<string, []>("op_44720_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_44720_end_0 = const()[name = tensor<string, []>("op_44720_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_44720_end_mask_0 = const()[name = tensor<string, []>("op_44720_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44720_cast_fp16 = slice_by_index(begin = var_44720_begin_0, end = var_44720_end_0, end_mask = var_44720_end_mask_0, x = q_199_cast_fp16)[name = tensor<string, []>("op_44720_cast_fp16")];
            tensor<int32, [4]> var_44724_begin_0 = const()[name = tensor<string, []>("op_44724_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_44724_end_0 = const()[name = tensor<string, []>("op_44724_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_44724_end_mask_0 = const()[name = tensor<string, []>("op_44724_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44724_cast_fp16 = slice_by_index(begin = var_44724_begin_0, end = var_44724_end_0, end_mask = var_44724_end_mask_0, x = q_199_cast_fp16)[name = tensor<string, []>("op_44724_cast_fp16")];
            tensor<int32, [4]> var_44728_begin_0 = const()[name = tensor<string, []>("op_44728_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_44728_end_0 = const()[name = tensor<string, []>("op_44728_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_44728_end_mask_0 = const()[name = tensor<string, []>("op_44728_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44728_cast_fp16 = slice_by_index(begin = var_44728_begin_0, end = var_44728_end_0, end_mask = var_44728_end_mask_0, x = q_199_cast_fp16)[name = tensor<string, []>("op_44728_cast_fp16")];
            tensor<int32, [4]> var_44732_begin_0 = const()[name = tensor<string, []>("op_44732_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_44732_end_0 = const()[name = tensor<string, []>("op_44732_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_44732_end_mask_0 = const()[name = tensor<string, []>("op_44732_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44732_cast_fp16 = slice_by_index(begin = var_44732_begin_0, end = var_44732_end_0, end_mask = var_44732_end_mask_0, x = q_199_cast_fp16)[name = tensor<string, []>("op_44732_cast_fp16")];
            tensor<int32, [4]> var_44736_begin_0 = const()[name = tensor<string, []>("op_44736_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_44736_end_0 = const()[name = tensor<string, []>("op_44736_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_44736_end_mask_0 = const()[name = tensor<string, []>("op_44736_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44736_cast_fp16 = slice_by_index(begin = var_44736_begin_0, end = var_44736_end_0, end_mask = var_44736_end_mask_0, x = q_199_cast_fp16)[name = tensor<string, []>("op_44736_cast_fp16")];
            tensor<int32, [4]> var_44740_begin_0 = const()[name = tensor<string, []>("op_44740_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_44740_end_0 = const()[name = tensor<string, []>("op_44740_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_44740_end_mask_0 = const()[name = tensor<string, []>("op_44740_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44740_cast_fp16 = slice_by_index(begin = var_44740_begin_0, end = var_44740_end_0, end_mask = var_44740_end_mask_0, x = q_199_cast_fp16)[name = tensor<string, []>("op_44740_cast_fp16")];
            tensor<int32, [4]> var_44744_begin_0 = const()[name = tensor<string, []>("op_44744_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_44744_end_0 = const()[name = tensor<string, []>("op_44744_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_44744_end_mask_0 = const()[name = tensor<string, []>("op_44744_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44744_cast_fp16 = slice_by_index(begin = var_44744_begin_0, end = var_44744_end_0, end_mask = var_44744_end_mask_0, x = q_199_cast_fp16)[name = tensor<string, []>("op_44744_cast_fp16")];
            tensor<int32, [4]> var_44748_begin_0 = const()[name = tensor<string, []>("op_44748_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_44748_end_0 = const()[name = tensor<string, []>("op_44748_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_44748_end_mask_0 = const()[name = tensor<string, []>("op_44748_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44748_cast_fp16 = slice_by_index(begin = var_44748_begin_0, end = var_44748_end_0, end_mask = var_44748_end_mask_0, x = q_199_cast_fp16)[name = tensor<string, []>("op_44748_cast_fp16")];
            tensor<int32, [4]> var_44752_begin_0 = const()[name = tensor<string, []>("op_44752_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_44752_end_0 = const()[name = tensor<string, []>("op_44752_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_44752_end_mask_0 = const()[name = tensor<string, []>("op_44752_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44752_cast_fp16 = slice_by_index(begin = var_44752_begin_0, end = var_44752_end_0, end_mask = var_44752_end_mask_0, x = q_199_cast_fp16)[name = tensor<string, []>("op_44752_cast_fp16")];
            tensor<int32, [4]> var_44756_begin_0 = const()[name = tensor<string, []>("op_44756_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_44756_end_0 = const()[name = tensor<string, []>("op_44756_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_44756_end_mask_0 = const()[name = tensor<string, []>("op_44756_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44756_cast_fp16 = slice_by_index(begin = var_44756_begin_0, end = var_44756_end_0, end_mask = var_44756_end_mask_0, x = q_199_cast_fp16)[name = tensor<string, []>("op_44756_cast_fp16")];
            tensor<int32, [4]> var_44760_begin_0 = const()[name = tensor<string, []>("op_44760_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_44760_end_0 = const()[name = tensor<string, []>("op_44760_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_44760_end_mask_0 = const()[name = tensor<string, []>("op_44760_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44760_cast_fp16 = slice_by_index(begin = var_44760_begin_0, end = var_44760_end_0, end_mask = var_44760_end_mask_0, x = q_199_cast_fp16)[name = tensor<string, []>("op_44760_cast_fp16")];
            tensor<int32, [4]> var_44764_begin_0 = const()[name = tensor<string, []>("op_44764_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_44764_end_0 = const()[name = tensor<string, []>("op_44764_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_44764_end_mask_0 = const()[name = tensor<string, []>("op_44764_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44764_cast_fp16 = slice_by_index(begin = var_44764_begin_0, end = var_44764_end_0, end_mask = var_44764_end_mask_0, x = q_199_cast_fp16)[name = tensor<string, []>("op_44764_cast_fp16")];
            tensor<int32, [4]> var_44768_begin_0 = const()[name = tensor<string, []>("op_44768_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_44768_end_0 = const()[name = tensor<string, []>("op_44768_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_44768_end_mask_0 = const()[name = tensor<string, []>("op_44768_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44768_cast_fp16 = slice_by_index(begin = var_44768_begin_0, end = var_44768_end_0, end_mask = var_44768_end_mask_0, x = q_199_cast_fp16)[name = tensor<string, []>("op_44768_cast_fp16")];
            tensor<int32, [4]> var_44772_begin_0 = const()[name = tensor<string, []>("op_44772_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_44772_end_0 = const()[name = tensor<string, []>("op_44772_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_44772_end_mask_0 = const()[name = tensor<string, []>("op_44772_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44772_cast_fp16 = slice_by_index(begin = var_44772_begin_0, end = var_44772_end_0, end_mask = var_44772_end_mask_0, x = q_199_cast_fp16)[name = tensor<string, []>("op_44772_cast_fp16")];
            tensor<int32, [4]> var_44776_begin_0 = const()[name = tensor<string, []>("op_44776_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_44776_end_0 = const()[name = tensor<string, []>("op_44776_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_44776_end_mask_0 = const()[name = tensor<string, []>("op_44776_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44776_cast_fp16 = slice_by_index(begin = var_44776_begin_0, end = var_44776_end_0, end_mask = var_44776_end_mask_0, x = q_199_cast_fp16)[name = tensor<string, []>("op_44776_cast_fp16")];
            tensor<int32, [4]> var_44780_begin_0 = const()[name = tensor<string, []>("op_44780_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_44780_end_0 = const()[name = tensor<string, []>("op_44780_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_44780_end_mask_0 = const()[name = tensor<string, []>("op_44780_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44780_cast_fp16 = slice_by_index(begin = var_44780_begin_0, end = var_44780_end_0, end_mask = var_44780_end_mask_0, x = q_199_cast_fp16)[name = tensor<string, []>("op_44780_cast_fp16")];
            tensor<int32, [4]> k_399_perm_0 = const()[name = tensor<string, []>("k_399_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_44787_begin_0 = const()[name = tensor<string, []>("op_44787_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_44787_end_0 = const()[name = tensor<string, []>("op_44787_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_44787_end_mask_0 = const()[name = tensor<string, []>("op_44787_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> k_399_cast_fp16 = transpose(perm = k_399_perm_0, x = k_397_cast_fp16)[name = tensor<string, []>("transpose_40")];
            tensor<fp16, [2, 77, 1, 64]> var_44787_cast_fp16 = slice_by_index(begin = var_44787_begin_0, end = var_44787_end_0, end_mask = var_44787_end_mask_0, x = k_399_cast_fp16)[name = tensor<string, []>("op_44787_cast_fp16")];
            tensor<int32, [4]> var_44791_begin_0 = const()[name = tensor<string, []>("op_44791_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_44791_end_0 = const()[name = tensor<string, []>("op_44791_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_44791_end_mask_0 = const()[name = tensor<string, []>("op_44791_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_44791_cast_fp16 = slice_by_index(begin = var_44791_begin_0, end = var_44791_end_0, end_mask = var_44791_end_mask_0, x = k_399_cast_fp16)[name = tensor<string, []>("op_44791_cast_fp16")];
            tensor<int32, [4]> var_44795_begin_0 = const()[name = tensor<string, []>("op_44795_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_44795_end_0 = const()[name = tensor<string, []>("op_44795_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_44795_end_mask_0 = const()[name = tensor<string, []>("op_44795_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_44795_cast_fp16 = slice_by_index(begin = var_44795_begin_0, end = var_44795_end_0, end_mask = var_44795_end_mask_0, x = k_399_cast_fp16)[name = tensor<string, []>("op_44795_cast_fp16")];
            tensor<int32, [4]> var_44799_begin_0 = const()[name = tensor<string, []>("op_44799_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_44799_end_0 = const()[name = tensor<string, []>("op_44799_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_44799_end_mask_0 = const()[name = tensor<string, []>("op_44799_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_44799_cast_fp16 = slice_by_index(begin = var_44799_begin_0, end = var_44799_end_0, end_mask = var_44799_end_mask_0, x = k_399_cast_fp16)[name = tensor<string, []>("op_44799_cast_fp16")];
            tensor<int32, [4]> var_44803_begin_0 = const()[name = tensor<string, []>("op_44803_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_44803_end_0 = const()[name = tensor<string, []>("op_44803_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_44803_end_mask_0 = const()[name = tensor<string, []>("op_44803_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_44803_cast_fp16 = slice_by_index(begin = var_44803_begin_0, end = var_44803_end_0, end_mask = var_44803_end_mask_0, x = k_399_cast_fp16)[name = tensor<string, []>("op_44803_cast_fp16")];
            tensor<int32, [4]> var_44807_begin_0 = const()[name = tensor<string, []>("op_44807_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_44807_end_0 = const()[name = tensor<string, []>("op_44807_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_44807_end_mask_0 = const()[name = tensor<string, []>("op_44807_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_44807_cast_fp16 = slice_by_index(begin = var_44807_begin_0, end = var_44807_end_0, end_mask = var_44807_end_mask_0, x = k_399_cast_fp16)[name = tensor<string, []>("op_44807_cast_fp16")];
            tensor<int32, [4]> var_44811_begin_0 = const()[name = tensor<string, []>("op_44811_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_44811_end_0 = const()[name = tensor<string, []>("op_44811_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_44811_end_mask_0 = const()[name = tensor<string, []>("op_44811_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_44811_cast_fp16 = slice_by_index(begin = var_44811_begin_0, end = var_44811_end_0, end_mask = var_44811_end_mask_0, x = k_399_cast_fp16)[name = tensor<string, []>("op_44811_cast_fp16")];
            tensor<int32, [4]> var_44815_begin_0 = const()[name = tensor<string, []>("op_44815_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_44815_end_0 = const()[name = tensor<string, []>("op_44815_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_44815_end_mask_0 = const()[name = tensor<string, []>("op_44815_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_44815_cast_fp16 = slice_by_index(begin = var_44815_begin_0, end = var_44815_end_0, end_mask = var_44815_end_mask_0, x = k_399_cast_fp16)[name = tensor<string, []>("op_44815_cast_fp16")];
            tensor<int32, [4]> var_44819_begin_0 = const()[name = tensor<string, []>("op_44819_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_44819_end_0 = const()[name = tensor<string, []>("op_44819_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_44819_end_mask_0 = const()[name = tensor<string, []>("op_44819_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_44819_cast_fp16 = slice_by_index(begin = var_44819_begin_0, end = var_44819_end_0, end_mask = var_44819_end_mask_0, x = k_399_cast_fp16)[name = tensor<string, []>("op_44819_cast_fp16")];
            tensor<int32, [4]> var_44823_begin_0 = const()[name = tensor<string, []>("op_44823_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_44823_end_0 = const()[name = tensor<string, []>("op_44823_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_44823_end_mask_0 = const()[name = tensor<string, []>("op_44823_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_44823_cast_fp16 = slice_by_index(begin = var_44823_begin_0, end = var_44823_end_0, end_mask = var_44823_end_mask_0, x = k_399_cast_fp16)[name = tensor<string, []>("op_44823_cast_fp16")];
            tensor<int32, [4]> var_44827_begin_0 = const()[name = tensor<string, []>("op_44827_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_44827_end_0 = const()[name = tensor<string, []>("op_44827_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_44827_end_mask_0 = const()[name = tensor<string, []>("op_44827_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_44827_cast_fp16 = slice_by_index(begin = var_44827_begin_0, end = var_44827_end_0, end_mask = var_44827_end_mask_0, x = k_399_cast_fp16)[name = tensor<string, []>("op_44827_cast_fp16")];
            tensor<int32, [4]> var_44831_begin_0 = const()[name = tensor<string, []>("op_44831_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_44831_end_0 = const()[name = tensor<string, []>("op_44831_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_44831_end_mask_0 = const()[name = tensor<string, []>("op_44831_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_44831_cast_fp16 = slice_by_index(begin = var_44831_begin_0, end = var_44831_end_0, end_mask = var_44831_end_mask_0, x = k_399_cast_fp16)[name = tensor<string, []>("op_44831_cast_fp16")];
            tensor<int32, [4]> var_44835_begin_0 = const()[name = tensor<string, []>("op_44835_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_44835_end_0 = const()[name = tensor<string, []>("op_44835_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_44835_end_mask_0 = const()[name = tensor<string, []>("op_44835_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_44835_cast_fp16 = slice_by_index(begin = var_44835_begin_0, end = var_44835_end_0, end_mask = var_44835_end_mask_0, x = k_399_cast_fp16)[name = tensor<string, []>("op_44835_cast_fp16")];
            tensor<int32, [4]> var_44839_begin_0 = const()[name = tensor<string, []>("op_44839_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_44839_end_0 = const()[name = tensor<string, []>("op_44839_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_44839_end_mask_0 = const()[name = tensor<string, []>("op_44839_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_44839_cast_fp16 = slice_by_index(begin = var_44839_begin_0, end = var_44839_end_0, end_mask = var_44839_end_mask_0, x = k_399_cast_fp16)[name = tensor<string, []>("op_44839_cast_fp16")];
            tensor<int32, [4]> var_44843_begin_0 = const()[name = tensor<string, []>("op_44843_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_44843_end_0 = const()[name = tensor<string, []>("op_44843_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_44843_end_mask_0 = const()[name = tensor<string, []>("op_44843_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_44843_cast_fp16 = slice_by_index(begin = var_44843_begin_0, end = var_44843_end_0, end_mask = var_44843_end_mask_0, x = k_399_cast_fp16)[name = tensor<string, []>("op_44843_cast_fp16")];
            tensor<int32, [4]> var_44847_begin_0 = const()[name = tensor<string, []>("op_44847_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_44847_end_0 = const()[name = tensor<string, []>("op_44847_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_44847_end_mask_0 = const()[name = tensor<string, []>("op_44847_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_44847_cast_fp16 = slice_by_index(begin = var_44847_begin_0, end = var_44847_end_0, end_mask = var_44847_end_mask_0, x = k_399_cast_fp16)[name = tensor<string, []>("op_44847_cast_fp16")];
            tensor<int32, [4]> var_44851_begin_0 = const()[name = tensor<string, []>("op_44851_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_44851_end_0 = const()[name = tensor<string, []>("op_44851_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_44851_end_mask_0 = const()[name = tensor<string, []>("op_44851_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_44851_cast_fp16 = slice_by_index(begin = var_44851_begin_0, end = var_44851_end_0, end_mask = var_44851_end_mask_0, x = k_399_cast_fp16)[name = tensor<string, []>("op_44851_cast_fp16")];
            tensor<int32, [4]> var_44855_begin_0 = const()[name = tensor<string, []>("op_44855_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_44855_end_0 = const()[name = tensor<string, []>("op_44855_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_44855_end_mask_0 = const()[name = tensor<string, []>("op_44855_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_44855_cast_fp16 = slice_by_index(begin = var_44855_begin_0, end = var_44855_end_0, end_mask = var_44855_end_mask_0, x = k_399_cast_fp16)[name = tensor<string, []>("op_44855_cast_fp16")];
            tensor<int32, [4]> var_44859_begin_0 = const()[name = tensor<string, []>("op_44859_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_44859_end_0 = const()[name = tensor<string, []>("op_44859_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_44859_end_mask_0 = const()[name = tensor<string, []>("op_44859_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_44859_cast_fp16 = slice_by_index(begin = var_44859_begin_0, end = var_44859_end_0, end_mask = var_44859_end_mask_0, x = k_399_cast_fp16)[name = tensor<string, []>("op_44859_cast_fp16")];
            tensor<int32, [4]> var_44863_begin_0 = const()[name = tensor<string, []>("op_44863_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_44863_end_0 = const()[name = tensor<string, []>("op_44863_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_44863_end_mask_0 = const()[name = tensor<string, []>("op_44863_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_44863_cast_fp16 = slice_by_index(begin = var_44863_begin_0, end = var_44863_end_0, end_mask = var_44863_end_mask_0, x = k_399_cast_fp16)[name = tensor<string, []>("op_44863_cast_fp16")];
            tensor<int32, [4]> var_44865_begin_0 = const()[name = tensor<string, []>("op_44865_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_44865_end_0 = const()[name = tensor<string, []>("op_44865_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_44865_end_mask_0 = const()[name = tensor<string, []>("op_44865_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44865_cast_fp16 = slice_by_index(begin = var_44865_begin_0, end = var_44865_end_0, end_mask = var_44865_end_mask_0, x = v_199_cast_fp16)[name = tensor<string, []>("op_44865_cast_fp16")];
            tensor<int32, [4]> var_44869_begin_0 = const()[name = tensor<string, []>("op_44869_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_44869_end_0 = const()[name = tensor<string, []>("op_44869_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_44869_end_mask_0 = const()[name = tensor<string, []>("op_44869_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44869_cast_fp16 = slice_by_index(begin = var_44869_begin_0, end = var_44869_end_0, end_mask = var_44869_end_mask_0, x = v_199_cast_fp16)[name = tensor<string, []>("op_44869_cast_fp16")];
            tensor<int32, [4]> var_44873_begin_0 = const()[name = tensor<string, []>("op_44873_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_44873_end_0 = const()[name = tensor<string, []>("op_44873_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_44873_end_mask_0 = const()[name = tensor<string, []>("op_44873_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44873_cast_fp16 = slice_by_index(begin = var_44873_begin_0, end = var_44873_end_0, end_mask = var_44873_end_mask_0, x = v_199_cast_fp16)[name = tensor<string, []>("op_44873_cast_fp16")];
            tensor<int32, [4]> var_44877_begin_0 = const()[name = tensor<string, []>("op_44877_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_44877_end_0 = const()[name = tensor<string, []>("op_44877_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_44877_end_mask_0 = const()[name = tensor<string, []>("op_44877_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44877_cast_fp16 = slice_by_index(begin = var_44877_begin_0, end = var_44877_end_0, end_mask = var_44877_end_mask_0, x = v_199_cast_fp16)[name = tensor<string, []>("op_44877_cast_fp16")];
            tensor<int32, [4]> var_44881_begin_0 = const()[name = tensor<string, []>("op_44881_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_44881_end_0 = const()[name = tensor<string, []>("op_44881_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_44881_end_mask_0 = const()[name = tensor<string, []>("op_44881_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44881_cast_fp16 = slice_by_index(begin = var_44881_begin_0, end = var_44881_end_0, end_mask = var_44881_end_mask_0, x = v_199_cast_fp16)[name = tensor<string, []>("op_44881_cast_fp16")];
            tensor<int32, [4]> var_44885_begin_0 = const()[name = tensor<string, []>("op_44885_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_44885_end_0 = const()[name = tensor<string, []>("op_44885_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_44885_end_mask_0 = const()[name = tensor<string, []>("op_44885_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44885_cast_fp16 = slice_by_index(begin = var_44885_begin_0, end = var_44885_end_0, end_mask = var_44885_end_mask_0, x = v_199_cast_fp16)[name = tensor<string, []>("op_44885_cast_fp16")];
            tensor<int32, [4]> var_44889_begin_0 = const()[name = tensor<string, []>("op_44889_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_44889_end_0 = const()[name = tensor<string, []>("op_44889_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_44889_end_mask_0 = const()[name = tensor<string, []>("op_44889_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44889_cast_fp16 = slice_by_index(begin = var_44889_begin_0, end = var_44889_end_0, end_mask = var_44889_end_mask_0, x = v_199_cast_fp16)[name = tensor<string, []>("op_44889_cast_fp16")];
            tensor<int32, [4]> var_44893_begin_0 = const()[name = tensor<string, []>("op_44893_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_44893_end_0 = const()[name = tensor<string, []>("op_44893_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_44893_end_mask_0 = const()[name = tensor<string, []>("op_44893_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44893_cast_fp16 = slice_by_index(begin = var_44893_begin_0, end = var_44893_end_0, end_mask = var_44893_end_mask_0, x = v_199_cast_fp16)[name = tensor<string, []>("op_44893_cast_fp16")];
            tensor<int32, [4]> var_44897_begin_0 = const()[name = tensor<string, []>("op_44897_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_44897_end_0 = const()[name = tensor<string, []>("op_44897_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_44897_end_mask_0 = const()[name = tensor<string, []>("op_44897_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44897_cast_fp16 = slice_by_index(begin = var_44897_begin_0, end = var_44897_end_0, end_mask = var_44897_end_mask_0, x = v_199_cast_fp16)[name = tensor<string, []>("op_44897_cast_fp16")];
            tensor<int32, [4]> var_44901_begin_0 = const()[name = tensor<string, []>("op_44901_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_44901_end_0 = const()[name = tensor<string, []>("op_44901_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_44901_end_mask_0 = const()[name = tensor<string, []>("op_44901_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44901_cast_fp16 = slice_by_index(begin = var_44901_begin_0, end = var_44901_end_0, end_mask = var_44901_end_mask_0, x = v_199_cast_fp16)[name = tensor<string, []>("op_44901_cast_fp16")];
            tensor<int32, [4]> var_44905_begin_0 = const()[name = tensor<string, []>("op_44905_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_44905_end_0 = const()[name = tensor<string, []>("op_44905_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_44905_end_mask_0 = const()[name = tensor<string, []>("op_44905_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44905_cast_fp16 = slice_by_index(begin = var_44905_begin_0, end = var_44905_end_0, end_mask = var_44905_end_mask_0, x = v_199_cast_fp16)[name = tensor<string, []>("op_44905_cast_fp16")];
            tensor<int32, [4]> var_44909_begin_0 = const()[name = tensor<string, []>("op_44909_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_44909_end_0 = const()[name = tensor<string, []>("op_44909_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_44909_end_mask_0 = const()[name = tensor<string, []>("op_44909_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44909_cast_fp16 = slice_by_index(begin = var_44909_begin_0, end = var_44909_end_0, end_mask = var_44909_end_mask_0, x = v_199_cast_fp16)[name = tensor<string, []>("op_44909_cast_fp16")];
            tensor<int32, [4]> var_44913_begin_0 = const()[name = tensor<string, []>("op_44913_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_44913_end_0 = const()[name = tensor<string, []>("op_44913_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_44913_end_mask_0 = const()[name = tensor<string, []>("op_44913_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44913_cast_fp16 = slice_by_index(begin = var_44913_begin_0, end = var_44913_end_0, end_mask = var_44913_end_mask_0, x = v_199_cast_fp16)[name = tensor<string, []>("op_44913_cast_fp16")];
            tensor<int32, [4]> var_44917_begin_0 = const()[name = tensor<string, []>("op_44917_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_44917_end_0 = const()[name = tensor<string, []>("op_44917_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_44917_end_mask_0 = const()[name = tensor<string, []>("op_44917_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44917_cast_fp16 = slice_by_index(begin = var_44917_begin_0, end = var_44917_end_0, end_mask = var_44917_end_mask_0, x = v_199_cast_fp16)[name = tensor<string, []>("op_44917_cast_fp16")];
            tensor<int32, [4]> var_44921_begin_0 = const()[name = tensor<string, []>("op_44921_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_44921_end_0 = const()[name = tensor<string, []>("op_44921_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_44921_end_mask_0 = const()[name = tensor<string, []>("op_44921_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44921_cast_fp16 = slice_by_index(begin = var_44921_begin_0, end = var_44921_end_0, end_mask = var_44921_end_mask_0, x = v_199_cast_fp16)[name = tensor<string, []>("op_44921_cast_fp16")];
            tensor<int32, [4]> var_44925_begin_0 = const()[name = tensor<string, []>("op_44925_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_44925_end_0 = const()[name = tensor<string, []>("op_44925_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_44925_end_mask_0 = const()[name = tensor<string, []>("op_44925_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44925_cast_fp16 = slice_by_index(begin = var_44925_begin_0, end = var_44925_end_0, end_mask = var_44925_end_mask_0, x = v_199_cast_fp16)[name = tensor<string, []>("op_44925_cast_fp16")];
            tensor<int32, [4]> var_44929_begin_0 = const()[name = tensor<string, []>("op_44929_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_44929_end_0 = const()[name = tensor<string, []>("op_44929_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_44929_end_mask_0 = const()[name = tensor<string, []>("op_44929_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44929_cast_fp16 = slice_by_index(begin = var_44929_begin_0, end = var_44929_end_0, end_mask = var_44929_end_mask_0, x = v_199_cast_fp16)[name = tensor<string, []>("op_44929_cast_fp16")];
            tensor<int32, [4]> var_44933_begin_0 = const()[name = tensor<string, []>("op_44933_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_44933_end_0 = const()[name = tensor<string, []>("op_44933_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_44933_end_mask_0 = const()[name = tensor<string, []>("op_44933_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44933_cast_fp16 = slice_by_index(begin = var_44933_begin_0, end = var_44933_end_0, end_mask = var_44933_end_mask_0, x = v_199_cast_fp16)[name = tensor<string, []>("op_44933_cast_fp16")];
            tensor<int32, [4]> var_44937_begin_0 = const()[name = tensor<string, []>("op_44937_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_44937_end_0 = const()[name = tensor<string, []>("op_44937_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_44937_end_mask_0 = const()[name = tensor<string, []>("op_44937_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44937_cast_fp16 = slice_by_index(begin = var_44937_begin_0, end = var_44937_end_0, end_mask = var_44937_end_mask_0, x = v_199_cast_fp16)[name = tensor<string, []>("op_44937_cast_fp16")];
            tensor<int32, [4]> var_44941_begin_0 = const()[name = tensor<string, []>("op_44941_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_44941_end_0 = const()[name = tensor<string, []>("op_44941_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_44941_end_mask_0 = const()[name = tensor<string, []>("op_44941_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44941_cast_fp16 = slice_by_index(begin = var_44941_begin_0, end = var_44941_end_0, end_mask = var_44941_end_mask_0, x = v_199_cast_fp16)[name = tensor<string, []>("op_44941_cast_fp16")];
            tensor<string, []> var_44945_equation_0 = const()[name = tensor<string, []>("op_44945_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44945_cast_fp16 = einsum(equation = var_44945_equation_0, values = (var_44787_cast_fp16, var_44704_cast_fp16))[name = tensor<string, []>("op_44945_cast_fp16")];
            tensor<fp16, []> var_44946_to_fp16 = const()[name = tensor<string, []>("op_44946_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3801_cast_fp16 = mul(x = var_44945_cast_fp16, y = var_44946_to_fp16)[name = tensor<string, []>("aw_3801_cast_fp16")];
            tensor<string, []> var_44949_equation_0 = const()[name = tensor<string, []>("op_44949_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44949_cast_fp16 = einsum(equation = var_44949_equation_0, values = (var_44791_cast_fp16, var_44708_cast_fp16))[name = tensor<string, []>("op_44949_cast_fp16")];
            tensor<fp16, []> var_44950_to_fp16 = const()[name = tensor<string, []>("op_44950_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3803_cast_fp16 = mul(x = var_44949_cast_fp16, y = var_44950_to_fp16)[name = tensor<string, []>("aw_3803_cast_fp16")];
            tensor<string, []> var_44953_equation_0 = const()[name = tensor<string, []>("op_44953_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44953_cast_fp16 = einsum(equation = var_44953_equation_0, values = (var_44795_cast_fp16, var_44712_cast_fp16))[name = tensor<string, []>("op_44953_cast_fp16")];
            tensor<fp16, []> var_44954_to_fp16 = const()[name = tensor<string, []>("op_44954_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3805_cast_fp16 = mul(x = var_44953_cast_fp16, y = var_44954_to_fp16)[name = tensor<string, []>("aw_3805_cast_fp16")];
            tensor<string, []> var_44957_equation_0 = const()[name = tensor<string, []>("op_44957_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44957_cast_fp16 = einsum(equation = var_44957_equation_0, values = (var_44799_cast_fp16, var_44716_cast_fp16))[name = tensor<string, []>("op_44957_cast_fp16")];
            tensor<fp16, []> var_44958_to_fp16 = const()[name = tensor<string, []>("op_44958_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3807_cast_fp16 = mul(x = var_44957_cast_fp16, y = var_44958_to_fp16)[name = tensor<string, []>("aw_3807_cast_fp16")];
            tensor<string, []> var_44961_equation_0 = const()[name = tensor<string, []>("op_44961_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44961_cast_fp16 = einsum(equation = var_44961_equation_0, values = (var_44803_cast_fp16, var_44720_cast_fp16))[name = tensor<string, []>("op_44961_cast_fp16")];
            tensor<fp16, []> var_44962_to_fp16 = const()[name = tensor<string, []>("op_44962_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3809_cast_fp16 = mul(x = var_44961_cast_fp16, y = var_44962_to_fp16)[name = tensor<string, []>("aw_3809_cast_fp16")];
            tensor<string, []> var_44965_equation_0 = const()[name = tensor<string, []>("op_44965_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44965_cast_fp16 = einsum(equation = var_44965_equation_0, values = (var_44807_cast_fp16, var_44724_cast_fp16))[name = tensor<string, []>("op_44965_cast_fp16")];
            tensor<fp16, []> var_44966_to_fp16 = const()[name = tensor<string, []>("op_44966_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3811_cast_fp16 = mul(x = var_44965_cast_fp16, y = var_44966_to_fp16)[name = tensor<string, []>("aw_3811_cast_fp16")];
            tensor<string, []> var_44969_equation_0 = const()[name = tensor<string, []>("op_44969_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44969_cast_fp16 = einsum(equation = var_44969_equation_0, values = (var_44811_cast_fp16, var_44728_cast_fp16))[name = tensor<string, []>("op_44969_cast_fp16")];
            tensor<fp16, []> var_44970_to_fp16 = const()[name = tensor<string, []>("op_44970_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3813_cast_fp16 = mul(x = var_44969_cast_fp16, y = var_44970_to_fp16)[name = tensor<string, []>("aw_3813_cast_fp16")];
            tensor<string, []> var_44973_equation_0 = const()[name = tensor<string, []>("op_44973_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44973_cast_fp16 = einsum(equation = var_44973_equation_0, values = (var_44815_cast_fp16, var_44732_cast_fp16))[name = tensor<string, []>("op_44973_cast_fp16")];
            tensor<fp16, []> var_44974_to_fp16 = const()[name = tensor<string, []>("op_44974_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3815_cast_fp16 = mul(x = var_44973_cast_fp16, y = var_44974_to_fp16)[name = tensor<string, []>("aw_3815_cast_fp16")];
            tensor<string, []> var_44977_equation_0 = const()[name = tensor<string, []>("op_44977_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44977_cast_fp16 = einsum(equation = var_44977_equation_0, values = (var_44819_cast_fp16, var_44736_cast_fp16))[name = tensor<string, []>("op_44977_cast_fp16")];
            tensor<fp16, []> var_44978_to_fp16 = const()[name = tensor<string, []>("op_44978_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3817_cast_fp16 = mul(x = var_44977_cast_fp16, y = var_44978_to_fp16)[name = tensor<string, []>("aw_3817_cast_fp16")];
            tensor<string, []> var_44981_equation_0 = const()[name = tensor<string, []>("op_44981_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44981_cast_fp16 = einsum(equation = var_44981_equation_0, values = (var_44823_cast_fp16, var_44740_cast_fp16))[name = tensor<string, []>("op_44981_cast_fp16")];
            tensor<fp16, []> var_44982_to_fp16 = const()[name = tensor<string, []>("op_44982_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3819_cast_fp16 = mul(x = var_44981_cast_fp16, y = var_44982_to_fp16)[name = tensor<string, []>("aw_3819_cast_fp16")];
            tensor<string, []> var_44985_equation_0 = const()[name = tensor<string, []>("op_44985_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44985_cast_fp16 = einsum(equation = var_44985_equation_0, values = (var_44827_cast_fp16, var_44744_cast_fp16))[name = tensor<string, []>("op_44985_cast_fp16")];
            tensor<fp16, []> var_44986_to_fp16 = const()[name = tensor<string, []>("op_44986_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3821_cast_fp16 = mul(x = var_44985_cast_fp16, y = var_44986_to_fp16)[name = tensor<string, []>("aw_3821_cast_fp16")];
            tensor<string, []> var_44989_equation_0 = const()[name = tensor<string, []>("op_44989_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44989_cast_fp16 = einsum(equation = var_44989_equation_0, values = (var_44831_cast_fp16, var_44748_cast_fp16))[name = tensor<string, []>("op_44989_cast_fp16")];
            tensor<fp16, []> var_44990_to_fp16 = const()[name = tensor<string, []>("op_44990_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3823_cast_fp16 = mul(x = var_44989_cast_fp16, y = var_44990_to_fp16)[name = tensor<string, []>("aw_3823_cast_fp16")];
            tensor<string, []> var_44993_equation_0 = const()[name = tensor<string, []>("op_44993_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44993_cast_fp16 = einsum(equation = var_44993_equation_0, values = (var_44835_cast_fp16, var_44752_cast_fp16))[name = tensor<string, []>("op_44993_cast_fp16")];
            tensor<fp16, []> var_44994_to_fp16 = const()[name = tensor<string, []>("op_44994_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3825_cast_fp16 = mul(x = var_44993_cast_fp16, y = var_44994_to_fp16)[name = tensor<string, []>("aw_3825_cast_fp16")];
            tensor<string, []> var_44997_equation_0 = const()[name = tensor<string, []>("op_44997_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44997_cast_fp16 = einsum(equation = var_44997_equation_0, values = (var_44839_cast_fp16, var_44756_cast_fp16))[name = tensor<string, []>("op_44997_cast_fp16")];
            tensor<fp16, []> var_44998_to_fp16 = const()[name = tensor<string, []>("op_44998_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3827_cast_fp16 = mul(x = var_44997_cast_fp16, y = var_44998_to_fp16)[name = tensor<string, []>("aw_3827_cast_fp16")];
            tensor<string, []> var_45001_equation_0 = const()[name = tensor<string, []>("op_45001_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45001_cast_fp16 = einsum(equation = var_45001_equation_0, values = (var_44843_cast_fp16, var_44760_cast_fp16))[name = tensor<string, []>("op_45001_cast_fp16")];
            tensor<fp16, []> var_45002_to_fp16 = const()[name = tensor<string, []>("op_45002_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3829_cast_fp16 = mul(x = var_45001_cast_fp16, y = var_45002_to_fp16)[name = tensor<string, []>("aw_3829_cast_fp16")];
            tensor<string, []> var_45005_equation_0 = const()[name = tensor<string, []>("op_45005_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45005_cast_fp16 = einsum(equation = var_45005_equation_0, values = (var_44847_cast_fp16, var_44764_cast_fp16))[name = tensor<string, []>("op_45005_cast_fp16")];
            tensor<fp16, []> var_45006_to_fp16 = const()[name = tensor<string, []>("op_45006_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3831_cast_fp16 = mul(x = var_45005_cast_fp16, y = var_45006_to_fp16)[name = tensor<string, []>("aw_3831_cast_fp16")];
            tensor<string, []> var_45009_equation_0 = const()[name = tensor<string, []>("op_45009_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45009_cast_fp16 = einsum(equation = var_45009_equation_0, values = (var_44851_cast_fp16, var_44768_cast_fp16))[name = tensor<string, []>("op_45009_cast_fp16")];
            tensor<fp16, []> var_45010_to_fp16 = const()[name = tensor<string, []>("op_45010_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3833_cast_fp16 = mul(x = var_45009_cast_fp16, y = var_45010_to_fp16)[name = tensor<string, []>("aw_3833_cast_fp16")];
            tensor<string, []> var_45013_equation_0 = const()[name = tensor<string, []>("op_45013_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45013_cast_fp16 = einsum(equation = var_45013_equation_0, values = (var_44855_cast_fp16, var_44772_cast_fp16))[name = tensor<string, []>("op_45013_cast_fp16")];
            tensor<fp16, []> var_45014_to_fp16 = const()[name = tensor<string, []>("op_45014_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3835_cast_fp16 = mul(x = var_45013_cast_fp16, y = var_45014_to_fp16)[name = tensor<string, []>("aw_3835_cast_fp16")];
            tensor<string, []> var_45017_equation_0 = const()[name = tensor<string, []>("op_45017_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45017_cast_fp16 = einsum(equation = var_45017_equation_0, values = (var_44859_cast_fp16, var_44776_cast_fp16))[name = tensor<string, []>("op_45017_cast_fp16")];
            tensor<fp16, []> var_45018_to_fp16 = const()[name = tensor<string, []>("op_45018_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3837_cast_fp16 = mul(x = var_45017_cast_fp16, y = var_45018_to_fp16)[name = tensor<string, []>("aw_3837_cast_fp16")];
            tensor<string, []> var_45021_equation_0 = const()[name = tensor<string, []>("op_45021_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45021_cast_fp16 = einsum(equation = var_45021_equation_0, values = (var_44863_cast_fp16, var_44780_cast_fp16))[name = tensor<string, []>("op_45021_cast_fp16")];
            tensor<fp16, []> var_45022_to_fp16 = const()[name = tensor<string, []>("op_45022_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3839_cast_fp16 = mul(x = var_45021_cast_fp16, y = var_45022_to_fp16)[name = tensor<string, []>("aw_3839_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_45024_cast_fp16 = softmax(axis = var_30355, x = aw_3801_cast_fp16)[name = tensor<string, []>("op_45024_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_45025_cast_fp16 = softmax(axis = var_30355, x = aw_3803_cast_fp16)[name = tensor<string, []>("op_45025_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_45026_cast_fp16 = softmax(axis = var_30355, x = aw_3805_cast_fp16)[name = tensor<string, []>("op_45026_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_45027_cast_fp16 = softmax(axis = var_30355, x = aw_3807_cast_fp16)[name = tensor<string, []>("op_45027_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_45028_cast_fp16 = softmax(axis = var_30355, x = aw_3809_cast_fp16)[name = tensor<string, []>("op_45028_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_45029_cast_fp16 = softmax(axis = var_30355, x = aw_3811_cast_fp16)[name = tensor<string, []>("op_45029_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_45030_cast_fp16 = softmax(axis = var_30355, x = aw_3813_cast_fp16)[name = tensor<string, []>("op_45030_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_45031_cast_fp16 = softmax(axis = var_30355, x = aw_3815_cast_fp16)[name = tensor<string, []>("op_45031_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_45032_cast_fp16 = softmax(axis = var_30355, x = aw_3817_cast_fp16)[name = tensor<string, []>("op_45032_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_45033_cast_fp16 = softmax(axis = var_30355, x = aw_3819_cast_fp16)[name = tensor<string, []>("op_45033_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_45034_cast_fp16 = softmax(axis = var_30355, x = aw_3821_cast_fp16)[name = tensor<string, []>("op_45034_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_45035_cast_fp16 = softmax(axis = var_30355, x = aw_3823_cast_fp16)[name = tensor<string, []>("op_45035_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_45036_cast_fp16 = softmax(axis = var_30355, x = aw_3825_cast_fp16)[name = tensor<string, []>("op_45036_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_45037_cast_fp16 = softmax(axis = var_30355, x = aw_3827_cast_fp16)[name = tensor<string, []>("op_45037_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_45038_cast_fp16 = softmax(axis = var_30355, x = aw_3829_cast_fp16)[name = tensor<string, []>("op_45038_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_45039_cast_fp16 = softmax(axis = var_30355, x = aw_3831_cast_fp16)[name = tensor<string, []>("op_45039_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_45040_cast_fp16 = softmax(axis = var_30355, x = aw_3833_cast_fp16)[name = tensor<string, []>("op_45040_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_45041_cast_fp16 = softmax(axis = var_30355, x = aw_3835_cast_fp16)[name = tensor<string, []>("op_45041_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_45042_cast_fp16 = softmax(axis = var_30355, x = aw_3837_cast_fp16)[name = tensor<string, []>("op_45042_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_45043_cast_fp16 = softmax(axis = var_30355, x = aw_3839_cast_fp16)[name = tensor<string, []>("op_45043_cast_fp16")];
            tensor<string, []> var_45045_equation_0 = const()[name = tensor<string, []>("op_45045_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45045_cast_fp16 = einsum(equation = var_45045_equation_0, values = (var_44865_cast_fp16, var_45024_cast_fp16))[name = tensor<string, []>("op_45045_cast_fp16")];
            tensor<string, []> var_45047_equation_0 = const()[name = tensor<string, []>("op_45047_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45047_cast_fp16 = einsum(equation = var_45047_equation_0, values = (var_44869_cast_fp16, var_45025_cast_fp16))[name = tensor<string, []>("op_45047_cast_fp16")];
            tensor<string, []> var_45049_equation_0 = const()[name = tensor<string, []>("op_45049_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45049_cast_fp16 = einsum(equation = var_45049_equation_0, values = (var_44873_cast_fp16, var_45026_cast_fp16))[name = tensor<string, []>("op_45049_cast_fp16")];
            tensor<string, []> var_45051_equation_0 = const()[name = tensor<string, []>("op_45051_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45051_cast_fp16 = einsum(equation = var_45051_equation_0, values = (var_44877_cast_fp16, var_45027_cast_fp16))[name = tensor<string, []>("op_45051_cast_fp16")];
            tensor<string, []> var_45053_equation_0 = const()[name = tensor<string, []>("op_45053_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45053_cast_fp16 = einsum(equation = var_45053_equation_0, values = (var_44881_cast_fp16, var_45028_cast_fp16))[name = tensor<string, []>("op_45053_cast_fp16")];
            tensor<string, []> var_45055_equation_0 = const()[name = tensor<string, []>("op_45055_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45055_cast_fp16 = einsum(equation = var_45055_equation_0, values = (var_44885_cast_fp16, var_45029_cast_fp16))[name = tensor<string, []>("op_45055_cast_fp16")];
            tensor<string, []> var_45057_equation_0 = const()[name = tensor<string, []>("op_45057_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45057_cast_fp16 = einsum(equation = var_45057_equation_0, values = (var_44889_cast_fp16, var_45030_cast_fp16))[name = tensor<string, []>("op_45057_cast_fp16")];
            tensor<string, []> var_45059_equation_0 = const()[name = tensor<string, []>("op_45059_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45059_cast_fp16 = einsum(equation = var_45059_equation_0, values = (var_44893_cast_fp16, var_45031_cast_fp16))[name = tensor<string, []>("op_45059_cast_fp16")];
            tensor<string, []> var_45061_equation_0 = const()[name = tensor<string, []>("op_45061_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45061_cast_fp16 = einsum(equation = var_45061_equation_0, values = (var_44897_cast_fp16, var_45032_cast_fp16))[name = tensor<string, []>("op_45061_cast_fp16")];
            tensor<string, []> var_45063_equation_0 = const()[name = tensor<string, []>("op_45063_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45063_cast_fp16 = einsum(equation = var_45063_equation_0, values = (var_44901_cast_fp16, var_45033_cast_fp16))[name = tensor<string, []>("op_45063_cast_fp16")];
            tensor<string, []> var_45065_equation_0 = const()[name = tensor<string, []>("op_45065_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45065_cast_fp16 = einsum(equation = var_45065_equation_0, values = (var_44905_cast_fp16, var_45034_cast_fp16))[name = tensor<string, []>("op_45065_cast_fp16")];
            tensor<string, []> var_45067_equation_0 = const()[name = tensor<string, []>("op_45067_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45067_cast_fp16 = einsum(equation = var_45067_equation_0, values = (var_44909_cast_fp16, var_45035_cast_fp16))[name = tensor<string, []>("op_45067_cast_fp16")];
            tensor<string, []> var_45069_equation_0 = const()[name = tensor<string, []>("op_45069_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45069_cast_fp16 = einsum(equation = var_45069_equation_0, values = (var_44913_cast_fp16, var_45036_cast_fp16))[name = tensor<string, []>("op_45069_cast_fp16")];
            tensor<string, []> var_45071_equation_0 = const()[name = tensor<string, []>("op_45071_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45071_cast_fp16 = einsum(equation = var_45071_equation_0, values = (var_44917_cast_fp16, var_45037_cast_fp16))[name = tensor<string, []>("op_45071_cast_fp16")];
            tensor<string, []> var_45073_equation_0 = const()[name = tensor<string, []>("op_45073_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45073_cast_fp16 = einsum(equation = var_45073_equation_0, values = (var_44921_cast_fp16, var_45038_cast_fp16))[name = tensor<string, []>("op_45073_cast_fp16")];
            tensor<string, []> var_45075_equation_0 = const()[name = tensor<string, []>("op_45075_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45075_cast_fp16 = einsum(equation = var_45075_equation_0, values = (var_44925_cast_fp16, var_45039_cast_fp16))[name = tensor<string, []>("op_45075_cast_fp16")];
            tensor<string, []> var_45077_equation_0 = const()[name = tensor<string, []>("op_45077_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45077_cast_fp16 = einsum(equation = var_45077_equation_0, values = (var_44929_cast_fp16, var_45040_cast_fp16))[name = tensor<string, []>("op_45077_cast_fp16")];
            tensor<string, []> var_45079_equation_0 = const()[name = tensor<string, []>("op_45079_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45079_cast_fp16 = einsum(equation = var_45079_equation_0, values = (var_44933_cast_fp16, var_45041_cast_fp16))[name = tensor<string, []>("op_45079_cast_fp16")];
            tensor<string, []> var_45081_equation_0 = const()[name = tensor<string, []>("op_45081_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45081_cast_fp16 = einsum(equation = var_45081_equation_0, values = (var_44937_cast_fp16, var_45042_cast_fp16))[name = tensor<string, []>("op_45081_cast_fp16")];
            tensor<string, []> var_45083_equation_0 = const()[name = tensor<string, []>("op_45083_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45083_cast_fp16 = einsum(equation = var_45083_equation_0, values = (var_44941_cast_fp16, var_45043_cast_fp16))[name = tensor<string, []>("op_45083_cast_fp16")];
            tensor<bool, []> input_579_interleave_0 = const()[name = tensor<string, []>("input_579_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_579_cast_fp16 = concat(axis = var_30355, interleave = input_579_interleave_0, values = (var_45045_cast_fp16, var_45047_cast_fp16, var_45049_cast_fp16, var_45051_cast_fp16, var_45053_cast_fp16, var_45055_cast_fp16, var_45057_cast_fp16, var_45059_cast_fp16, var_45061_cast_fp16, var_45063_cast_fp16, var_45065_cast_fp16, var_45067_cast_fp16, var_45069_cast_fp16, var_45071_cast_fp16, var_45073_cast_fp16, var_45075_cast_fp16, var_45077_cast_fp16, var_45079_cast_fp16, var_45081_cast_fp16, var_45083_cast_fp16))[name = tensor<string, []>("input_579_cast_fp16")];
            tensor<string, []> var_45093_pad_type_0 = const()[name = tensor<string, []>("op_45093_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_45093_strides_0 = const()[name = tensor<string, []>("op_45093_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_45093_pad_0 = const()[name = tensor<string, []>("op_45093_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_45093_dilations_0 = const()[name = tensor<string, []>("op_45093_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_45093_groups_0 = const()[name = tensor<string, []>("op_45093_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_5_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(444627200))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(445856064))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_5_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_5_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_5_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(445856256)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_45093_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_5_attn2_to_out_0_bias_to_fp16, dilations = var_45093_dilations_0, groups = var_45093_groups_0, pad = var_45093_pad_0, pad_type = var_45093_pad_type_0, strides = var_45093_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_5_attn2_to_out_0_weight_to_fp16_palettized, x = input_579_cast_fp16)[name = tensor<string, []>("op_45093_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_299_cast_fp16 = add(x = var_45093_cast_fp16, y = inputs_297_cast_fp16)[name = tensor<string, []>("inputs_299_cast_fp16")];
            tensor<int32, [1]> input_581_axes_0 = const()[name = tensor<string, []>("input_581_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> input_581_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_581_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(445858880)))];
            tensor<fp16, [1280]> input_581_beta_0_to_fp16 = const()[name = tensor<string, []>("input_581_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(445861504)))];
            tensor<fp16, []> var_45103_to_fp16 = const()[name = tensor<string, []>("op_45103_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> input_581_cast_fp16 = layer_norm(axes = input_581_axes_0, beta = input_581_beta_0_to_fp16, epsilon = var_45103_to_fp16, gamma = input_581_gamma_0_to_fp16, x = inputs_299_cast_fp16)[name = tensor<string, []>("input_581_cast_fp16")];
            tensor<string, []> var_45123_pad_type_0 = const()[name = tensor<string, []>("op_45123_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_45123_strides_0 = const()[name = tensor<string, []>("op_45123_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_45123_pad_0 = const()[name = tensor<string, []>("op_45123_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_45123_dilations_0 = const()[name = tensor<string, []>("op_45123_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_45123_groups_0 = const()[name = tensor<string, []>("op_45123_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_5_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [9830400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(445864128))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(455694592))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_5_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([10240, 1280, 1, 1])];
            tensor<fp16, [10240]> up_blocks_0_attentions_1_transformer_blocks_5_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_5_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(455694784)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_45123_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_5_ff_net_0_proj_bias_to_fp16, dilations = var_45123_dilations_0, groups = var_45123_groups_0, pad = var_45123_pad_0, pad_type = var_45123_pad_type_0, strides = var_45123_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_5_ff_net_0_proj_weight_to_fp16_palettized, x = input_581_cast_fp16)[name = tensor<string, []>("op_45123_cast_fp16")];
            tensor<int32, [2]> var_45124_split_sizes_0 = const()[name = tensor<string, []>("op_45124_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_45124_axis_0 = const()[name = tensor<string, []>("op_45124_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_45124_cast_fp16_0, tensor<fp16, [2, 5120, 1, 1024]> var_45124_cast_fp16_1 = split(axis = var_45124_axis_0, split_sizes = var_45124_split_sizes_0, x = var_45123_cast_fp16)[name = tensor<string, []>("op_45124_cast_fp16")];
            tensor<string, []> var_45126_mode_0 = const()[name = tensor<string, []>("op_45126_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_45126_cast_fp16 = gelu(mode = var_45126_mode_0, x = var_45124_cast_fp16_1)[name = tensor<string, []>("op_45126_cast_fp16")];
            tensor<fp16, [2, 5120, 1, 1024]> input_583_cast_fp16 = mul(x = var_45124_cast_fp16_0, y = var_45126_cast_fp16)[name = tensor<string, []>("input_583_cast_fp16")];
            tensor<string, []> var_45134_pad_type_0 = const()[name = tensor<string, []>("op_45134_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_45134_strides_0 = const()[name = tensor<string, []>("op_45134_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_45134_pad_0 = const()[name = tensor<string, []>("op_45134_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_45134_dilations_0 = const()[name = tensor<string, []>("op_45134_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_45134_groups_0 = const()[name = tensor<string, []>("op_45134_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_5_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [4915200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(455715328))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(460630592))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_5_ff_net_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 5120, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_5_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_5_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(460630784)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_45134_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_5_ff_net_2_bias_to_fp16, dilations = var_45134_dilations_0, groups = var_45134_groups_0, pad = var_45134_pad_0, pad_type = var_45134_pad_type_0, strides = var_45134_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_5_ff_net_2_weight_to_fp16_palettized, x = input_583_cast_fp16)[name = tensor<string, []>("op_45134_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_301_cast_fp16 = add(x = var_45134_cast_fp16, y = inputs_299_cast_fp16)[name = tensor<string, []>("inputs_301_cast_fp16")];
            tensor<int32, [1]> hidden_states_397_axes_0 = const()[name = tensor<string, []>("hidden_states_397_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_397_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_397_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(460633408)))];
            tensor<fp16, [1280]> hidden_states_397_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_397_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(460636032)))];
            tensor<fp16, []> var_45150_to_fp16 = const()[name = tensor<string, []>("op_45150_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_397_cast_fp16 = layer_norm(axes = hidden_states_397_axes_0, beta = hidden_states_397_beta_0_to_fp16, epsilon = var_45150_to_fp16, gamma = hidden_states_397_gamma_0_to_fp16, x = inputs_301_cast_fp16)[name = tensor<string, []>("hidden_states_397_cast_fp16")];
            tensor<string, []> q_201_pad_type_0 = const()[name = tensor<string, []>("q_201_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_201_strides_0 = const()[name = tensor<string, []>("q_201_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_201_pad_0 = const()[name = tensor<string, []>("q_201_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_201_dilations_0 = const()[name = tensor<string, []>("q_201_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_201_groups_0 = const()[name = tensor<string, []>("q_201_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_6_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(460638656))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(461867520))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_6_attn1_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_201_cast_fp16 = conv(dilations = q_201_dilations_0, groups = q_201_groups_0, pad = q_201_pad_0, pad_type = q_201_pad_type_0, strides = q_201_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_6_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_397_cast_fp16)[name = tensor<string, []>("q_201_cast_fp16")];
            tensor<string, []> k_401_pad_type_0 = const()[name = tensor<string, []>("k_401_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_401_strides_0 = const()[name = tensor<string, []>("k_401_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_401_pad_0 = const()[name = tensor<string, []>("k_401_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_401_dilations_0 = const()[name = tensor<string, []>("k_401_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_401_groups_0 = const()[name = tensor<string, []>("k_401_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_6_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(461867712))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(463096576))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_6_attn1_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> k_401_cast_fp16 = conv(dilations = k_401_dilations_0, groups = k_401_groups_0, pad = k_401_pad_0, pad_type = k_401_pad_type_0, strides = k_401_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_6_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_397_cast_fp16)[name = tensor<string, []>("k_401_cast_fp16")];
            tensor<string, []> v_201_pad_type_0 = const()[name = tensor<string, []>("v_201_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_201_strides_0 = const()[name = tensor<string, []>("v_201_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_201_pad_0 = const()[name = tensor<string, []>("v_201_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_201_dilations_0 = const()[name = tensor<string, []>("v_201_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_201_groups_0 = const()[name = tensor<string, []>("v_201_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_6_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(463096768))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(464325632))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_6_attn1_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> v_201_cast_fp16 = conv(dilations = v_201_dilations_0, groups = v_201_groups_0, pad = v_201_pad_0, pad_type = v_201_pad_type_0, strides = v_201_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_6_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_397_cast_fp16)[name = tensor<string, []>("v_201_cast_fp16")];
            tensor<int32, [4]> var_45183_begin_0 = const()[name = tensor<string, []>("op_45183_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_45183_end_0 = const()[name = tensor<string, []>("op_45183_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_45183_end_mask_0 = const()[name = tensor<string, []>("op_45183_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45183_cast_fp16 = slice_by_index(begin = var_45183_begin_0, end = var_45183_end_0, end_mask = var_45183_end_mask_0, x = q_201_cast_fp16)[name = tensor<string, []>("op_45183_cast_fp16")];
            tensor<int32, [4]> var_45187_begin_0 = const()[name = tensor<string, []>("op_45187_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_45187_end_0 = const()[name = tensor<string, []>("op_45187_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_45187_end_mask_0 = const()[name = tensor<string, []>("op_45187_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45187_cast_fp16 = slice_by_index(begin = var_45187_begin_0, end = var_45187_end_0, end_mask = var_45187_end_mask_0, x = q_201_cast_fp16)[name = tensor<string, []>("op_45187_cast_fp16")];
            tensor<int32, [4]> var_45191_begin_0 = const()[name = tensor<string, []>("op_45191_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_45191_end_0 = const()[name = tensor<string, []>("op_45191_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_45191_end_mask_0 = const()[name = tensor<string, []>("op_45191_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45191_cast_fp16 = slice_by_index(begin = var_45191_begin_0, end = var_45191_end_0, end_mask = var_45191_end_mask_0, x = q_201_cast_fp16)[name = tensor<string, []>("op_45191_cast_fp16")];
            tensor<int32, [4]> var_45195_begin_0 = const()[name = tensor<string, []>("op_45195_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_45195_end_0 = const()[name = tensor<string, []>("op_45195_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_45195_end_mask_0 = const()[name = tensor<string, []>("op_45195_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45195_cast_fp16 = slice_by_index(begin = var_45195_begin_0, end = var_45195_end_0, end_mask = var_45195_end_mask_0, x = q_201_cast_fp16)[name = tensor<string, []>("op_45195_cast_fp16")];
            tensor<int32, [4]> var_45199_begin_0 = const()[name = tensor<string, []>("op_45199_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_45199_end_0 = const()[name = tensor<string, []>("op_45199_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_45199_end_mask_0 = const()[name = tensor<string, []>("op_45199_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45199_cast_fp16 = slice_by_index(begin = var_45199_begin_0, end = var_45199_end_0, end_mask = var_45199_end_mask_0, x = q_201_cast_fp16)[name = tensor<string, []>("op_45199_cast_fp16")];
            tensor<int32, [4]> var_45203_begin_0 = const()[name = tensor<string, []>("op_45203_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_45203_end_0 = const()[name = tensor<string, []>("op_45203_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_45203_end_mask_0 = const()[name = tensor<string, []>("op_45203_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45203_cast_fp16 = slice_by_index(begin = var_45203_begin_0, end = var_45203_end_0, end_mask = var_45203_end_mask_0, x = q_201_cast_fp16)[name = tensor<string, []>("op_45203_cast_fp16")];
            tensor<int32, [4]> var_45207_begin_0 = const()[name = tensor<string, []>("op_45207_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_45207_end_0 = const()[name = tensor<string, []>("op_45207_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_45207_end_mask_0 = const()[name = tensor<string, []>("op_45207_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45207_cast_fp16 = slice_by_index(begin = var_45207_begin_0, end = var_45207_end_0, end_mask = var_45207_end_mask_0, x = q_201_cast_fp16)[name = tensor<string, []>("op_45207_cast_fp16")];
            tensor<int32, [4]> var_45211_begin_0 = const()[name = tensor<string, []>("op_45211_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_45211_end_0 = const()[name = tensor<string, []>("op_45211_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_45211_end_mask_0 = const()[name = tensor<string, []>("op_45211_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45211_cast_fp16 = slice_by_index(begin = var_45211_begin_0, end = var_45211_end_0, end_mask = var_45211_end_mask_0, x = q_201_cast_fp16)[name = tensor<string, []>("op_45211_cast_fp16")];
            tensor<int32, [4]> var_45215_begin_0 = const()[name = tensor<string, []>("op_45215_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_45215_end_0 = const()[name = tensor<string, []>("op_45215_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_45215_end_mask_0 = const()[name = tensor<string, []>("op_45215_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45215_cast_fp16 = slice_by_index(begin = var_45215_begin_0, end = var_45215_end_0, end_mask = var_45215_end_mask_0, x = q_201_cast_fp16)[name = tensor<string, []>("op_45215_cast_fp16")];
            tensor<int32, [4]> var_45219_begin_0 = const()[name = tensor<string, []>("op_45219_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_45219_end_0 = const()[name = tensor<string, []>("op_45219_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_45219_end_mask_0 = const()[name = tensor<string, []>("op_45219_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45219_cast_fp16 = slice_by_index(begin = var_45219_begin_0, end = var_45219_end_0, end_mask = var_45219_end_mask_0, x = q_201_cast_fp16)[name = tensor<string, []>("op_45219_cast_fp16")];
            tensor<int32, [4]> var_45223_begin_0 = const()[name = tensor<string, []>("op_45223_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_45223_end_0 = const()[name = tensor<string, []>("op_45223_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_45223_end_mask_0 = const()[name = tensor<string, []>("op_45223_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45223_cast_fp16 = slice_by_index(begin = var_45223_begin_0, end = var_45223_end_0, end_mask = var_45223_end_mask_0, x = q_201_cast_fp16)[name = tensor<string, []>("op_45223_cast_fp16")];
            tensor<int32, [4]> var_45227_begin_0 = const()[name = tensor<string, []>("op_45227_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_45227_end_0 = const()[name = tensor<string, []>("op_45227_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_45227_end_mask_0 = const()[name = tensor<string, []>("op_45227_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45227_cast_fp16 = slice_by_index(begin = var_45227_begin_0, end = var_45227_end_0, end_mask = var_45227_end_mask_0, x = q_201_cast_fp16)[name = tensor<string, []>("op_45227_cast_fp16")];
            tensor<int32, [4]> var_45231_begin_0 = const()[name = tensor<string, []>("op_45231_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_45231_end_0 = const()[name = tensor<string, []>("op_45231_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_45231_end_mask_0 = const()[name = tensor<string, []>("op_45231_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45231_cast_fp16 = slice_by_index(begin = var_45231_begin_0, end = var_45231_end_0, end_mask = var_45231_end_mask_0, x = q_201_cast_fp16)[name = tensor<string, []>("op_45231_cast_fp16")];
            tensor<int32, [4]> var_45235_begin_0 = const()[name = tensor<string, []>("op_45235_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_45235_end_0 = const()[name = tensor<string, []>("op_45235_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_45235_end_mask_0 = const()[name = tensor<string, []>("op_45235_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45235_cast_fp16 = slice_by_index(begin = var_45235_begin_0, end = var_45235_end_0, end_mask = var_45235_end_mask_0, x = q_201_cast_fp16)[name = tensor<string, []>("op_45235_cast_fp16")];
            tensor<int32, [4]> var_45239_begin_0 = const()[name = tensor<string, []>("op_45239_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_45239_end_0 = const()[name = tensor<string, []>("op_45239_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_45239_end_mask_0 = const()[name = tensor<string, []>("op_45239_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45239_cast_fp16 = slice_by_index(begin = var_45239_begin_0, end = var_45239_end_0, end_mask = var_45239_end_mask_0, x = q_201_cast_fp16)[name = tensor<string, []>("op_45239_cast_fp16")];
            tensor<int32, [4]> var_45243_begin_0 = const()[name = tensor<string, []>("op_45243_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_45243_end_0 = const()[name = tensor<string, []>("op_45243_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_45243_end_mask_0 = const()[name = tensor<string, []>("op_45243_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45243_cast_fp16 = slice_by_index(begin = var_45243_begin_0, end = var_45243_end_0, end_mask = var_45243_end_mask_0, x = q_201_cast_fp16)[name = tensor<string, []>("op_45243_cast_fp16")];
            tensor<int32, [4]> var_45247_begin_0 = const()[name = tensor<string, []>("op_45247_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_45247_end_0 = const()[name = tensor<string, []>("op_45247_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_45247_end_mask_0 = const()[name = tensor<string, []>("op_45247_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45247_cast_fp16 = slice_by_index(begin = var_45247_begin_0, end = var_45247_end_0, end_mask = var_45247_end_mask_0, x = q_201_cast_fp16)[name = tensor<string, []>("op_45247_cast_fp16")];
            tensor<int32, [4]> var_45251_begin_0 = const()[name = tensor<string, []>("op_45251_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_45251_end_0 = const()[name = tensor<string, []>("op_45251_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_45251_end_mask_0 = const()[name = tensor<string, []>("op_45251_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45251_cast_fp16 = slice_by_index(begin = var_45251_begin_0, end = var_45251_end_0, end_mask = var_45251_end_mask_0, x = q_201_cast_fp16)[name = tensor<string, []>("op_45251_cast_fp16")];
            tensor<int32, [4]> var_45255_begin_0 = const()[name = tensor<string, []>("op_45255_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_45255_end_0 = const()[name = tensor<string, []>("op_45255_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_45255_end_mask_0 = const()[name = tensor<string, []>("op_45255_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45255_cast_fp16 = slice_by_index(begin = var_45255_begin_0, end = var_45255_end_0, end_mask = var_45255_end_mask_0, x = q_201_cast_fp16)[name = tensor<string, []>("op_45255_cast_fp16")];
            tensor<int32, [4]> var_45259_begin_0 = const()[name = tensor<string, []>("op_45259_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_45259_end_0 = const()[name = tensor<string, []>("op_45259_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_45259_end_mask_0 = const()[name = tensor<string, []>("op_45259_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45259_cast_fp16 = slice_by_index(begin = var_45259_begin_0, end = var_45259_end_0, end_mask = var_45259_end_mask_0, x = q_201_cast_fp16)[name = tensor<string, []>("op_45259_cast_fp16")];
            tensor<int32, [4]> k_403_perm_0 = const()[name = tensor<string, []>("k_403_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_45266_begin_0 = const()[name = tensor<string, []>("op_45266_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_45266_end_0 = const()[name = tensor<string, []>("op_45266_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_45266_end_mask_0 = const()[name = tensor<string, []>("op_45266_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> k_403_cast_fp16 = transpose(perm = k_403_perm_0, x = k_401_cast_fp16)[name = tensor<string, []>("transpose_39")];
            tensor<fp16, [2, 1024, 1, 64]> var_45266_cast_fp16 = slice_by_index(begin = var_45266_begin_0, end = var_45266_end_0, end_mask = var_45266_end_mask_0, x = k_403_cast_fp16)[name = tensor<string, []>("op_45266_cast_fp16")];
            tensor<int32, [4]> var_45270_begin_0 = const()[name = tensor<string, []>("op_45270_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_45270_end_0 = const()[name = tensor<string, []>("op_45270_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_45270_end_mask_0 = const()[name = tensor<string, []>("op_45270_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_45270_cast_fp16 = slice_by_index(begin = var_45270_begin_0, end = var_45270_end_0, end_mask = var_45270_end_mask_0, x = k_403_cast_fp16)[name = tensor<string, []>("op_45270_cast_fp16")];
            tensor<int32, [4]> var_45274_begin_0 = const()[name = tensor<string, []>("op_45274_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_45274_end_0 = const()[name = tensor<string, []>("op_45274_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_45274_end_mask_0 = const()[name = tensor<string, []>("op_45274_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_45274_cast_fp16 = slice_by_index(begin = var_45274_begin_0, end = var_45274_end_0, end_mask = var_45274_end_mask_0, x = k_403_cast_fp16)[name = tensor<string, []>("op_45274_cast_fp16")];
            tensor<int32, [4]> var_45278_begin_0 = const()[name = tensor<string, []>("op_45278_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_45278_end_0 = const()[name = tensor<string, []>("op_45278_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_45278_end_mask_0 = const()[name = tensor<string, []>("op_45278_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_45278_cast_fp16 = slice_by_index(begin = var_45278_begin_0, end = var_45278_end_0, end_mask = var_45278_end_mask_0, x = k_403_cast_fp16)[name = tensor<string, []>("op_45278_cast_fp16")];
            tensor<int32, [4]> var_45282_begin_0 = const()[name = tensor<string, []>("op_45282_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_45282_end_0 = const()[name = tensor<string, []>("op_45282_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_45282_end_mask_0 = const()[name = tensor<string, []>("op_45282_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_45282_cast_fp16 = slice_by_index(begin = var_45282_begin_0, end = var_45282_end_0, end_mask = var_45282_end_mask_0, x = k_403_cast_fp16)[name = tensor<string, []>("op_45282_cast_fp16")];
            tensor<int32, [4]> var_45286_begin_0 = const()[name = tensor<string, []>("op_45286_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_45286_end_0 = const()[name = tensor<string, []>("op_45286_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_45286_end_mask_0 = const()[name = tensor<string, []>("op_45286_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_45286_cast_fp16 = slice_by_index(begin = var_45286_begin_0, end = var_45286_end_0, end_mask = var_45286_end_mask_0, x = k_403_cast_fp16)[name = tensor<string, []>("op_45286_cast_fp16")];
            tensor<int32, [4]> var_45290_begin_0 = const()[name = tensor<string, []>("op_45290_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_45290_end_0 = const()[name = tensor<string, []>("op_45290_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_45290_end_mask_0 = const()[name = tensor<string, []>("op_45290_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_45290_cast_fp16 = slice_by_index(begin = var_45290_begin_0, end = var_45290_end_0, end_mask = var_45290_end_mask_0, x = k_403_cast_fp16)[name = tensor<string, []>("op_45290_cast_fp16")];
            tensor<int32, [4]> var_45294_begin_0 = const()[name = tensor<string, []>("op_45294_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_45294_end_0 = const()[name = tensor<string, []>("op_45294_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_45294_end_mask_0 = const()[name = tensor<string, []>("op_45294_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_45294_cast_fp16 = slice_by_index(begin = var_45294_begin_0, end = var_45294_end_0, end_mask = var_45294_end_mask_0, x = k_403_cast_fp16)[name = tensor<string, []>("op_45294_cast_fp16")];
            tensor<int32, [4]> var_45298_begin_0 = const()[name = tensor<string, []>("op_45298_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_45298_end_0 = const()[name = tensor<string, []>("op_45298_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_45298_end_mask_0 = const()[name = tensor<string, []>("op_45298_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_45298_cast_fp16 = slice_by_index(begin = var_45298_begin_0, end = var_45298_end_0, end_mask = var_45298_end_mask_0, x = k_403_cast_fp16)[name = tensor<string, []>("op_45298_cast_fp16")];
            tensor<int32, [4]> var_45302_begin_0 = const()[name = tensor<string, []>("op_45302_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_45302_end_0 = const()[name = tensor<string, []>("op_45302_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_45302_end_mask_0 = const()[name = tensor<string, []>("op_45302_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_45302_cast_fp16 = slice_by_index(begin = var_45302_begin_0, end = var_45302_end_0, end_mask = var_45302_end_mask_0, x = k_403_cast_fp16)[name = tensor<string, []>("op_45302_cast_fp16")];
            tensor<int32, [4]> var_45306_begin_0 = const()[name = tensor<string, []>("op_45306_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_45306_end_0 = const()[name = tensor<string, []>("op_45306_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_45306_end_mask_0 = const()[name = tensor<string, []>("op_45306_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_45306_cast_fp16 = slice_by_index(begin = var_45306_begin_0, end = var_45306_end_0, end_mask = var_45306_end_mask_0, x = k_403_cast_fp16)[name = tensor<string, []>("op_45306_cast_fp16")];
            tensor<int32, [4]> var_45310_begin_0 = const()[name = tensor<string, []>("op_45310_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_45310_end_0 = const()[name = tensor<string, []>("op_45310_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_45310_end_mask_0 = const()[name = tensor<string, []>("op_45310_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_45310_cast_fp16 = slice_by_index(begin = var_45310_begin_0, end = var_45310_end_0, end_mask = var_45310_end_mask_0, x = k_403_cast_fp16)[name = tensor<string, []>("op_45310_cast_fp16")];
            tensor<int32, [4]> var_45314_begin_0 = const()[name = tensor<string, []>("op_45314_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_45314_end_0 = const()[name = tensor<string, []>("op_45314_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_45314_end_mask_0 = const()[name = tensor<string, []>("op_45314_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_45314_cast_fp16 = slice_by_index(begin = var_45314_begin_0, end = var_45314_end_0, end_mask = var_45314_end_mask_0, x = k_403_cast_fp16)[name = tensor<string, []>("op_45314_cast_fp16")];
            tensor<int32, [4]> var_45318_begin_0 = const()[name = tensor<string, []>("op_45318_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_45318_end_0 = const()[name = tensor<string, []>("op_45318_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_45318_end_mask_0 = const()[name = tensor<string, []>("op_45318_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_45318_cast_fp16 = slice_by_index(begin = var_45318_begin_0, end = var_45318_end_0, end_mask = var_45318_end_mask_0, x = k_403_cast_fp16)[name = tensor<string, []>("op_45318_cast_fp16")];
            tensor<int32, [4]> var_45322_begin_0 = const()[name = tensor<string, []>("op_45322_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_45322_end_0 = const()[name = tensor<string, []>("op_45322_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_45322_end_mask_0 = const()[name = tensor<string, []>("op_45322_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_45322_cast_fp16 = slice_by_index(begin = var_45322_begin_0, end = var_45322_end_0, end_mask = var_45322_end_mask_0, x = k_403_cast_fp16)[name = tensor<string, []>("op_45322_cast_fp16")];
            tensor<int32, [4]> var_45326_begin_0 = const()[name = tensor<string, []>("op_45326_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_45326_end_0 = const()[name = tensor<string, []>("op_45326_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_45326_end_mask_0 = const()[name = tensor<string, []>("op_45326_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_45326_cast_fp16 = slice_by_index(begin = var_45326_begin_0, end = var_45326_end_0, end_mask = var_45326_end_mask_0, x = k_403_cast_fp16)[name = tensor<string, []>("op_45326_cast_fp16")];
            tensor<int32, [4]> var_45330_begin_0 = const()[name = tensor<string, []>("op_45330_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_45330_end_0 = const()[name = tensor<string, []>("op_45330_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_45330_end_mask_0 = const()[name = tensor<string, []>("op_45330_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_45330_cast_fp16 = slice_by_index(begin = var_45330_begin_0, end = var_45330_end_0, end_mask = var_45330_end_mask_0, x = k_403_cast_fp16)[name = tensor<string, []>("op_45330_cast_fp16")];
            tensor<int32, [4]> var_45334_begin_0 = const()[name = tensor<string, []>("op_45334_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_45334_end_0 = const()[name = tensor<string, []>("op_45334_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_45334_end_mask_0 = const()[name = tensor<string, []>("op_45334_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_45334_cast_fp16 = slice_by_index(begin = var_45334_begin_0, end = var_45334_end_0, end_mask = var_45334_end_mask_0, x = k_403_cast_fp16)[name = tensor<string, []>("op_45334_cast_fp16")];
            tensor<int32, [4]> var_45338_begin_0 = const()[name = tensor<string, []>("op_45338_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_45338_end_0 = const()[name = tensor<string, []>("op_45338_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_45338_end_mask_0 = const()[name = tensor<string, []>("op_45338_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_45338_cast_fp16 = slice_by_index(begin = var_45338_begin_0, end = var_45338_end_0, end_mask = var_45338_end_mask_0, x = k_403_cast_fp16)[name = tensor<string, []>("op_45338_cast_fp16")];
            tensor<int32, [4]> var_45342_begin_0 = const()[name = tensor<string, []>("op_45342_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_45342_end_0 = const()[name = tensor<string, []>("op_45342_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_45342_end_mask_0 = const()[name = tensor<string, []>("op_45342_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_45342_cast_fp16 = slice_by_index(begin = var_45342_begin_0, end = var_45342_end_0, end_mask = var_45342_end_mask_0, x = k_403_cast_fp16)[name = tensor<string, []>("op_45342_cast_fp16")];
            tensor<int32, [4]> var_45344_begin_0 = const()[name = tensor<string, []>("op_45344_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_45344_end_0 = const()[name = tensor<string, []>("op_45344_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_45344_end_mask_0 = const()[name = tensor<string, []>("op_45344_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45344_cast_fp16 = slice_by_index(begin = var_45344_begin_0, end = var_45344_end_0, end_mask = var_45344_end_mask_0, x = v_201_cast_fp16)[name = tensor<string, []>("op_45344_cast_fp16")];
            tensor<int32, [4]> var_45348_begin_0 = const()[name = tensor<string, []>("op_45348_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_45348_end_0 = const()[name = tensor<string, []>("op_45348_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_45348_end_mask_0 = const()[name = tensor<string, []>("op_45348_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45348_cast_fp16 = slice_by_index(begin = var_45348_begin_0, end = var_45348_end_0, end_mask = var_45348_end_mask_0, x = v_201_cast_fp16)[name = tensor<string, []>("op_45348_cast_fp16")];
            tensor<int32, [4]> var_45352_begin_0 = const()[name = tensor<string, []>("op_45352_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_45352_end_0 = const()[name = tensor<string, []>("op_45352_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_45352_end_mask_0 = const()[name = tensor<string, []>("op_45352_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45352_cast_fp16 = slice_by_index(begin = var_45352_begin_0, end = var_45352_end_0, end_mask = var_45352_end_mask_0, x = v_201_cast_fp16)[name = tensor<string, []>("op_45352_cast_fp16")];
            tensor<int32, [4]> var_45356_begin_0 = const()[name = tensor<string, []>("op_45356_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_45356_end_0 = const()[name = tensor<string, []>("op_45356_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_45356_end_mask_0 = const()[name = tensor<string, []>("op_45356_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45356_cast_fp16 = slice_by_index(begin = var_45356_begin_0, end = var_45356_end_0, end_mask = var_45356_end_mask_0, x = v_201_cast_fp16)[name = tensor<string, []>("op_45356_cast_fp16")];
            tensor<int32, [4]> var_45360_begin_0 = const()[name = tensor<string, []>("op_45360_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_45360_end_0 = const()[name = tensor<string, []>("op_45360_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_45360_end_mask_0 = const()[name = tensor<string, []>("op_45360_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45360_cast_fp16 = slice_by_index(begin = var_45360_begin_0, end = var_45360_end_0, end_mask = var_45360_end_mask_0, x = v_201_cast_fp16)[name = tensor<string, []>("op_45360_cast_fp16")];
            tensor<int32, [4]> var_45364_begin_0 = const()[name = tensor<string, []>("op_45364_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_45364_end_0 = const()[name = tensor<string, []>("op_45364_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_45364_end_mask_0 = const()[name = tensor<string, []>("op_45364_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45364_cast_fp16 = slice_by_index(begin = var_45364_begin_0, end = var_45364_end_0, end_mask = var_45364_end_mask_0, x = v_201_cast_fp16)[name = tensor<string, []>("op_45364_cast_fp16")];
            tensor<int32, [4]> var_45368_begin_0 = const()[name = tensor<string, []>("op_45368_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_45368_end_0 = const()[name = tensor<string, []>("op_45368_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_45368_end_mask_0 = const()[name = tensor<string, []>("op_45368_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45368_cast_fp16 = slice_by_index(begin = var_45368_begin_0, end = var_45368_end_0, end_mask = var_45368_end_mask_0, x = v_201_cast_fp16)[name = tensor<string, []>("op_45368_cast_fp16")];
            tensor<int32, [4]> var_45372_begin_0 = const()[name = tensor<string, []>("op_45372_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_45372_end_0 = const()[name = tensor<string, []>("op_45372_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_45372_end_mask_0 = const()[name = tensor<string, []>("op_45372_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45372_cast_fp16 = slice_by_index(begin = var_45372_begin_0, end = var_45372_end_0, end_mask = var_45372_end_mask_0, x = v_201_cast_fp16)[name = tensor<string, []>("op_45372_cast_fp16")];
            tensor<int32, [4]> var_45376_begin_0 = const()[name = tensor<string, []>("op_45376_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_45376_end_0 = const()[name = tensor<string, []>("op_45376_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_45376_end_mask_0 = const()[name = tensor<string, []>("op_45376_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45376_cast_fp16 = slice_by_index(begin = var_45376_begin_0, end = var_45376_end_0, end_mask = var_45376_end_mask_0, x = v_201_cast_fp16)[name = tensor<string, []>("op_45376_cast_fp16")];
            tensor<int32, [4]> var_45380_begin_0 = const()[name = tensor<string, []>("op_45380_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_45380_end_0 = const()[name = tensor<string, []>("op_45380_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_45380_end_mask_0 = const()[name = tensor<string, []>("op_45380_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45380_cast_fp16 = slice_by_index(begin = var_45380_begin_0, end = var_45380_end_0, end_mask = var_45380_end_mask_0, x = v_201_cast_fp16)[name = tensor<string, []>("op_45380_cast_fp16")];
            tensor<int32, [4]> var_45384_begin_0 = const()[name = tensor<string, []>("op_45384_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_45384_end_0 = const()[name = tensor<string, []>("op_45384_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_45384_end_mask_0 = const()[name = tensor<string, []>("op_45384_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45384_cast_fp16 = slice_by_index(begin = var_45384_begin_0, end = var_45384_end_0, end_mask = var_45384_end_mask_0, x = v_201_cast_fp16)[name = tensor<string, []>("op_45384_cast_fp16")];
            tensor<int32, [4]> var_45388_begin_0 = const()[name = tensor<string, []>("op_45388_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_45388_end_0 = const()[name = tensor<string, []>("op_45388_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_45388_end_mask_0 = const()[name = tensor<string, []>("op_45388_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45388_cast_fp16 = slice_by_index(begin = var_45388_begin_0, end = var_45388_end_0, end_mask = var_45388_end_mask_0, x = v_201_cast_fp16)[name = tensor<string, []>("op_45388_cast_fp16")];
            tensor<int32, [4]> var_45392_begin_0 = const()[name = tensor<string, []>("op_45392_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_45392_end_0 = const()[name = tensor<string, []>("op_45392_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_45392_end_mask_0 = const()[name = tensor<string, []>("op_45392_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45392_cast_fp16 = slice_by_index(begin = var_45392_begin_0, end = var_45392_end_0, end_mask = var_45392_end_mask_0, x = v_201_cast_fp16)[name = tensor<string, []>("op_45392_cast_fp16")];
            tensor<int32, [4]> var_45396_begin_0 = const()[name = tensor<string, []>("op_45396_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_45396_end_0 = const()[name = tensor<string, []>("op_45396_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_45396_end_mask_0 = const()[name = tensor<string, []>("op_45396_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45396_cast_fp16 = slice_by_index(begin = var_45396_begin_0, end = var_45396_end_0, end_mask = var_45396_end_mask_0, x = v_201_cast_fp16)[name = tensor<string, []>("op_45396_cast_fp16")];
            tensor<int32, [4]> var_45400_begin_0 = const()[name = tensor<string, []>("op_45400_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_45400_end_0 = const()[name = tensor<string, []>("op_45400_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_45400_end_mask_0 = const()[name = tensor<string, []>("op_45400_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45400_cast_fp16 = slice_by_index(begin = var_45400_begin_0, end = var_45400_end_0, end_mask = var_45400_end_mask_0, x = v_201_cast_fp16)[name = tensor<string, []>("op_45400_cast_fp16")];
            tensor<int32, [4]> var_45404_begin_0 = const()[name = tensor<string, []>("op_45404_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_45404_end_0 = const()[name = tensor<string, []>("op_45404_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_45404_end_mask_0 = const()[name = tensor<string, []>("op_45404_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45404_cast_fp16 = slice_by_index(begin = var_45404_begin_0, end = var_45404_end_0, end_mask = var_45404_end_mask_0, x = v_201_cast_fp16)[name = tensor<string, []>("op_45404_cast_fp16")];
            tensor<int32, [4]> var_45408_begin_0 = const()[name = tensor<string, []>("op_45408_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_45408_end_0 = const()[name = tensor<string, []>("op_45408_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_45408_end_mask_0 = const()[name = tensor<string, []>("op_45408_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45408_cast_fp16 = slice_by_index(begin = var_45408_begin_0, end = var_45408_end_0, end_mask = var_45408_end_mask_0, x = v_201_cast_fp16)[name = tensor<string, []>("op_45408_cast_fp16")];
            tensor<int32, [4]> var_45412_begin_0 = const()[name = tensor<string, []>("op_45412_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_45412_end_0 = const()[name = tensor<string, []>("op_45412_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_45412_end_mask_0 = const()[name = tensor<string, []>("op_45412_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45412_cast_fp16 = slice_by_index(begin = var_45412_begin_0, end = var_45412_end_0, end_mask = var_45412_end_mask_0, x = v_201_cast_fp16)[name = tensor<string, []>("op_45412_cast_fp16")];
            tensor<int32, [4]> var_45416_begin_0 = const()[name = tensor<string, []>("op_45416_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_45416_end_0 = const()[name = tensor<string, []>("op_45416_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_45416_end_mask_0 = const()[name = tensor<string, []>("op_45416_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45416_cast_fp16 = slice_by_index(begin = var_45416_begin_0, end = var_45416_end_0, end_mask = var_45416_end_mask_0, x = v_201_cast_fp16)[name = tensor<string, []>("op_45416_cast_fp16")];
            tensor<int32, [4]> var_45420_begin_0 = const()[name = tensor<string, []>("op_45420_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_45420_end_0 = const()[name = tensor<string, []>("op_45420_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_45420_end_mask_0 = const()[name = tensor<string, []>("op_45420_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45420_cast_fp16 = slice_by_index(begin = var_45420_begin_0, end = var_45420_end_0, end_mask = var_45420_end_mask_0, x = v_201_cast_fp16)[name = tensor<string, []>("op_45420_cast_fp16")];
            tensor<string, []> var_45424_equation_0 = const()[name = tensor<string, []>("op_45424_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45424_cast_fp16 = einsum(equation = var_45424_equation_0, values = (var_45266_cast_fp16, var_45183_cast_fp16))[name = tensor<string, []>("op_45424_cast_fp16")];
            tensor<fp16, []> var_45425_to_fp16 = const()[name = tensor<string, []>("op_45425_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3841_cast_fp16 = mul(x = var_45424_cast_fp16, y = var_45425_to_fp16)[name = tensor<string, []>("aw_3841_cast_fp16")];
            tensor<string, []> var_45428_equation_0 = const()[name = tensor<string, []>("op_45428_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45428_cast_fp16 = einsum(equation = var_45428_equation_0, values = (var_45270_cast_fp16, var_45187_cast_fp16))[name = tensor<string, []>("op_45428_cast_fp16")];
            tensor<fp16, []> var_45429_to_fp16 = const()[name = tensor<string, []>("op_45429_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3843_cast_fp16 = mul(x = var_45428_cast_fp16, y = var_45429_to_fp16)[name = tensor<string, []>("aw_3843_cast_fp16")];
            tensor<string, []> var_45432_equation_0 = const()[name = tensor<string, []>("op_45432_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45432_cast_fp16 = einsum(equation = var_45432_equation_0, values = (var_45274_cast_fp16, var_45191_cast_fp16))[name = tensor<string, []>("op_45432_cast_fp16")];
            tensor<fp16, []> var_45433_to_fp16 = const()[name = tensor<string, []>("op_45433_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3845_cast_fp16 = mul(x = var_45432_cast_fp16, y = var_45433_to_fp16)[name = tensor<string, []>("aw_3845_cast_fp16")];
            tensor<string, []> var_45436_equation_0 = const()[name = tensor<string, []>("op_45436_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45436_cast_fp16 = einsum(equation = var_45436_equation_0, values = (var_45278_cast_fp16, var_45195_cast_fp16))[name = tensor<string, []>("op_45436_cast_fp16")];
            tensor<fp16, []> var_45437_to_fp16 = const()[name = tensor<string, []>("op_45437_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3847_cast_fp16 = mul(x = var_45436_cast_fp16, y = var_45437_to_fp16)[name = tensor<string, []>("aw_3847_cast_fp16")];
            tensor<string, []> var_45440_equation_0 = const()[name = tensor<string, []>("op_45440_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45440_cast_fp16 = einsum(equation = var_45440_equation_0, values = (var_45282_cast_fp16, var_45199_cast_fp16))[name = tensor<string, []>("op_45440_cast_fp16")];
            tensor<fp16, []> var_45441_to_fp16 = const()[name = tensor<string, []>("op_45441_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3849_cast_fp16 = mul(x = var_45440_cast_fp16, y = var_45441_to_fp16)[name = tensor<string, []>("aw_3849_cast_fp16")];
            tensor<string, []> var_45444_equation_0 = const()[name = tensor<string, []>("op_45444_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45444_cast_fp16 = einsum(equation = var_45444_equation_0, values = (var_45286_cast_fp16, var_45203_cast_fp16))[name = tensor<string, []>("op_45444_cast_fp16")];
            tensor<fp16, []> var_45445_to_fp16 = const()[name = tensor<string, []>("op_45445_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3851_cast_fp16 = mul(x = var_45444_cast_fp16, y = var_45445_to_fp16)[name = tensor<string, []>("aw_3851_cast_fp16")];
            tensor<string, []> var_45448_equation_0 = const()[name = tensor<string, []>("op_45448_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45448_cast_fp16 = einsum(equation = var_45448_equation_0, values = (var_45290_cast_fp16, var_45207_cast_fp16))[name = tensor<string, []>("op_45448_cast_fp16")];
            tensor<fp16, []> var_45449_to_fp16 = const()[name = tensor<string, []>("op_45449_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3853_cast_fp16 = mul(x = var_45448_cast_fp16, y = var_45449_to_fp16)[name = tensor<string, []>("aw_3853_cast_fp16")];
            tensor<string, []> var_45452_equation_0 = const()[name = tensor<string, []>("op_45452_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45452_cast_fp16 = einsum(equation = var_45452_equation_0, values = (var_45294_cast_fp16, var_45211_cast_fp16))[name = tensor<string, []>("op_45452_cast_fp16")];
            tensor<fp16, []> var_45453_to_fp16 = const()[name = tensor<string, []>("op_45453_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3855_cast_fp16 = mul(x = var_45452_cast_fp16, y = var_45453_to_fp16)[name = tensor<string, []>("aw_3855_cast_fp16")];
            tensor<string, []> var_45456_equation_0 = const()[name = tensor<string, []>("op_45456_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45456_cast_fp16 = einsum(equation = var_45456_equation_0, values = (var_45298_cast_fp16, var_45215_cast_fp16))[name = tensor<string, []>("op_45456_cast_fp16")];
            tensor<fp16, []> var_45457_to_fp16 = const()[name = tensor<string, []>("op_45457_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3857_cast_fp16 = mul(x = var_45456_cast_fp16, y = var_45457_to_fp16)[name = tensor<string, []>("aw_3857_cast_fp16")];
            tensor<string, []> var_45460_equation_0 = const()[name = tensor<string, []>("op_45460_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45460_cast_fp16 = einsum(equation = var_45460_equation_0, values = (var_45302_cast_fp16, var_45219_cast_fp16))[name = tensor<string, []>("op_45460_cast_fp16")];
            tensor<fp16, []> var_45461_to_fp16 = const()[name = tensor<string, []>("op_45461_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3859_cast_fp16 = mul(x = var_45460_cast_fp16, y = var_45461_to_fp16)[name = tensor<string, []>("aw_3859_cast_fp16")];
            tensor<string, []> var_45464_equation_0 = const()[name = tensor<string, []>("op_45464_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45464_cast_fp16 = einsum(equation = var_45464_equation_0, values = (var_45306_cast_fp16, var_45223_cast_fp16))[name = tensor<string, []>("op_45464_cast_fp16")];
            tensor<fp16, []> var_45465_to_fp16 = const()[name = tensor<string, []>("op_45465_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3861_cast_fp16 = mul(x = var_45464_cast_fp16, y = var_45465_to_fp16)[name = tensor<string, []>("aw_3861_cast_fp16")];
            tensor<string, []> var_45468_equation_0 = const()[name = tensor<string, []>("op_45468_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45468_cast_fp16 = einsum(equation = var_45468_equation_0, values = (var_45310_cast_fp16, var_45227_cast_fp16))[name = tensor<string, []>("op_45468_cast_fp16")];
            tensor<fp16, []> var_45469_to_fp16 = const()[name = tensor<string, []>("op_45469_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3863_cast_fp16 = mul(x = var_45468_cast_fp16, y = var_45469_to_fp16)[name = tensor<string, []>("aw_3863_cast_fp16")];
            tensor<string, []> var_45472_equation_0 = const()[name = tensor<string, []>("op_45472_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45472_cast_fp16 = einsum(equation = var_45472_equation_0, values = (var_45314_cast_fp16, var_45231_cast_fp16))[name = tensor<string, []>("op_45472_cast_fp16")];
            tensor<fp16, []> var_45473_to_fp16 = const()[name = tensor<string, []>("op_45473_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3865_cast_fp16 = mul(x = var_45472_cast_fp16, y = var_45473_to_fp16)[name = tensor<string, []>("aw_3865_cast_fp16")];
            tensor<string, []> var_45476_equation_0 = const()[name = tensor<string, []>("op_45476_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45476_cast_fp16 = einsum(equation = var_45476_equation_0, values = (var_45318_cast_fp16, var_45235_cast_fp16))[name = tensor<string, []>("op_45476_cast_fp16")];
            tensor<fp16, []> var_45477_to_fp16 = const()[name = tensor<string, []>("op_45477_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3867_cast_fp16 = mul(x = var_45476_cast_fp16, y = var_45477_to_fp16)[name = tensor<string, []>("aw_3867_cast_fp16")];
            tensor<string, []> var_45480_equation_0 = const()[name = tensor<string, []>("op_45480_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45480_cast_fp16 = einsum(equation = var_45480_equation_0, values = (var_45322_cast_fp16, var_45239_cast_fp16))[name = tensor<string, []>("op_45480_cast_fp16")];
            tensor<fp16, []> var_45481_to_fp16 = const()[name = tensor<string, []>("op_45481_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3869_cast_fp16 = mul(x = var_45480_cast_fp16, y = var_45481_to_fp16)[name = tensor<string, []>("aw_3869_cast_fp16")];
            tensor<string, []> var_45484_equation_0 = const()[name = tensor<string, []>("op_45484_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45484_cast_fp16 = einsum(equation = var_45484_equation_0, values = (var_45326_cast_fp16, var_45243_cast_fp16))[name = tensor<string, []>("op_45484_cast_fp16")];
            tensor<fp16, []> var_45485_to_fp16 = const()[name = tensor<string, []>("op_45485_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3871_cast_fp16 = mul(x = var_45484_cast_fp16, y = var_45485_to_fp16)[name = tensor<string, []>("aw_3871_cast_fp16")];
            tensor<string, []> var_45488_equation_0 = const()[name = tensor<string, []>("op_45488_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45488_cast_fp16 = einsum(equation = var_45488_equation_0, values = (var_45330_cast_fp16, var_45247_cast_fp16))[name = tensor<string, []>("op_45488_cast_fp16")];
            tensor<fp16, []> var_45489_to_fp16 = const()[name = tensor<string, []>("op_45489_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3873_cast_fp16 = mul(x = var_45488_cast_fp16, y = var_45489_to_fp16)[name = tensor<string, []>("aw_3873_cast_fp16")];
            tensor<string, []> var_45492_equation_0 = const()[name = tensor<string, []>("op_45492_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45492_cast_fp16 = einsum(equation = var_45492_equation_0, values = (var_45334_cast_fp16, var_45251_cast_fp16))[name = tensor<string, []>("op_45492_cast_fp16")];
            tensor<fp16, []> var_45493_to_fp16 = const()[name = tensor<string, []>("op_45493_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3875_cast_fp16 = mul(x = var_45492_cast_fp16, y = var_45493_to_fp16)[name = tensor<string, []>("aw_3875_cast_fp16")];
            tensor<string, []> var_45496_equation_0 = const()[name = tensor<string, []>("op_45496_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45496_cast_fp16 = einsum(equation = var_45496_equation_0, values = (var_45338_cast_fp16, var_45255_cast_fp16))[name = tensor<string, []>("op_45496_cast_fp16")];
            tensor<fp16, []> var_45497_to_fp16 = const()[name = tensor<string, []>("op_45497_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3877_cast_fp16 = mul(x = var_45496_cast_fp16, y = var_45497_to_fp16)[name = tensor<string, []>("aw_3877_cast_fp16")];
            tensor<string, []> var_45500_equation_0 = const()[name = tensor<string, []>("op_45500_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45500_cast_fp16 = einsum(equation = var_45500_equation_0, values = (var_45342_cast_fp16, var_45259_cast_fp16))[name = tensor<string, []>("op_45500_cast_fp16")];
            tensor<fp16, []> var_45501_to_fp16 = const()[name = tensor<string, []>("op_45501_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3879_cast_fp16 = mul(x = var_45500_cast_fp16, y = var_45501_to_fp16)[name = tensor<string, []>("aw_3879_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45503_cast_fp16 = softmax(axis = var_30355, x = aw_3841_cast_fp16)[name = tensor<string, []>("op_45503_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45504_cast_fp16 = softmax(axis = var_30355, x = aw_3843_cast_fp16)[name = tensor<string, []>("op_45504_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45505_cast_fp16 = softmax(axis = var_30355, x = aw_3845_cast_fp16)[name = tensor<string, []>("op_45505_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45506_cast_fp16 = softmax(axis = var_30355, x = aw_3847_cast_fp16)[name = tensor<string, []>("op_45506_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45507_cast_fp16 = softmax(axis = var_30355, x = aw_3849_cast_fp16)[name = tensor<string, []>("op_45507_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45508_cast_fp16 = softmax(axis = var_30355, x = aw_3851_cast_fp16)[name = tensor<string, []>("op_45508_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45509_cast_fp16 = softmax(axis = var_30355, x = aw_3853_cast_fp16)[name = tensor<string, []>("op_45509_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45510_cast_fp16 = softmax(axis = var_30355, x = aw_3855_cast_fp16)[name = tensor<string, []>("op_45510_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45511_cast_fp16 = softmax(axis = var_30355, x = aw_3857_cast_fp16)[name = tensor<string, []>("op_45511_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45512_cast_fp16 = softmax(axis = var_30355, x = aw_3859_cast_fp16)[name = tensor<string, []>("op_45512_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45513_cast_fp16 = softmax(axis = var_30355, x = aw_3861_cast_fp16)[name = tensor<string, []>("op_45513_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45514_cast_fp16 = softmax(axis = var_30355, x = aw_3863_cast_fp16)[name = tensor<string, []>("op_45514_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45515_cast_fp16 = softmax(axis = var_30355, x = aw_3865_cast_fp16)[name = tensor<string, []>("op_45515_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45516_cast_fp16 = softmax(axis = var_30355, x = aw_3867_cast_fp16)[name = tensor<string, []>("op_45516_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45517_cast_fp16 = softmax(axis = var_30355, x = aw_3869_cast_fp16)[name = tensor<string, []>("op_45517_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45518_cast_fp16 = softmax(axis = var_30355, x = aw_3871_cast_fp16)[name = tensor<string, []>("op_45518_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45519_cast_fp16 = softmax(axis = var_30355, x = aw_3873_cast_fp16)[name = tensor<string, []>("op_45519_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45520_cast_fp16 = softmax(axis = var_30355, x = aw_3875_cast_fp16)[name = tensor<string, []>("op_45520_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45521_cast_fp16 = softmax(axis = var_30355, x = aw_3877_cast_fp16)[name = tensor<string, []>("op_45521_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45522_cast_fp16 = softmax(axis = var_30355, x = aw_3879_cast_fp16)[name = tensor<string, []>("op_45522_cast_fp16")];
            tensor<string, []> var_45524_equation_0 = const()[name = tensor<string, []>("op_45524_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45524_cast_fp16 = einsum(equation = var_45524_equation_0, values = (var_45344_cast_fp16, var_45503_cast_fp16))[name = tensor<string, []>("op_45524_cast_fp16")];
            tensor<string, []> var_45526_equation_0 = const()[name = tensor<string, []>("op_45526_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45526_cast_fp16 = einsum(equation = var_45526_equation_0, values = (var_45348_cast_fp16, var_45504_cast_fp16))[name = tensor<string, []>("op_45526_cast_fp16")];
            tensor<string, []> var_45528_equation_0 = const()[name = tensor<string, []>("op_45528_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45528_cast_fp16 = einsum(equation = var_45528_equation_0, values = (var_45352_cast_fp16, var_45505_cast_fp16))[name = tensor<string, []>("op_45528_cast_fp16")];
            tensor<string, []> var_45530_equation_0 = const()[name = tensor<string, []>("op_45530_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45530_cast_fp16 = einsum(equation = var_45530_equation_0, values = (var_45356_cast_fp16, var_45506_cast_fp16))[name = tensor<string, []>("op_45530_cast_fp16")];
            tensor<string, []> var_45532_equation_0 = const()[name = tensor<string, []>("op_45532_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45532_cast_fp16 = einsum(equation = var_45532_equation_0, values = (var_45360_cast_fp16, var_45507_cast_fp16))[name = tensor<string, []>("op_45532_cast_fp16")];
            tensor<string, []> var_45534_equation_0 = const()[name = tensor<string, []>("op_45534_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45534_cast_fp16 = einsum(equation = var_45534_equation_0, values = (var_45364_cast_fp16, var_45508_cast_fp16))[name = tensor<string, []>("op_45534_cast_fp16")];
            tensor<string, []> var_45536_equation_0 = const()[name = tensor<string, []>("op_45536_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45536_cast_fp16 = einsum(equation = var_45536_equation_0, values = (var_45368_cast_fp16, var_45509_cast_fp16))[name = tensor<string, []>("op_45536_cast_fp16")];
            tensor<string, []> var_45538_equation_0 = const()[name = tensor<string, []>("op_45538_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45538_cast_fp16 = einsum(equation = var_45538_equation_0, values = (var_45372_cast_fp16, var_45510_cast_fp16))[name = tensor<string, []>("op_45538_cast_fp16")];
            tensor<string, []> var_45540_equation_0 = const()[name = tensor<string, []>("op_45540_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45540_cast_fp16 = einsum(equation = var_45540_equation_0, values = (var_45376_cast_fp16, var_45511_cast_fp16))[name = tensor<string, []>("op_45540_cast_fp16")];
            tensor<string, []> var_45542_equation_0 = const()[name = tensor<string, []>("op_45542_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45542_cast_fp16 = einsum(equation = var_45542_equation_0, values = (var_45380_cast_fp16, var_45512_cast_fp16))[name = tensor<string, []>("op_45542_cast_fp16")];
            tensor<string, []> var_45544_equation_0 = const()[name = tensor<string, []>("op_45544_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45544_cast_fp16 = einsum(equation = var_45544_equation_0, values = (var_45384_cast_fp16, var_45513_cast_fp16))[name = tensor<string, []>("op_45544_cast_fp16")];
            tensor<string, []> var_45546_equation_0 = const()[name = tensor<string, []>("op_45546_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45546_cast_fp16 = einsum(equation = var_45546_equation_0, values = (var_45388_cast_fp16, var_45514_cast_fp16))[name = tensor<string, []>("op_45546_cast_fp16")];
            tensor<string, []> var_45548_equation_0 = const()[name = tensor<string, []>("op_45548_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45548_cast_fp16 = einsum(equation = var_45548_equation_0, values = (var_45392_cast_fp16, var_45515_cast_fp16))[name = tensor<string, []>("op_45548_cast_fp16")];
            tensor<string, []> var_45550_equation_0 = const()[name = tensor<string, []>("op_45550_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45550_cast_fp16 = einsum(equation = var_45550_equation_0, values = (var_45396_cast_fp16, var_45516_cast_fp16))[name = tensor<string, []>("op_45550_cast_fp16")];
            tensor<string, []> var_45552_equation_0 = const()[name = tensor<string, []>("op_45552_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45552_cast_fp16 = einsum(equation = var_45552_equation_0, values = (var_45400_cast_fp16, var_45517_cast_fp16))[name = tensor<string, []>("op_45552_cast_fp16")];
            tensor<string, []> var_45554_equation_0 = const()[name = tensor<string, []>("op_45554_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45554_cast_fp16 = einsum(equation = var_45554_equation_0, values = (var_45404_cast_fp16, var_45518_cast_fp16))[name = tensor<string, []>("op_45554_cast_fp16")];
            tensor<string, []> var_45556_equation_0 = const()[name = tensor<string, []>("op_45556_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45556_cast_fp16 = einsum(equation = var_45556_equation_0, values = (var_45408_cast_fp16, var_45519_cast_fp16))[name = tensor<string, []>("op_45556_cast_fp16")];
            tensor<string, []> var_45558_equation_0 = const()[name = tensor<string, []>("op_45558_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45558_cast_fp16 = einsum(equation = var_45558_equation_0, values = (var_45412_cast_fp16, var_45520_cast_fp16))[name = tensor<string, []>("op_45558_cast_fp16")];
            tensor<string, []> var_45560_equation_0 = const()[name = tensor<string, []>("op_45560_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45560_cast_fp16 = einsum(equation = var_45560_equation_0, values = (var_45416_cast_fp16, var_45521_cast_fp16))[name = tensor<string, []>("op_45560_cast_fp16")];
            tensor<string, []> var_45562_equation_0 = const()[name = tensor<string, []>("op_45562_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45562_cast_fp16 = einsum(equation = var_45562_equation_0, values = (var_45420_cast_fp16, var_45522_cast_fp16))[name = tensor<string, []>("op_45562_cast_fp16")];
            tensor<bool, []> input_585_interleave_0 = const()[name = tensor<string, []>("input_585_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_585_cast_fp16 = concat(axis = var_30355, interleave = input_585_interleave_0, values = (var_45524_cast_fp16, var_45526_cast_fp16, var_45528_cast_fp16, var_45530_cast_fp16, var_45532_cast_fp16, var_45534_cast_fp16, var_45536_cast_fp16, var_45538_cast_fp16, var_45540_cast_fp16, var_45542_cast_fp16, var_45544_cast_fp16, var_45546_cast_fp16, var_45548_cast_fp16, var_45550_cast_fp16, var_45552_cast_fp16, var_45554_cast_fp16, var_45556_cast_fp16, var_45558_cast_fp16, var_45560_cast_fp16, var_45562_cast_fp16))[name = tensor<string, []>("input_585_cast_fp16")];
            tensor<string, []> var_45572_pad_type_0 = const()[name = tensor<string, []>("op_45572_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_45572_strides_0 = const()[name = tensor<string, []>("op_45572_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_45572_pad_0 = const()[name = tensor<string, []>("op_45572_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_45572_dilations_0 = const()[name = tensor<string, []>("op_45572_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_45572_groups_0 = const()[name = tensor<string, []>("op_45572_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_6_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(464325824))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(465554688))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_6_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_6_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_6_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(465554880)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_45572_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_6_attn1_to_out_0_bias_to_fp16, dilations = var_45572_dilations_0, groups = var_45572_groups_0, pad = var_45572_pad_0, pad_type = var_45572_pad_type_0, strides = var_45572_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_6_attn1_to_out_0_weight_to_fp16_palettized, x = input_585_cast_fp16)[name = tensor<string, []>("op_45572_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_303_cast_fp16 = add(x = var_45572_cast_fp16, y = inputs_301_cast_fp16)[name = tensor<string, []>("inputs_303_cast_fp16")];
            tensor<int32, [1]> hidden_states_399_axes_0 = const()[name = tensor<string, []>("hidden_states_399_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_399_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_399_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(465557504)))];
            tensor<fp16, [1280]> hidden_states_399_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_399_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(465560128)))];
            tensor<fp16, []> var_45582_to_fp16 = const()[name = tensor<string, []>("op_45582_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_399_cast_fp16 = layer_norm(axes = hidden_states_399_axes_0, beta = hidden_states_399_beta_0_to_fp16, epsilon = var_45582_to_fp16, gamma = hidden_states_399_gamma_0_to_fp16, x = inputs_303_cast_fp16)[name = tensor<string, []>("hidden_states_399_cast_fp16")];
            tensor<string, []> q_203_pad_type_0 = const()[name = tensor<string, []>("q_203_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_203_strides_0 = const()[name = tensor<string, []>("q_203_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_203_pad_0 = const()[name = tensor<string, []>("q_203_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_203_dilations_0 = const()[name = tensor<string, []>("q_203_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_203_groups_0 = const()[name = tensor<string, []>("q_203_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_6_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(465562752))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(466791616))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_6_attn2_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_203_cast_fp16 = conv(dilations = q_203_dilations_0, groups = q_203_groups_0, pad = q_203_pad_0, pad_type = q_203_pad_type_0, strides = q_203_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_6_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_399_cast_fp16)[name = tensor<string, []>("q_203_cast_fp16")];
            tensor<string, []> k_405_pad_type_0 = const()[name = tensor<string, []>("k_405_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_405_strides_0 = const()[name = tensor<string, []>("k_405_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_405_pad_0 = const()[name = tensor<string, []>("k_405_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_405_dilations_0 = const()[name = tensor<string, []>("k_405_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_405_groups_0 = const()[name = tensor<string, []>("k_405_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_6_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(466791808))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(468757952))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_6_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> k_405_cast_fp16 = conv(dilations = k_405_dilations_0, groups = k_405_groups_0, pad = k_405_pad_0, pad_type = k_405_pad_type_0, strides = k_405_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_6_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("k_405_cast_fp16")];
            tensor<string, []> v_203_pad_type_0 = const()[name = tensor<string, []>("v_203_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_203_strides_0 = const()[name = tensor<string, []>("v_203_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_203_pad_0 = const()[name = tensor<string, []>("v_203_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_203_dilations_0 = const()[name = tensor<string, []>("v_203_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_203_groups_0 = const()[name = tensor<string, []>("v_203_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_6_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(468758144))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(470724288))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_6_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> v_203_cast_fp16 = conv(dilations = v_203_dilations_0, groups = v_203_groups_0, pad = v_203_pad_0, pad_type = v_203_pad_type_0, strides = v_203_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_6_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("v_203_cast_fp16")];
            tensor<int32, [4]> var_45615_begin_0 = const()[name = tensor<string, []>("op_45615_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_45615_end_0 = const()[name = tensor<string, []>("op_45615_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_45615_end_mask_0 = const()[name = tensor<string, []>("op_45615_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45615_cast_fp16 = slice_by_index(begin = var_45615_begin_0, end = var_45615_end_0, end_mask = var_45615_end_mask_0, x = q_203_cast_fp16)[name = tensor<string, []>("op_45615_cast_fp16")];
            tensor<int32, [4]> var_45619_begin_0 = const()[name = tensor<string, []>("op_45619_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_45619_end_0 = const()[name = tensor<string, []>("op_45619_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_45619_end_mask_0 = const()[name = tensor<string, []>("op_45619_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45619_cast_fp16 = slice_by_index(begin = var_45619_begin_0, end = var_45619_end_0, end_mask = var_45619_end_mask_0, x = q_203_cast_fp16)[name = tensor<string, []>("op_45619_cast_fp16")];
            tensor<int32, [4]> var_45623_begin_0 = const()[name = tensor<string, []>("op_45623_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_45623_end_0 = const()[name = tensor<string, []>("op_45623_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_45623_end_mask_0 = const()[name = tensor<string, []>("op_45623_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45623_cast_fp16 = slice_by_index(begin = var_45623_begin_0, end = var_45623_end_0, end_mask = var_45623_end_mask_0, x = q_203_cast_fp16)[name = tensor<string, []>("op_45623_cast_fp16")];
            tensor<int32, [4]> var_45627_begin_0 = const()[name = tensor<string, []>("op_45627_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_45627_end_0 = const()[name = tensor<string, []>("op_45627_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_45627_end_mask_0 = const()[name = tensor<string, []>("op_45627_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45627_cast_fp16 = slice_by_index(begin = var_45627_begin_0, end = var_45627_end_0, end_mask = var_45627_end_mask_0, x = q_203_cast_fp16)[name = tensor<string, []>("op_45627_cast_fp16")];
            tensor<int32, [4]> var_45631_begin_0 = const()[name = tensor<string, []>("op_45631_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_45631_end_0 = const()[name = tensor<string, []>("op_45631_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_45631_end_mask_0 = const()[name = tensor<string, []>("op_45631_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45631_cast_fp16 = slice_by_index(begin = var_45631_begin_0, end = var_45631_end_0, end_mask = var_45631_end_mask_0, x = q_203_cast_fp16)[name = tensor<string, []>("op_45631_cast_fp16")];
            tensor<int32, [4]> var_45635_begin_0 = const()[name = tensor<string, []>("op_45635_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_45635_end_0 = const()[name = tensor<string, []>("op_45635_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_45635_end_mask_0 = const()[name = tensor<string, []>("op_45635_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45635_cast_fp16 = slice_by_index(begin = var_45635_begin_0, end = var_45635_end_0, end_mask = var_45635_end_mask_0, x = q_203_cast_fp16)[name = tensor<string, []>("op_45635_cast_fp16")];
            tensor<int32, [4]> var_45639_begin_0 = const()[name = tensor<string, []>("op_45639_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_45639_end_0 = const()[name = tensor<string, []>("op_45639_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_45639_end_mask_0 = const()[name = tensor<string, []>("op_45639_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45639_cast_fp16 = slice_by_index(begin = var_45639_begin_0, end = var_45639_end_0, end_mask = var_45639_end_mask_0, x = q_203_cast_fp16)[name = tensor<string, []>("op_45639_cast_fp16")];
            tensor<int32, [4]> var_45643_begin_0 = const()[name = tensor<string, []>("op_45643_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_45643_end_0 = const()[name = tensor<string, []>("op_45643_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_45643_end_mask_0 = const()[name = tensor<string, []>("op_45643_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45643_cast_fp16 = slice_by_index(begin = var_45643_begin_0, end = var_45643_end_0, end_mask = var_45643_end_mask_0, x = q_203_cast_fp16)[name = tensor<string, []>("op_45643_cast_fp16")];
            tensor<int32, [4]> var_45647_begin_0 = const()[name = tensor<string, []>("op_45647_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_45647_end_0 = const()[name = tensor<string, []>("op_45647_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_45647_end_mask_0 = const()[name = tensor<string, []>("op_45647_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45647_cast_fp16 = slice_by_index(begin = var_45647_begin_0, end = var_45647_end_0, end_mask = var_45647_end_mask_0, x = q_203_cast_fp16)[name = tensor<string, []>("op_45647_cast_fp16")];
            tensor<int32, [4]> var_45651_begin_0 = const()[name = tensor<string, []>("op_45651_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_45651_end_0 = const()[name = tensor<string, []>("op_45651_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_45651_end_mask_0 = const()[name = tensor<string, []>("op_45651_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45651_cast_fp16 = slice_by_index(begin = var_45651_begin_0, end = var_45651_end_0, end_mask = var_45651_end_mask_0, x = q_203_cast_fp16)[name = tensor<string, []>("op_45651_cast_fp16")];
            tensor<int32, [4]> var_45655_begin_0 = const()[name = tensor<string, []>("op_45655_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_45655_end_0 = const()[name = tensor<string, []>("op_45655_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_45655_end_mask_0 = const()[name = tensor<string, []>("op_45655_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45655_cast_fp16 = slice_by_index(begin = var_45655_begin_0, end = var_45655_end_0, end_mask = var_45655_end_mask_0, x = q_203_cast_fp16)[name = tensor<string, []>("op_45655_cast_fp16")];
            tensor<int32, [4]> var_45659_begin_0 = const()[name = tensor<string, []>("op_45659_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_45659_end_0 = const()[name = tensor<string, []>("op_45659_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_45659_end_mask_0 = const()[name = tensor<string, []>("op_45659_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45659_cast_fp16 = slice_by_index(begin = var_45659_begin_0, end = var_45659_end_0, end_mask = var_45659_end_mask_0, x = q_203_cast_fp16)[name = tensor<string, []>("op_45659_cast_fp16")];
            tensor<int32, [4]> var_45663_begin_0 = const()[name = tensor<string, []>("op_45663_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_45663_end_0 = const()[name = tensor<string, []>("op_45663_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_45663_end_mask_0 = const()[name = tensor<string, []>("op_45663_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45663_cast_fp16 = slice_by_index(begin = var_45663_begin_0, end = var_45663_end_0, end_mask = var_45663_end_mask_0, x = q_203_cast_fp16)[name = tensor<string, []>("op_45663_cast_fp16")];
            tensor<int32, [4]> var_45667_begin_0 = const()[name = tensor<string, []>("op_45667_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_45667_end_0 = const()[name = tensor<string, []>("op_45667_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_45667_end_mask_0 = const()[name = tensor<string, []>("op_45667_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45667_cast_fp16 = slice_by_index(begin = var_45667_begin_0, end = var_45667_end_0, end_mask = var_45667_end_mask_0, x = q_203_cast_fp16)[name = tensor<string, []>("op_45667_cast_fp16")];
            tensor<int32, [4]> var_45671_begin_0 = const()[name = tensor<string, []>("op_45671_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_45671_end_0 = const()[name = tensor<string, []>("op_45671_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_45671_end_mask_0 = const()[name = tensor<string, []>("op_45671_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45671_cast_fp16 = slice_by_index(begin = var_45671_begin_0, end = var_45671_end_0, end_mask = var_45671_end_mask_0, x = q_203_cast_fp16)[name = tensor<string, []>("op_45671_cast_fp16")];
            tensor<int32, [4]> var_45675_begin_0 = const()[name = tensor<string, []>("op_45675_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_45675_end_0 = const()[name = tensor<string, []>("op_45675_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_45675_end_mask_0 = const()[name = tensor<string, []>("op_45675_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45675_cast_fp16 = slice_by_index(begin = var_45675_begin_0, end = var_45675_end_0, end_mask = var_45675_end_mask_0, x = q_203_cast_fp16)[name = tensor<string, []>("op_45675_cast_fp16")];
            tensor<int32, [4]> var_45679_begin_0 = const()[name = tensor<string, []>("op_45679_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_45679_end_0 = const()[name = tensor<string, []>("op_45679_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_45679_end_mask_0 = const()[name = tensor<string, []>("op_45679_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45679_cast_fp16 = slice_by_index(begin = var_45679_begin_0, end = var_45679_end_0, end_mask = var_45679_end_mask_0, x = q_203_cast_fp16)[name = tensor<string, []>("op_45679_cast_fp16")];
            tensor<int32, [4]> var_45683_begin_0 = const()[name = tensor<string, []>("op_45683_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_45683_end_0 = const()[name = tensor<string, []>("op_45683_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_45683_end_mask_0 = const()[name = tensor<string, []>("op_45683_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45683_cast_fp16 = slice_by_index(begin = var_45683_begin_0, end = var_45683_end_0, end_mask = var_45683_end_mask_0, x = q_203_cast_fp16)[name = tensor<string, []>("op_45683_cast_fp16")];
            tensor<int32, [4]> var_45687_begin_0 = const()[name = tensor<string, []>("op_45687_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_45687_end_0 = const()[name = tensor<string, []>("op_45687_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_45687_end_mask_0 = const()[name = tensor<string, []>("op_45687_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45687_cast_fp16 = slice_by_index(begin = var_45687_begin_0, end = var_45687_end_0, end_mask = var_45687_end_mask_0, x = q_203_cast_fp16)[name = tensor<string, []>("op_45687_cast_fp16")];
            tensor<int32, [4]> var_45691_begin_0 = const()[name = tensor<string, []>("op_45691_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_45691_end_0 = const()[name = tensor<string, []>("op_45691_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_45691_end_mask_0 = const()[name = tensor<string, []>("op_45691_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45691_cast_fp16 = slice_by_index(begin = var_45691_begin_0, end = var_45691_end_0, end_mask = var_45691_end_mask_0, x = q_203_cast_fp16)[name = tensor<string, []>("op_45691_cast_fp16")];
            tensor<int32, [4]> k_407_perm_0 = const()[name = tensor<string, []>("k_407_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_45698_begin_0 = const()[name = tensor<string, []>("op_45698_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_45698_end_0 = const()[name = tensor<string, []>("op_45698_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_45698_end_mask_0 = const()[name = tensor<string, []>("op_45698_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> k_407_cast_fp16 = transpose(perm = k_407_perm_0, x = k_405_cast_fp16)[name = tensor<string, []>("transpose_38")];
            tensor<fp16, [2, 77, 1, 64]> var_45698_cast_fp16 = slice_by_index(begin = var_45698_begin_0, end = var_45698_end_0, end_mask = var_45698_end_mask_0, x = k_407_cast_fp16)[name = tensor<string, []>("op_45698_cast_fp16")];
            tensor<int32, [4]> var_45702_begin_0 = const()[name = tensor<string, []>("op_45702_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_45702_end_0 = const()[name = tensor<string, []>("op_45702_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_45702_end_mask_0 = const()[name = tensor<string, []>("op_45702_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_45702_cast_fp16 = slice_by_index(begin = var_45702_begin_0, end = var_45702_end_0, end_mask = var_45702_end_mask_0, x = k_407_cast_fp16)[name = tensor<string, []>("op_45702_cast_fp16")];
            tensor<int32, [4]> var_45706_begin_0 = const()[name = tensor<string, []>("op_45706_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_45706_end_0 = const()[name = tensor<string, []>("op_45706_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_45706_end_mask_0 = const()[name = tensor<string, []>("op_45706_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_45706_cast_fp16 = slice_by_index(begin = var_45706_begin_0, end = var_45706_end_0, end_mask = var_45706_end_mask_0, x = k_407_cast_fp16)[name = tensor<string, []>("op_45706_cast_fp16")];
            tensor<int32, [4]> var_45710_begin_0 = const()[name = tensor<string, []>("op_45710_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_45710_end_0 = const()[name = tensor<string, []>("op_45710_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_45710_end_mask_0 = const()[name = tensor<string, []>("op_45710_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_45710_cast_fp16 = slice_by_index(begin = var_45710_begin_0, end = var_45710_end_0, end_mask = var_45710_end_mask_0, x = k_407_cast_fp16)[name = tensor<string, []>("op_45710_cast_fp16")];
            tensor<int32, [4]> var_45714_begin_0 = const()[name = tensor<string, []>("op_45714_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_45714_end_0 = const()[name = tensor<string, []>("op_45714_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_45714_end_mask_0 = const()[name = tensor<string, []>("op_45714_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_45714_cast_fp16 = slice_by_index(begin = var_45714_begin_0, end = var_45714_end_0, end_mask = var_45714_end_mask_0, x = k_407_cast_fp16)[name = tensor<string, []>("op_45714_cast_fp16")];
            tensor<int32, [4]> var_45718_begin_0 = const()[name = tensor<string, []>("op_45718_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_45718_end_0 = const()[name = tensor<string, []>("op_45718_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_45718_end_mask_0 = const()[name = tensor<string, []>("op_45718_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_45718_cast_fp16 = slice_by_index(begin = var_45718_begin_0, end = var_45718_end_0, end_mask = var_45718_end_mask_0, x = k_407_cast_fp16)[name = tensor<string, []>("op_45718_cast_fp16")];
            tensor<int32, [4]> var_45722_begin_0 = const()[name = tensor<string, []>("op_45722_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_45722_end_0 = const()[name = tensor<string, []>("op_45722_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_45722_end_mask_0 = const()[name = tensor<string, []>("op_45722_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_45722_cast_fp16 = slice_by_index(begin = var_45722_begin_0, end = var_45722_end_0, end_mask = var_45722_end_mask_0, x = k_407_cast_fp16)[name = tensor<string, []>("op_45722_cast_fp16")];
            tensor<int32, [4]> var_45726_begin_0 = const()[name = tensor<string, []>("op_45726_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_45726_end_0 = const()[name = tensor<string, []>("op_45726_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_45726_end_mask_0 = const()[name = tensor<string, []>("op_45726_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_45726_cast_fp16 = slice_by_index(begin = var_45726_begin_0, end = var_45726_end_0, end_mask = var_45726_end_mask_0, x = k_407_cast_fp16)[name = tensor<string, []>("op_45726_cast_fp16")];
            tensor<int32, [4]> var_45730_begin_0 = const()[name = tensor<string, []>("op_45730_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_45730_end_0 = const()[name = tensor<string, []>("op_45730_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_45730_end_mask_0 = const()[name = tensor<string, []>("op_45730_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_45730_cast_fp16 = slice_by_index(begin = var_45730_begin_0, end = var_45730_end_0, end_mask = var_45730_end_mask_0, x = k_407_cast_fp16)[name = tensor<string, []>("op_45730_cast_fp16")];
            tensor<int32, [4]> var_45734_begin_0 = const()[name = tensor<string, []>("op_45734_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_45734_end_0 = const()[name = tensor<string, []>("op_45734_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_45734_end_mask_0 = const()[name = tensor<string, []>("op_45734_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_45734_cast_fp16 = slice_by_index(begin = var_45734_begin_0, end = var_45734_end_0, end_mask = var_45734_end_mask_0, x = k_407_cast_fp16)[name = tensor<string, []>("op_45734_cast_fp16")];
            tensor<int32, [4]> var_45738_begin_0 = const()[name = tensor<string, []>("op_45738_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_45738_end_0 = const()[name = tensor<string, []>("op_45738_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_45738_end_mask_0 = const()[name = tensor<string, []>("op_45738_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_45738_cast_fp16 = slice_by_index(begin = var_45738_begin_0, end = var_45738_end_0, end_mask = var_45738_end_mask_0, x = k_407_cast_fp16)[name = tensor<string, []>("op_45738_cast_fp16")];
            tensor<int32, [4]> var_45742_begin_0 = const()[name = tensor<string, []>("op_45742_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_45742_end_0 = const()[name = tensor<string, []>("op_45742_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_45742_end_mask_0 = const()[name = tensor<string, []>("op_45742_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_45742_cast_fp16 = slice_by_index(begin = var_45742_begin_0, end = var_45742_end_0, end_mask = var_45742_end_mask_0, x = k_407_cast_fp16)[name = tensor<string, []>("op_45742_cast_fp16")];
            tensor<int32, [4]> var_45746_begin_0 = const()[name = tensor<string, []>("op_45746_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_45746_end_0 = const()[name = tensor<string, []>("op_45746_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_45746_end_mask_0 = const()[name = tensor<string, []>("op_45746_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_45746_cast_fp16 = slice_by_index(begin = var_45746_begin_0, end = var_45746_end_0, end_mask = var_45746_end_mask_0, x = k_407_cast_fp16)[name = tensor<string, []>("op_45746_cast_fp16")];
            tensor<int32, [4]> var_45750_begin_0 = const()[name = tensor<string, []>("op_45750_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_45750_end_0 = const()[name = tensor<string, []>("op_45750_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_45750_end_mask_0 = const()[name = tensor<string, []>("op_45750_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_45750_cast_fp16 = slice_by_index(begin = var_45750_begin_0, end = var_45750_end_0, end_mask = var_45750_end_mask_0, x = k_407_cast_fp16)[name = tensor<string, []>("op_45750_cast_fp16")];
            tensor<int32, [4]> var_45754_begin_0 = const()[name = tensor<string, []>("op_45754_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_45754_end_0 = const()[name = tensor<string, []>("op_45754_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_45754_end_mask_0 = const()[name = tensor<string, []>("op_45754_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_45754_cast_fp16 = slice_by_index(begin = var_45754_begin_0, end = var_45754_end_0, end_mask = var_45754_end_mask_0, x = k_407_cast_fp16)[name = tensor<string, []>("op_45754_cast_fp16")];
            tensor<int32, [4]> var_45758_begin_0 = const()[name = tensor<string, []>("op_45758_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_45758_end_0 = const()[name = tensor<string, []>("op_45758_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_45758_end_mask_0 = const()[name = tensor<string, []>("op_45758_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_45758_cast_fp16 = slice_by_index(begin = var_45758_begin_0, end = var_45758_end_0, end_mask = var_45758_end_mask_0, x = k_407_cast_fp16)[name = tensor<string, []>("op_45758_cast_fp16")];
            tensor<int32, [4]> var_45762_begin_0 = const()[name = tensor<string, []>("op_45762_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_45762_end_0 = const()[name = tensor<string, []>("op_45762_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_45762_end_mask_0 = const()[name = tensor<string, []>("op_45762_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_45762_cast_fp16 = slice_by_index(begin = var_45762_begin_0, end = var_45762_end_0, end_mask = var_45762_end_mask_0, x = k_407_cast_fp16)[name = tensor<string, []>("op_45762_cast_fp16")];
            tensor<int32, [4]> var_45766_begin_0 = const()[name = tensor<string, []>("op_45766_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_45766_end_0 = const()[name = tensor<string, []>("op_45766_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_45766_end_mask_0 = const()[name = tensor<string, []>("op_45766_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_45766_cast_fp16 = slice_by_index(begin = var_45766_begin_0, end = var_45766_end_0, end_mask = var_45766_end_mask_0, x = k_407_cast_fp16)[name = tensor<string, []>("op_45766_cast_fp16")];
            tensor<int32, [4]> var_45770_begin_0 = const()[name = tensor<string, []>("op_45770_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_45770_end_0 = const()[name = tensor<string, []>("op_45770_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_45770_end_mask_0 = const()[name = tensor<string, []>("op_45770_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_45770_cast_fp16 = slice_by_index(begin = var_45770_begin_0, end = var_45770_end_0, end_mask = var_45770_end_mask_0, x = k_407_cast_fp16)[name = tensor<string, []>("op_45770_cast_fp16")];
            tensor<int32, [4]> var_45774_begin_0 = const()[name = tensor<string, []>("op_45774_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_45774_end_0 = const()[name = tensor<string, []>("op_45774_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_45774_end_mask_0 = const()[name = tensor<string, []>("op_45774_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_45774_cast_fp16 = slice_by_index(begin = var_45774_begin_0, end = var_45774_end_0, end_mask = var_45774_end_mask_0, x = k_407_cast_fp16)[name = tensor<string, []>("op_45774_cast_fp16")];
            tensor<int32, [4]> var_45776_begin_0 = const()[name = tensor<string, []>("op_45776_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_45776_end_0 = const()[name = tensor<string, []>("op_45776_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_45776_end_mask_0 = const()[name = tensor<string, []>("op_45776_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_45776_cast_fp16 = slice_by_index(begin = var_45776_begin_0, end = var_45776_end_0, end_mask = var_45776_end_mask_0, x = v_203_cast_fp16)[name = tensor<string, []>("op_45776_cast_fp16")];
            tensor<int32, [4]> var_45780_begin_0 = const()[name = tensor<string, []>("op_45780_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_45780_end_0 = const()[name = tensor<string, []>("op_45780_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_45780_end_mask_0 = const()[name = tensor<string, []>("op_45780_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_45780_cast_fp16 = slice_by_index(begin = var_45780_begin_0, end = var_45780_end_0, end_mask = var_45780_end_mask_0, x = v_203_cast_fp16)[name = tensor<string, []>("op_45780_cast_fp16")];
            tensor<int32, [4]> var_45784_begin_0 = const()[name = tensor<string, []>("op_45784_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_45784_end_0 = const()[name = tensor<string, []>("op_45784_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_45784_end_mask_0 = const()[name = tensor<string, []>("op_45784_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_45784_cast_fp16 = slice_by_index(begin = var_45784_begin_0, end = var_45784_end_0, end_mask = var_45784_end_mask_0, x = v_203_cast_fp16)[name = tensor<string, []>("op_45784_cast_fp16")];
            tensor<int32, [4]> var_45788_begin_0 = const()[name = tensor<string, []>("op_45788_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_45788_end_0 = const()[name = tensor<string, []>("op_45788_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_45788_end_mask_0 = const()[name = tensor<string, []>("op_45788_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_45788_cast_fp16 = slice_by_index(begin = var_45788_begin_0, end = var_45788_end_0, end_mask = var_45788_end_mask_0, x = v_203_cast_fp16)[name = tensor<string, []>("op_45788_cast_fp16")];
            tensor<int32, [4]> var_45792_begin_0 = const()[name = tensor<string, []>("op_45792_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_45792_end_0 = const()[name = tensor<string, []>("op_45792_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_45792_end_mask_0 = const()[name = tensor<string, []>("op_45792_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_45792_cast_fp16 = slice_by_index(begin = var_45792_begin_0, end = var_45792_end_0, end_mask = var_45792_end_mask_0, x = v_203_cast_fp16)[name = tensor<string, []>("op_45792_cast_fp16")];
            tensor<int32, [4]> var_45796_begin_0 = const()[name = tensor<string, []>("op_45796_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_45796_end_0 = const()[name = tensor<string, []>("op_45796_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_45796_end_mask_0 = const()[name = tensor<string, []>("op_45796_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_45796_cast_fp16 = slice_by_index(begin = var_45796_begin_0, end = var_45796_end_0, end_mask = var_45796_end_mask_0, x = v_203_cast_fp16)[name = tensor<string, []>("op_45796_cast_fp16")];
            tensor<int32, [4]> var_45800_begin_0 = const()[name = tensor<string, []>("op_45800_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_45800_end_0 = const()[name = tensor<string, []>("op_45800_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_45800_end_mask_0 = const()[name = tensor<string, []>("op_45800_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_45800_cast_fp16 = slice_by_index(begin = var_45800_begin_0, end = var_45800_end_0, end_mask = var_45800_end_mask_0, x = v_203_cast_fp16)[name = tensor<string, []>("op_45800_cast_fp16")];
            tensor<int32, [4]> var_45804_begin_0 = const()[name = tensor<string, []>("op_45804_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_45804_end_0 = const()[name = tensor<string, []>("op_45804_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_45804_end_mask_0 = const()[name = tensor<string, []>("op_45804_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_45804_cast_fp16 = slice_by_index(begin = var_45804_begin_0, end = var_45804_end_0, end_mask = var_45804_end_mask_0, x = v_203_cast_fp16)[name = tensor<string, []>("op_45804_cast_fp16")];
            tensor<int32, [4]> var_45808_begin_0 = const()[name = tensor<string, []>("op_45808_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_45808_end_0 = const()[name = tensor<string, []>("op_45808_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_45808_end_mask_0 = const()[name = tensor<string, []>("op_45808_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_45808_cast_fp16 = slice_by_index(begin = var_45808_begin_0, end = var_45808_end_0, end_mask = var_45808_end_mask_0, x = v_203_cast_fp16)[name = tensor<string, []>("op_45808_cast_fp16")];
            tensor<int32, [4]> var_45812_begin_0 = const()[name = tensor<string, []>("op_45812_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_45812_end_0 = const()[name = tensor<string, []>("op_45812_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_45812_end_mask_0 = const()[name = tensor<string, []>("op_45812_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_45812_cast_fp16 = slice_by_index(begin = var_45812_begin_0, end = var_45812_end_0, end_mask = var_45812_end_mask_0, x = v_203_cast_fp16)[name = tensor<string, []>("op_45812_cast_fp16")];
            tensor<int32, [4]> var_45816_begin_0 = const()[name = tensor<string, []>("op_45816_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_45816_end_0 = const()[name = tensor<string, []>("op_45816_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_45816_end_mask_0 = const()[name = tensor<string, []>("op_45816_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_45816_cast_fp16 = slice_by_index(begin = var_45816_begin_0, end = var_45816_end_0, end_mask = var_45816_end_mask_0, x = v_203_cast_fp16)[name = tensor<string, []>("op_45816_cast_fp16")];
            tensor<int32, [4]> var_45820_begin_0 = const()[name = tensor<string, []>("op_45820_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_45820_end_0 = const()[name = tensor<string, []>("op_45820_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_45820_end_mask_0 = const()[name = tensor<string, []>("op_45820_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_45820_cast_fp16 = slice_by_index(begin = var_45820_begin_0, end = var_45820_end_0, end_mask = var_45820_end_mask_0, x = v_203_cast_fp16)[name = tensor<string, []>("op_45820_cast_fp16")];
            tensor<int32, [4]> var_45824_begin_0 = const()[name = tensor<string, []>("op_45824_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_45824_end_0 = const()[name = tensor<string, []>("op_45824_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_45824_end_mask_0 = const()[name = tensor<string, []>("op_45824_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_45824_cast_fp16 = slice_by_index(begin = var_45824_begin_0, end = var_45824_end_0, end_mask = var_45824_end_mask_0, x = v_203_cast_fp16)[name = tensor<string, []>("op_45824_cast_fp16")];
            tensor<int32, [4]> var_45828_begin_0 = const()[name = tensor<string, []>("op_45828_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_45828_end_0 = const()[name = tensor<string, []>("op_45828_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_45828_end_mask_0 = const()[name = tensor<string, []>("op_45828_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_45828_cast_fp16 = slice_by_index(begin = var_45828_begin_0, end = var_45828_end_0, end_mask = var_45828_end_mask_0, x = v_203_cast_fp16)[name = tensor<string, []>("op_45828_cast_fp16")];
            tensor<int32, [4]> var_45832_begin_0 = const()[name = tensor<string, []>("op_45832_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_45832_end_0 = const()[name = tensor<string, []>("op_45832_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_45832_end_mask_0 = const()[name = tensor<string, []>("op_45832_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_45832_cast_fp16 = slice_by_index(begin = var_45832_begin_0, end = var_45832_end_0, end_mask = var_45832_end_mask_0, x = v_203_cast_fp16)[name = tensor<string, []>("op_45832_cast_fp16")];
            tensor<int32, [4]> var_45836_begin_0 = const()[name = tensor<string, []>("op_45836_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_45836_end_0 = const()[name = tensor<string, []>("op_45836_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_45836_end_mask_0 = const()[name = tensor<string, []>("op_45836_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_45836_cast_fp16 = slice_by_index(begin = var_45836_begin_0, end = var_45836_end_0, end_mask = var_45836_end_mask_0, x = v_203_cast_fp16)[name = tensor<string, []>("op_45836_cast_fp16")];
            tensor<int32, [4]> var_45840_begin_0 = const()[name = tensor<string, []>("op_45840_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_45840_end_0 = const()[name = tensor<string, []>("op_45840_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_45840_end_mask_0 = const()[name = tensor<string, []>("op_45840_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_45840_cast_fp16 = slice_by_index(begin = var_45840_begin_0, end = var_45840_end_0, end_mask = var_45840_end_mask_0, x = v_203_cast_fp16)[name = tensor<string, []>("op_45840_cast_fp16")];
            tensor<int32, [4]> var_45844_begin_0 = const()[name = tensor<string, []>("op_45844_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_45844_end_0 = const()[name = tensor<string, []>("op_45844_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_45844_end_mask_0 = const()[name = tensor<string, []>("op_45844_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_45844_cast_fp16 = slice_by_index(begin = var_45844_begin_0, end = var_45844_end_0, end_mask = var_45844_end_mask_0, x = v_203_cast_fp16)[name = tensor<string, []>("op_45844_cast_fp16")];
            tensor<int32, [4]> var_45848_begin_0 = const()[name = tensor<string, []>("op_45848_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_45848_end_0 = const()[name = tensor<string, []>("op_45848_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_45848_end_mask_0 = const()[name = tensor<string, []>("op_45848_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_45848_cast_fp16 = slice_by_index(begin = var_45848_begin_0, end = var_45848_end_0, end_mask = var_45848_end_mask_0, x = v_203_cast_fp16)[name = tensor<string, []>("op_45848_cast_fp16")];
            tensor<int32, [4]> var_45852_begin_0 = const()[name = tensor<string, []>("op_45852_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_45852_end_0 = const()[name = tensor<string, []>("op_45852_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_45852_end_mask_0 = const()[name = tensor<string, []>("op_45852_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_45852_cast_fp16 = slice_by_index(begin = var_45852_begin_0, end = var_45852_end_0, end_mask = var_45852_end_mask_0, x = v_203_cast_fp16)[name = tensor<string, []>("op_45852_cast_fp16")];
            tensor<string, []> var_45856_equation_0 = const()[name = tensor<string, []>("op_45856_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45856_cast_fp16 = einsum(equation = var_45856_equation_0, values = (var_45698_cast_fp16, var_45615_cast_fp16))[name = tensor<string, []>("op_45856_cast_fp16")];
            tensor<fp16, []> var_45857_to_fp16 = const()[name = tensor<string, []>("op_45857_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3881_cast_fp16 = mul(x = var_45856_cast_fp16, y = var_45857_to_fp16)[name = tensor<string, []>("aw_3881_cast_fp16")];
            tensor<string, []> var_45860_equation_0 = const()[name = tensor<string, []>("op_45860_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45860_cast_fp16 = einsum(equation = var_45860_equation_0, values = (var_45702_cast_fp16, var_45619_cast_fp16))[name = tensor<string, []>("op_45860_cast_fp16")];
            tensor<fp16, []> var_45861_to_fp16 = const()[name = tensor<string, []>("op_45861_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3883_cast_fp16 = mul(x = var_45860_cast_fp16, y = var_45861_to_fp16)[name = tensor<string, []>("aw_3883_cast_fp16")];
            tensor<string, []> var_45864_equation_0 = const()[name = tensor<string, []>("op_45864_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45864_cast_fp16 = einsum(equation = var_45864_equation_0, values = (var_45706_cast_fp16, var_45623_cast_fp16))[name = tensor<string, []>("op_45864_cast_fp16")];
            tensor<fp16, []> var_45865_to_fp16 = const()[name = tensor<string, []>("op_45865_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3885_cast_fp16 = mul(x = var_45864_cast_fp16, y = var_45865_to_fp16)[name = tensor<string, []>("aw_3885_cast_fp16")];
            tensor<string, []> var_45868_equation_0 = const()[name = tensor<string, []>("op_45868_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45868_cast_fp16 = einsum(equation = var_45868_equation_0, values = (var_45710_cast_fp16, var_45627_cast_fp16))[name = tensor<string, []>("op_45868_cast_fp16")];
            tensor<fp16, []> var_45869_to_fp16 = const()[name = tensor<string, []>("op_45869_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3887_cast_fp16 = mul(x = var_45868_cast_fp16, y = var_45869_to_fp16)[name = tensor<string, []>("aw_3887_cast_fp16")];
            tensor<string, []> var_45872_equation_0 = const()[name = tensor<string, []>("op_45872_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45872_cast_fp16 = einsum(equation = var_45872_equation_0, values = (var_45714_cast_fp16, var_45631_cast_fp16))[name = tensor<string, []>("op_45872_cast_fp16")];
            tensor<fp16, []> var_45873_to_fp16 = const()[name = tensor<string, []>("op_45873_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3889_cast_fp16 = mul(x = var_45872_cast_fp16, y = var_45873_to_fp16)[name = tensor<string, []>("aw_3889_cast_fp16")];
            tensor<string, []> var_45876_equation_0 = const()[name = tensor<string, []>("op_45876_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45876_cast_fp16 = einsum(equation = var_45876_equation_0, values = (var_45718_cast_fp16, var_45635_cast_fp16))[name = tensor<string, []>("op_45876_cast_fp16")];
            tensor<fp16, []> var_45877_to_fp16 = const()[name = tensor<string, []>("op_45877_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3891_cast_fp16 = mul(x = var_45876_cast_fp16, y = var_45877_to_fp16)[name = tensor<string, []>("aw_3891_cast_fp16")];
            tensor<string, []> var_45880_equation_0 = const()[name = tensor<string, []>("op_45880_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45880_cast_fp16 = einsum(equation = var_45880_equation_0, values = (var_45722_cast_fp16, var_45639_cast_fp16))[name = tensor<string, []>("op_45880_cast_fp16")];
            tensor<fp16, []> var_45881_to_fp16 = const()[name = tensor<string, []>("op_45881_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3893_cast_fp16 = mul(x = var_45880_cast_fp16, y = var_45881_to_fp16)[name = tensor<string, []>("aw_3893_cast_fp16")];
            tensor<string, []> var_45884_equation_0 = const()[name = tensor<string, []>("op_45884_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45884_cast_fp16 = einsum(equation = var_45884_equation_0, values = (var_45726_cast_fp16, var_45643_cast_fp16))[name = tensor<string, []>("op_45884_cast_fp16")];
            tensor<fp16, []> var_45885_to_fp16 = const()[name = tensor<string, []>("op_45885_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3895_cast_fp16 = mul(x = var_45884_cast_fp16, y = var_45885_to_fp16)[name = tensor<string, []>("aw_3895_cast_fp16")];
            tensor<string, []> var_45888_equation_0 = const()[name = tensor<string, []>("op_45888_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45888_cast_fp16 = einsum(equation = var_45888_equation_0, values = (var_45730_cast_fp16, var_45647_cast_fp16))[name = tensor<string, []>("op_45888_cast_fp16")];
            tensor<fp16, []> var_45889_to_fp16 = const()[name = tensor<string, []>("op_45889_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3897_cast_fp16 = mul(x = var_45888_cast_fp16, y = var_45889_to_fp16)[name = tensor<string, []>("aw_3897_cast_fp16")];
            tensor<string, []> var_45892_equation_0 = const()[name = tensor<string, []>("op_45892_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45892_cast_fp16 = einsum(equation = var_45892_equation_0, values = (var_45734_cast_fp16, var_45651_cast_fp16))[name = tensor<string, []>("op_45892_cast_fp16")];
            tensor<fp16, []> var_45893_to_fp16 = const()[name = tensor<string, []>("op_45893_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3899_cast_fp16 = mul(x = var_45892_cast_fp16, y = var_45893_to_fp16)[name = tensor<string, []>("aw_3899_cast_fp16")];
            tensor<string, []> var_45896_equation_0 = const()[name = tensor<string, []>("op_45896_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45896_cast_fp16 = einsum(equation = var_45896_equation_0, values = (var_45738_cast_fp16, var_45655_cast_fp16))[name = tensor<string, []>("op_45896_cast_fp16")];
            tensor<fp16, []> var_45897_to_fp16 = const()[name = tensor<string, []>("op_45897_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3901_cast_fp16 = mul(x = var_45896_cast_fp16, y = var_45897_to_fp16)[name = tensor<string, []>("aw_3901_cast_fp16")];
            tensor<string, []> var_45900_equation_0 = const()[name = tensor<string, []>("op_45900_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45900_cast_fp16 = einsum(equation = var_45900_equation_0, values = (var_45742_cast_fp16, var_45659_cast_fp16))[name = tensor<string, []>("op_45900_cast_fp16")];
            tensor<fp16, []> var_45901_to_fp16 = const()[name = tensor<string, []>("op_45901_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3903_cast_fp16 = mul(x = var_45900_cast_fp16, y = var_45901_to_fp16)[name = tensor<string, []>("aw_3903_cast_fp16")];
            tensor<string, []> var_45904_equation_0 = const()[name = tensor<string, []>("op_45904_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45904_cast_fp16 = einsum(equation = var_45904_equation_0, values = (var_45746_cast_fp16, var_45663_cast_fp16))[name = tensor<string, []>("op_45904_cast_fp16")];
            tensor<fp16, []> var_45905_to_fp16 = const()[name = tensor<string, []>("op_45905_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3905_cast_fp16 = mul(x = var_45904_cast_fp16, y = var_45905_to_fp16)[name = tensor<string, []>("aw_3905_cast_fp16")];
            tensor<string, []> var_45908_equation_0 = const()[name = tensor<string, []>("op_45908_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45908_cast_fp16 = einsum(equation = var_45908_equation_0, values = (var_45750_cast_fp16, var_45667_cast_fp16))[name = tensor<string, []>("op_45908_cast_fp16")];
            tensor<fp16, []> var_45909_to_fp16 = const()[name = tensor<string, []>("op_45909_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3907_cast_fp16 = mul(x = var_45908_cast_fp16, y = var_45909_to_fp16)[name = tensor<string, []>("aw_3907_cast_fp16")];
            tensor<string, []> var_45912_equation_0 = const()[name = tensor<string, []>("op_45912_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45912_cast_fp16 = einsum(equation = var_45912_equation_0, values = (var_45754_cast_fp16, var_45671_cast_fp16))[name = tensor<string, []>("op_45912_cast_fp16")];
            tensor<fp16, []> var_45913_to_fp16 = const()[name = tensor<string, []>("op_45913_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3909_cast_fp16 = mul(x = var_45912_cast_fp16, y = var_45913_to_fp16)[name = tensor<string, []>("aw_3909_cast_fp16")];
            tensor<string, []> var_45916_equation_0 = const()[name = tensor<string, []>("op_45916_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45916_cast_fp16 = einsum(equation = var_45916_equation_0, values = (var_45758_cast_fp16, var_45675_cast_fp16))[name = tensor<string, []>("op_45916_cast_fp16")];
            tensor<fp16, []> var_45917_to_fp16 = const()[name = tensor<string, []>("op_45917_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3911_cast_fp16 = mul(x = var_45916_cast_fp16, y = var_45917_to_fp16)[name = tensor<string, []>("aw_3911_cast_fp16")];
            tensor<string, []> var_45920_equation_0 = const()[name = tensor<string, []>("op_45920_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45920_cast_fp16 = einsum(equation = var_45920_equation_0, values = (var_45762_cast_fp16, var_45679_cast_fp16))[name = tensor<string, []>("op_45920_cast_fp16")];
            tensor<fp16, []> var_45921_to_fp16 = const()[name = tensor<string, []>("op_45921_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3913_cast_fp16 = mul(x = var_45920_cast_fp16, y = var_45921_to_fp16)[name = tensor<string, []>("aw_3913_cast_fp16")];
            tensor<string, []> var_45924_equation_0 = const()[name = tensor<string, []>("op_45924_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45924_cast_fp16 = einsum(equation = var_45924_equation_0, values = (var_45766_cast_fp16, var_45683_cast_fp16))[name = tensor<string, []>("op_45924_cast_fp16")];
            tensor<fp16, []> var_45925_to_fp16 = const()[name = tensor<string, []>("op_45925_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3915_cast_fp16 = mul(x = var_45924_cast_fp16, y = var_45925_to_fp16)[name = tensor<string, []>("aw_3915_cast_fp16")];
            tensor<string, []> var_45928_equation_0 = const()[name = tensor<string, []>("op_45928_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45928_cast_fp16 = einsum(equation = var_45928_equation_0, values = (var_45770_cast_fp16, var_45687_cast_fp16))[name = tensor<string, []>("op_45928_cast_fp16")];
            tensor<fp16, []> var_45929_to_fp16 = const()[name = tensor<string, []>("op_45929_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3917_cast_fp16 = mul(x = var_45928_cast_fp16, y = var_45929_to_fp16)[name = tensor<string, []>("aw_3917_cast_fp16")];
            tensor<string, []> var_45932_equation_0 = const()[name = tensor<string, []>("op_45932_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45932_cast_fp16 = einsum(equation = var_45932_equation_0, values = (var_45774_cast_fp16, var_45691_cast_fp16))[name = tensor<string, []>("op_45932_cast_fp16")];
            tensor<fp16, []> var_45933_to_fp16 = const()[name = tensor<string, []>("op_45933_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3919_cast_fp16 = mul(x = var_45932_cast_fp16, y = var_45933_to_fp16)[name = tensor<string, []>("aw_3919_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_45935_cast_fp16 = softmax(axis = var_30355, x = aw_3881_cast_fp16)[name = tensor<string, []>("op_45935_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_45936_cast_fp16 = softmax(axis = var_30355, x = aw_3883_cast_fp16)[name = tensor<string, []>("op_45936_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_45937_cast_fp16 = softmax(axis = var_30355, x = aw_3885_cast_fp16)[name = tensor<string, []>("op_45937_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_45938_cast_fp16 = softmax(axis = var_30355, x = aw_3887_cast_fp16)[name = tensor<string, []>("op_45938_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_45939_cast_fp16 = softmax(axis = var_30355, x = aw_3889_cast_fp16)[name = tensor<string, []>("op_45939_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_45940_cast_fp16 = softmax(axis = var_30355, x = aw_3891_cast_fp16)[name = tensor<string, []>("op_45940_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_45941_cast_fp16 = softmax(axis = var_30355, x = aw_3893_cast_fp16)[name = tensor<string, []>("op_45941_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_45942_cast_fp16 = softmax(axis = var_30355, x = aw_3895_cast_fp16)[name = tensor<string, []>("op_45942_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_45943_cast_fp16 = softmax(axis = var_30355, x = aw_3897_cast_fp16)[name = tensor<string, []>("op_45943_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_45944_cast_fp16 = softmax(axis = var_30355, x = aw_3899_cast_fp16)[name = tensor<string, []>("op_45944_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_45945_cast_fp16 = softmax(axis = var_30355, x = aw_3901_cast_fp16)[name = tensor<string, []>("op_45945_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_45946_cast_fp16 = softmax(axis = var_30355, x = aw_3903_cast_fp16)[name = tensor<string, []>("op_45946_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_45947_cast_fp16 = softmax(axis = var_30355, x = aw_3905_cast_fp16)[name = tensor<string, []>("op_45947_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_45948_cast_fp16 = softmax(axis = var_30355, x = aw_3907_cast_fp16)[name = tensor<string, []>("op_45948_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_45949_cast_fp16 = softmax(axis = var_30355, x = aw_3909_cast_fp16)[name = tensor<string, []>("op_45949_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_45950_cast_fp16 = softmax(axis = var_30355, x = aw_3911_cast_fp16)[name = tensor<string, []>("op_45950_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_45951_cast_fp16 = softmax(axis = var_30355, x = aw_3913_cast_fp16)[name = tensor<string, []>("op_45951_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_45952_cast_fp16 = softmax(axis = var_30355, x = aw_3915_cast_fp16)[name = tensor<string, []>("op_45952_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_45953_cast_fp16 = softmax(axis = var_30355, x = aw_3917_cast_fp16)[name = tensor<string, []>("op_45953_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_45954_cast_fp16 = softmax(axis = var_30355, x = aw_3919_cast_fp16)[name = tensor<string, []>("op_45954_cast_fp16")];
            tensor<string, []> var_45956_equation_0 = const()[name = tensor<string, []>("op_45956_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45956_cast_fp16 = einsum(equation = var_45956_equation_0, values = (var_45776_cast_fp16, var_45935_cast_fp16))[name = tensor<string, []>("op_45956_cast_fp16")];
            tensor<string, []> var_45958_equation_0 = const()[name = tensor<string, []>("op_45958_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45958_cast_fp16 = einsum(equation = var_45958_equation_0, values = (var_45780_cast_fp16, var_45936_cast_fp16))[name = tensor<string, []>("op_45958_cast_fp16")];
            tensor<string, []> var_45960_equation_0 = const()[name = tensor<string, []>("op_45960_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45960_cast_fp16 = einsum(equation = var_45960_equation_0, values = (var_45784_cast_fp16, var_45937_cast_fp16))[name = tensor<string, []>("op_45960_cast_fp16")];
            tensor<string, []> var_45962_equation_0 = const()[name = tensor<string, []>("op_45962_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45962_cast_fp16 = einsum(equation = var_45962_equation_0, values = (var_45788_cast_fp16, var_45938_cast_fp16))[name = tensor<string, []>("op_45962_cast_fp16")];
            tensor<string, []> var_45964_equation_0 = const()[name = tensor<string, []>("op_45964_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45964_cast_fp16 = einsum(equation = var_45964_equation_0, values = (var_45792_cast_fp16, var_45939_cast_fp16))[name = tensor<string, []>("op_45964_cast_fp16")];
            tensor<string, []> var_45966_equation_0 = const()[name = tensor<string, []>("op_45966_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45966_cast_fp16 = einsum(equation = var_45966_equation_0, values = (var_45796_cast_fp16, var_45940_cast_fp16))[name = tensor<string, []>("op_45966_cast_fp16")];
            tensor<string, []> var_45968_equation_0 = const()[name = tensor<string, []>("op_45968_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45968_cast_fp16 = einsum(equation = var_45968_equation_0, values = (var_45800_cast_fp16, var_45941_cast_fp16))[name = tensor<string, []>("op_45968_cast_fp16")];
            tensor<string, []> var_45970_equation_0 = const()[name = tensor<string, []>("op_45970_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45970_cast_fp16 = einsum(equation = var_45970_equation_0, values = (var_45804_cast_fp16, var_45942_cast_fp16))[name = tensor<string, []>("op_45970_cast_fp16")];
            tensor<string, []> var_45972_equation_0 = const()[name = tensor<string, []>("op_45972_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45972_cast_fp16 = einsum(equation = var_45972_equation_0, values = (var_45808_cast_fp16, var_45943_cast_fp16))[name = tensor<string, []>("op_45972_cast_fp16")];
            tensor<string, []> var_45974_equation_0 = const()[name = tensor<string, []>("op_45974_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45974_cast_fp16 = einsum(equation = var_45974_equation_0, values = (var_45812_cast_fp16, var_45944_cast_fp16))[name = tensor<string, []>("op_45974_cast_fp16")];
            tensor<string, []> var_45976_equation_0 = const()[name = tensor<string, []>("op_45976_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45976_cast_fp16 = einsum(equation = var_45976_equation_0, values = (var_45816_cast_fp16, var_45945_cast_fp16))[name = tensor<string, []>("op_45976_cast_fp16")];
            tensor<string, []> var_45978_equation_0 = const()[name = tensor<string, []>("op_45978_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45978_cast_fp16 = einsum(equation = var_45978_equation_0, values = (var_45820_cast_fp16, var_45946_cast_fp16))[name = tensor<string, []>("op_45978_cast_fp16")];
            tensor<string, []> var_45980_equation_0 = const()[name = tensor<string, []>("op_45980_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45980_cast_fp16 = einsum(equation = var_45980_equation_0, values = (var_45824_cast_fp16, var_45947_cast_fp16))[name = tensor<string, []>("op_45980_cast_fp16")];
            tensor<string, []> var_45982_equation_0 = const()[name = tensor<string, []>("op_45982_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45982_cast_fp16 = einsum(equation = var_45982_equation_0, values = (var_45828_cast_fp16, var_45948_cast_fp16))[name = tensor<string, []>("op_45982_cast_fp16")];
            tensor<string, []> var_45984_equation_0 = const()[name = tensor<string, []>("op_45984_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45984_cast_fp16 = einsum(equation = var_45984_equation_0, values = (var_45832_cast_fp16, var_45949_cast_fp16))[name = tensor<string, []>("op_45984_cast_fp16")];
            tensor<string, []> var_45986_equation_0 = const()[name = tensor<string, []>("op_45986_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45986_cast_fp16 = einsum(equation = var_45986_equation_0, values = (var_45836_cast_fp16, var_45950_cast_fp16))[name = tensor<string, []>("op_45986_cast_fp16")];
            tensor<string, []> var_45988_equation_0 = const()[name = tensor<string, []>("op_45988_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45988_cast_fp16 = einsum(equation = var_45988_equation_0, values = (var_45840_cast_fp16, var_45951_cast_fp16))[name = tensor<string, []>("op_45988_cast_fp16")];
            tensor<string, []> var_45990_equation_0 = const()[name = tensor<string, []>("op_45990_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45990_cast_fp16 = einsum(equation = var_45990_equation_0, values = (var_45844_cast_fp16, var_45952_cast_fp16))[name = tensor<string, []>("op_45990_cast_fp16")];
            tensor<string, []> var_45992_equation_0 = const()[name = tensor<string, []>("op_45992_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45992_cast_fp16 = einsum(equation = var_45992_equation_0, values = (var_45848_cast_fp16, var_45953_cast_fp16))[name = tensor<string, []>("op_45992_cast_fp16")];
            tensor<string, []> var_45994_equation_0 = const()[name = tensor<string, []>("op_45994_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45994_cast_fp16 = einsum(equation = var_45994_equation_0, values = (var_45852_cast_fp16, var_45954_cast_fp16))[name = tensor<string, []>("op_45994_cast_fp16")];
            tensor<bool, []> input_587_interleave_0 = const()[name = tensor<string, []>("input_587_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_587_cast_fp16 = concat(axis = var_30355, interleave = input_587_interleave_0, values = (var_45956_cast_fp16, var_45958_cast_fp16, var_45960_cast_fp16, var_45962_cast_fp16, var_45964_cast_fp16, var_45966_cast_fp16, var_45968_cast_fp16, var_45970_cast_fp16, var_45972_cast_fp16, var_45974_cast_fp16, var_45976_cast_fp16, var_45978_cast_fp16, var_45980_cast_fp16, var_45982_cast_fp16, var_45984_cast_fp16, var_45986_cast_fp16, var_45988_cast_fp16, var_45990_cast_fp16, var_45992_cast_fp16, var_45994_cast_fp16))[name = tensor<string, []>("input_587_cast_fp16")];
            tensor<string, []> var_46004_pad_type_0 = const()[name = tensor<string, []>("op_46004_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_46004_strides_0 = const()[name = tensor<string, []>("op_46004_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_46004_pad_0 = const()[name = tensor<string, []>("op_46004_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_46004_dilations_0 = const()[name = tensor<string, []>("op_46004_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_46004_groups_0 = const()[name = tensor<string, []>("op_46004_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_6_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(470724480))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(471953344))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_6_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_6_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_6_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(471953536)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_46004_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_6_attn2_to_out_0_bias_to_fp16, dilations = var_46004_dilations_0, groups = var_46004_groups_0, pad = var_46004_pad_0, pad_type = var_46004_pad_type_0, strides = var_46004_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_6_attn2_to_out_0_weight_to_fp16_palettized, x = input_587_cast_fp16)[name = tensor<string, []>("op_46004_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_305_cast_fp16 = add(x = var_46004_cast_fp16, y = inputs_303_cast_fp16)[name = tensor<string, []>("inputs_305_cast_fp16")];
            tensor<int32, [1]> input_589_axes_0 = const()[name = tensor<string, []>("input_589_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> input_589_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_589_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(471956160)))];
            tensor<fp16, [1280]> input_589_beta_0_to_fp16 = const()[name = tensor<string, []>("input_589_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(471958784)))];
            tensor<fp16, []> var_46014_to_fp16 = const()[name = tensor<string, []>("op_46014_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> input_589_cast_fp16 = layer_norm(axes = input_589_axes_0, beta = input_589_beta_0_to_fp16, epsilon = var_46014_to_fp16, gamma = input_589_gamma_0_to_fp16, x = inputs_305_cast_fp16)[name = tensor<string, []>("input_589_cast_fp16")];
            tensor<string, []> var_46034_pad_type_0 = const()[name = tensor<string, []>("op_46034_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_46034_strides_0 = const()[name = tensor<string, []>("op_46034_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_46034_pad_0 = const()[name = tensor<string, []>("op_46034_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_46034_dilations_0 = const()[name = tensor<string, []>("op_46034_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_46034_groups_0 = const()[name = tensor<string, []>("op_46034_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_6_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [9830400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(471961408))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(481791872))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_6_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([10240, 1280, 1, 1])];
            tensor<fp16, [10240]> up_blocks_0_attentions_1_transformer_blocks_6_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_6_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(481792064)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_46034_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_6_ff_net_0_proj_bias_to_fp16, dilations = var_46034_dilations_0, groups = var_46034_groups_0, pad = var_46034_pad_0, pad_type = var_46034_pad_type_0, strides = var_46034_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_6_ff_net_0_proj_weight_to_fp16_palettized, x = input_589_cast_fp16)[name = tensor<string, []>("op_46034_cast_fp16")];
            tensor<int32, [2]> var_46035_split_sizes_0 = const()[name = tensor<string, []>("op_46035_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_46035_axis_0 = const()[name = tensor<string, []>("op_46035_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_46035_cast_fp16_0, tensor<fp16, [2, 5120, 1, 1024]> var_46035_cast_fp16_1 = split(axis = var_46035_axis_0, split_sizes = var_46035_split_sizes_0, x = var_46034_cast_fp16)[name = tensor<string, []>("op_46035_cast_fp16")];
            tensor<string, []> var_46037_mode_0 = const()[name = tensor<string, []>("op_46037_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_46037_cast_fp16 = gelu(mode = var_46037_mode_0, x = var_46035_cast_fp16_1)[name = tensor<string, []>("op_46037_cast_fp16")];
            tensor<fp16, [2, 5120, 1, 1024]> input_591_cast_fp16 = mul(x = var_46035_cast_fp16_0, y = var_46037_cast_fp16)[name = tensor<string, []>("input_591_cast_fp16")];
            tensor<string, []> var_46045_pad_type_0 = const()[name = tensor<string, []>("op_46045_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_46045_strides_0 = const()[name = tensor<string, []>("op_46045_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_46045_pad_0 = const()[name = tensor<string, []>("op_46045_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_46045_dilations_0 = const()[name = tensor<string, []>("op_46045_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_46045_groups_0 = const()[name = tensor<string, []>("op_46045_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_6_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [4915200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(481812608))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(486727872))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_6_ff_net_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 5120, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_6_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_6_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(486728064)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_46045_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_6_ff_net_2_bias_to_fp16, dilations = var_46045_dilations_0, groups = var_46045_groups_0, pad = var_46045_pad_0, pad_type = var_46045_pad_type_0, strides = var_46045_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_6_ff_net_2_weight_to_fp16_palettized, x = input_591_cast_fp16)[name = tensor<string, []>("op_46045_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_307_cast_fp16 = add(x = var_46045_cast_fp16, y = inputs_305_cast_fp16)[name = tensor<string, []>("inputs_307_cast_fp16")];
            tensor<int32, [1]> hidden_states_403_axes_0 = const()[name = tensor<string, []>("hidden_states_403_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_403_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_403_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(486730688)))];
            tensor<fp16, [1280]> hidden_states_403_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_403_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(486733312)))];
            tensor<fp16, []> var_46061_to_fp16 = const()[name = tensor<string, []>("op_46061_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_403_cast_fp16 = layer_norm(axes = hidden_states_403_axes_0, beta = hidden_states_403_beta_0_to_fp16, epsilon = var_46061_to_fp16, gamma = hidden_states_403_gamma_0_to_fp16, x = inputs_307_cast_fp16)[name = tensor<string, []>("hidden_states_403_cast_fp16")];
            tensor<string, []> q_205_pad_type_0 = const()[name = tensor<string, []>("q_205_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_205_strides_0 = const()[name = tensor<string, []>("q_205_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_205_pad_0 = const()[name = tensor<string, []>("q_205_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_205_dilations_0 = const()[name = tensor<string, []>("q_205_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_205_groups_0 = const()[name = tensor<string, []>("q_205_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_7_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(486735936))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(487964800))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_7_attn1_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_205_cast_fp16 = conv(dilations = q_205_dilations_0, groups = q_205_groups_0, pad = q_205_pad_0, pad_type = q_205_pad_type_0, strides = q_205_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_7_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_403_cast_fp16)[name = tensor<string, []>("q_205_cast_fp16")];
            tensor<string, []> k_409_pad_type_0 = const()[name = tensor<string, []>("k_409_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_409_strides_0 = const()[name = tensor<string, []>("k_409_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_409_pad_0 = const()[name = tensor<string, []>("k_409_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_409_dilations_0 = const()[name = tensor<string, []>("k_409_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_409_groups_0 = const()[name = tensor<string, []>("k_409_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_7_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(487964992))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(489193856))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_7_attn1_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> k_409_cast_fp16 = conv(dilations = k_409_dilations_0, groups = k_409_groups_0, pad = k_409_pad_0, pad_type = k_409_pad_type_0, strides = k_409_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_7_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_403_cast_fp16)[name = tensor<string, []>("k_409_cast_fp16")];
            tensor<string, []> v_205_pad_type_0 = const()[name = tensor<string, []>("v_205_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_205_strides_0 = const()[name = tensor<string, []>("v_205_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_205_pad_0 = const()[name = tensor<string, []>("v_205_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_205_dilations_0 = const()[name = tensor<string, []>("v_205_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_205_groups_0 = const()[name = tensor<string, []>("v_205_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_7_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(489194048))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(490422912))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_7_attn1_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> v_205_cast_fp16 = conv(dilations = v_205_dilations_0, groups = v_205_groups_0, pad = v_205_pad_0, pad_type = v_205_pad_type_0, strides = v_205_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_7_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_403_cast_fp16)[name = tensor<string, []>("v_205_cast_fp16")];
            tensor<int32, [4]> var_46094_begin_0 = const()[name = tensor<string, []>("op_46094_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_46094_end_0 = const()[name = tensor<string, []>("op_46094_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_46094_end_mask_0 = const()[name = tensor<string, []>("op_46094_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46094_cast_fp16 = slice_by_index(begin = var_46094_begin_0, end = var_46094_end_0, end_mask = var_46094_end_mask_0, x = q_205_cast_fp16)[name = tensor<string, []>("op_46094_cast_fp16")];
            tensor<int32, [4]> var_46098_begin_0 = const()[name = tensor<string, []>("op_46098_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_46098_end_0 = const()[name = tensor<string, []>("op_46098_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_46098_end_mask_0 = const()[name = tensor<string, []>("op_46098_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46098_cast_fp16 = slice_by_index(begin = var_46098_begin_0, end = var_46098_end_0, end_mask = var_46098_end_mask_0, x = q_205_cast_fp16)[name = tensor<string, []>("op_46098_cast_fp16")];
            tensor<int32, [4]> var_46102_begin_0 = const()[name = tensor<string, []>("op_46102_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_46102_end_0 = const()[name = tensor<string, []>("op_46102_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_46102_end_mask_0 = const()[name = tensor<string, []>("op_46102_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46102_cast_fp16 = slice_by_index(begin = var_46102_begin_0, end = var_46102_end_0, end_mask = var_46102_end_mask_0, x = q_205_cast_fp16)[name = tensor<string, []>("op_46102_cast_fp16")];
            tensor<int32, [4]> var_46106_begin_0 = const()[name = tensor<string, []>("op_46106_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_46106_end_0 = const()[name = tensor<string, []>("op_46106_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_46106_end_mask_0 = const()[name = tensor<string, []>("op_46106_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46106_cast_fp16 = slice_by_index(begin = var_46106_begin_0, end = var_46106_end_0, end_mask = var_46106_end_mask_0, x = q_205_cast_fp16)[name = tensor<string, []>("op_46106_cast_fp16")];
            tensor<int32, [4]> var_46110_begin_0 = const()[name = tensor<string, []>("op_46110_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_46110_end_0 = const()[name = tensor<string, []>("op_46110_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_46110_end_mask_0 = const()[name = tensor<string, []>("op_46110_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46110_cast_fp16 = slice_by_index(begin = var_46110_begin_0, end = var_46110_end_0, end_mask = var_46110_end_mask_0, x = q_205_cast_fp16)[name = tensor<string, []>("op_46110_cast_fp16")];
            tensor<int32, [4]> var_46114_begin_0 = const()[name = tensor<string, []>("op_46114_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_46114_end_0 = const()[name = tensor<string, []>("op_46114_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_46114_end_mask_0 = const()[name = tensor<string, []>("op_46114_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46114_cast_fp16 = slice_by_index(begin = var_46114_begin_0, end = var_46114_end_0, end_mask = var_46114_end_mask_0, x = q_205_cast_fp16)[name = tensor<string, []>("op_46114_cast_fp16")];
            tensor<int32, [4]> var_46118_begin_0 = const()[name = tensor<string, []>("op_46118_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_46118_end_0 = const()[name = tensor<string, []>("op_46118_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_46118_end_mask_0 = const()[name = tensor<string, []>("op_46118_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46118_cast_fp16 = slice_by_index(begin = var_46118_begin_0, end = var_46118_end_0, end_mask = var_46118_end_mask_0, x = q_205_cast_fp16)[name = tensor<string, []>("op_46118_cast_fp16")];
            tensor<int32, [4]> var_46122_begin_0 = const()[name = tensor<string, []>("op_46122_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_46122_end_0 = const()[name = tensor<string, []>("op_46122_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_46122_end_mask_0 = const()[name = tensor<string, []>("op_46122_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46122_cast_fp16 = slice_by_index(begin = var_46122_begin_0, end = var_46122_end_0, end_mask = var_46122_end_mask_0, x = q_205_cast_fp16)[name = tensor<string, []>("op_46122_cast_fp16")];
            tensor<int32, [4]> var_46126_begin_0 = const()[name = tensor<string, []>("op_46126_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_46126_end_0 = const()[name = tensor<string, []>("op_46126_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_46126_end_mask_0 = const()[name = tensor<string, []>("op_46126_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46126_cast_fp16 = slice_by_index(begin = var_46126_begin_0, end = var_46126_end_0, end_mask = var_46126_end_mask_0, x = q_205_cast_fp16)[name = tensor<string, []>("op_46126_cast_fp16")];
            tensor<int32, [4]> var_46130_begin_0 = const()[name = tensor<string, []>("op_46130_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_46130_end_0 = const()[name = tensor<string, []>("op_46130_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_46130_end_mask_0 = const()[name = tensor<string, []>("op_46130_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46130_cast_fp16 = slice_by_index(begin = var_46130_begin_0, end = var_46130_end_0, end_mask = var_46130_end_mask_0, x = q_205_cast_fp16)[name = tensor<string, []>("op_46130_cast_fp16")];
            tensor<int32, [4]> var_46134_begin_0 = const()[name = tensor<string, []>("op_46134_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_46134_end_0 = const()[name = tensor<string, []>("op_46134_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_46134_end_mask_0 = const()[name = tensor<string, []>("op_46134_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46134_cast_fp16 = slice_by_index(begin = var_46134_begin_0, end = var_46134_end_0, end_mask = var_46134_end_mask_0, x = q_205_cast_fp16)[name = tensor<string, []>("op_46134_cast_fp16")];
            tensor<int32, [4]> var_46138_begin_0 = const()[name = tensor<string, []>("op_46138_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_46138_end_0 = const()[name = tensor<string, []>("op_46138_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_46138_end_mask_0 = const()[name = tensor<string, []>("op_46138_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46138_cast_fp16 = slice_by_index(begin = var_46138_begin_0, end = var_46138_end_0, end_mask = var_46138_end_mask_0, x = q_205_cast_fp16)[name = tensor<string, []>("op_46138_cast_fp16")];
            tensor<int32, [4]> var_46142_begin_0 = const()[name = tensor<string, []>("op_46142_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_46142_end_0 = const()[name = tensor<string, []>("op_46142_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_46142_end_mask_0 = const()[name = tensor<string, []>("op_46142_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46142_cast_fp16 = slice_by_index(begin = var_46142_begin_0, end = var_46142_end_0, end_mask = var_46142_end_mask_0, x = q_205_cast_fp16)[name = tensor<string, []>("op_46142_cast_fp16")];
            tensor<int32, [4]> var_46146_begin_0 = const()[name = tensor<string, []>("op_46146_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_46146_end_0 = const()[name = tensor<string, []>("op_46146_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_46146_end_mask_0 = const()[name = tensor<string, []>("op_46146_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46146_cast_fp16 = slice_by_index(begin = var_46146_begin_0, end = var_46146_end_0, end_mask = var_46146_end_mask_0, x = q_205_cast_fp16)[name = tensor<string, []>("op_46146_cast_fp16")];
            tensor<int32, [4]> var_46150_begin_0 = const()[name = tensor<string, []>("op_46150_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_46150_end_0 = const()[name = tensor<string, []>("op_46150_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_46150_end_mask_0 = const()[name = tensor<string, []>("op_46150_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46150_cast_fp16 = slice_by_index(begin = var_46150_begin_0, end = var_46150_end_0, end_mask = var_46150_end_mask_0, x = q_205_cast_fp16)[name = tensor<string, []>("op_46150_cast_fp16")];
            tensor<int32, [4]> var_46154_begin_0 = const()[name = tensor<string, []>("op_46154_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_46154_end_0 = const()[name = tensor<string, []>("op_46154_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_46154_end_mask_0 = const()[name = tensor<string, []>("op_46154_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46154_cast_fp16 = slice_by_index(begin = var_46154_begin_0, end = var_46154_end_0, end_mask = var_46154_end_mask_0, x = q_205_cast_fp16)[name = tensor<string, []>("op_46154_cast_fp16")];
            tensor<int32, [4]> var_46158_begin_0 = const()[name = tensor<string, []>("op_46158_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_46158_end_0 = const()[name = tensor<string, []>("op_46158_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_46158_end_mask_0 = const()[name = tensor<string, []>("op_46158_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46158_cast_fp16 = slice_by_index(begin = var_46158_begin_0, end = var_46158_end_0, end_mask = var_46158_end_mask_0, x = q_205_cast_fp16)[name = tensor<string, []>("op_46158_cast_fp16")];
            tensor<int32, [4]> var_46162_begin_0 = const()[name = tensor<string, []>("op_46162_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_46162_end_0 = const()[name = tensor<string, []>("op_46162_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_46162_end_mask_0 = const()[name = tensor<string, []>("op_46162_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46162_cast_fp16 = slice_by_index(begin = var_46162_begin_0, end = var_46162_end_0, end_mask = var_46162_end_mask_0, x = q_205_cast_fp16)[name = tensor<string, []>("op_46162_cast_fp16")];
            tensor<int32, [4]> var_46166_begin_0 = const()[name = tensor<string, []>("op_46166_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_46166_end_0 = const()[name = tensor<string, []>("op_46166_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_46166_end_mask_0 = const()[name = tensor<string, []>("op_46166_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46166_cast_fp16 = slice_by_index(begin = var_46166_begin_0, end = var_46166_end_0, end_mask = var_46166_end_mask_0, x = q_205_cast_fp16)[name = tensor<string, []>("op_46166_cast_fp16")];
            tensor<int32, [4]> var_46170_begin_0 = const()[name = tensor<string, []>("op_46170_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_46170_end_0 = const()[name = tensor<string, []>("op_46170_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_46170_end_mask_0 = const()[name = tensor<string, []>("op_46170_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46170_cast_fp16 = slice_by_index(begin = var_46170_begin_0, end = var_46170_end_0, end_mask = var_46170_end_mask_0, x = q_205_cast_fp16)[name = tensor<string, []>("op_46170_cast_fp16")];
            tensor<int32, [4]> k_411_perm_0 = const()[name = tensor<string, []>("k_411_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_46177_begin_0 = const()[name = tensor<string, []>("op_46177_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_46177_end_0 = const()[name = tensor<string, []>("op_46177_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_46177_end_mask_0 = const()[name = tensor<string, []>("op_46177_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> k_411_cast_fp16 = transpose(perm = k_411_perm_0, x = k_409_cast_fp16)[name = tensor<string, []>("transpose_37")];
            tensor<fp16, [2, 1024, 1, 64]> var_46177_cast_fp16 = slice_by_index(begin = var_46177_begin_0, end = var_46177_end_0, end_mask = var_46177_end_mask_0, x = k_411_cast_fp16)[name = tensor<string, []>("op_46177_cast_fp16")];
            tensor<int32, [4]> var_46181_begin_0 = const()[name = tensor<string, []>("op_46181_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_46181_end_0 = const()[name = tensor<string, []>("op_46181_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_46181_end_mask_0 = const()[name = tensor<string, []>("op_46181_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_46181_cast_fp16 = slice_by_index(begin = var_46181_begin_0, end = var_46181_end_0, end_mask = var_46181_end_mask_0, x = k_411_cast_fp16)[name = tensor<string, []>("op_46181_cast_fp16")];
            tensor<int32, [4]> var_46185_begin_0 = const()[name = tensor<string, []>("op_46185_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_46185_end_0 = const()[name = tensor<string, []>("op_46185_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_46185_end_mask_0 = const()[name = tensor<string, []>("op_46185_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_46185_cast_fp16 = slice_by_index(begin = var_46185_begin_0, end = var_46185_end_0, end_mask = var_46185_end_mask_0, x = k_411_cast_fp16)[name = tensor<string, []>("op_46185_cast_fp16")];
            tensor<int32, [4]> var_46189_begin_0 = const()[name = tensor<string, []>("op_46189_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_46189_end_0 = const()[name = tensor<string, []>("op_46189_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_46189_end_mask_0 = const()[name = tensor<string, []>("op_46189_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_46189_cast_fp16 = slice_by_index(begin = var_46189_begin_0, end = var_46189_end_0, end_mask = var_46189_end_mask_0, x = k_411_cast_fp16)[name = tensor<string, []>("op_46189_cast_fp16")];
            tensor<int32, [4]> var_46193_begin_0 = const()[name = tensor<string, []>("op_46193_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_46193_end_0 = const()[name = tensor<string, []>("op_46193_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_46193_end_mask_0 = const()[name = tensor<string, []>("op_46193_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_46193_cast_fp16 = slice_by_index(begin = var_46193_begin_0, end = var_46193_end_0, end_mask = var_46193_end_mask_0, x = k_411_cast_fp16)[name = tensor<string, []>("op_46193_cast_fp16")];
            tensor<int32, [4]> var_46197_begin_0 = const()[name = tensor<string, []>("op_46197_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_46197_end_0 = const()[name = tensor<string, []>("op_46197_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_46197_end_mask_0 = const()[name = tensor<string, []>("op_46197_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_46197_cast_fp16 = slice_by_index(begin = var_46197_begin_0, end = var_46197_end_0, end_mask = var_46197_end_mask_0, x = k_411_cast_fp16)[name = tensor<string, []>("op_46197_cast_fp16")];
            tensor<int32, [4]> var_46201_begin_0 = const()[name = tensor<string, []>("op_46201_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_46201_end_0 = const()[name = tensor<string, []>("op_46201_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_46201_end_mask_0 = const()[name = tensor<string, []>("op_46201_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_46201_cast_fp16 = slice_by_index(begin = var_46201_begin_0, end = var_46201_end_0, end_mask = var_46201_end_mask_0, x = k_411_cast_fp16)[name = tensor<string, []>("op_46201_cast_fp16")];
            tensor<int32, [4]> var_46205_begin_0 = const()[name = tensor<string, []>("op_46205_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_46205_end_0 = const()[name = tensor<string, []>("op_46205_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_46205_end_mask_0 = const()[name = tensor<string, []>("op_46205_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_46205_cast_fp16 = slice_by_index(begin = var_46205_begin_0, end = var_46205_end_0, end_mask = var_46205_end_mask_0, x = k_411_cast_fp16)[name = tensor<string, []>("op_46205_cast_fp16")];
            tensor<int32, [4]> var_46209_begin_0 = const()[name = tensor<string, []>("op_46209_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_46209_end_0 = const()[name = tensor<string, []>("op_46209_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_46209_end_mask_0 = const()[name = tensor<string, []>("op_46209_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_46209_cast_fp16 = slice_by_index(begin = var_46209_begin_0, end = var_46209_end_0, end_mask = var_46209_end_mask_0, x = k_411_cast_fp16)[name = tensor<string, []>("op_46209_cast_fp16")];
            tensor<int32, [4]> var_46213_begin_0 = const()[name = tensor<string, []>("op_46213_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_46213_end_0 = const()[name = tensor<string, []>("op_46213_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_46213_end_mask_0 = const()[name = tensor<string, []>("op_46213_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_46213_cast_fp16 = slice_by_index(begin = var_46213_begin_0, end = var_46213_end_0, end_mask = var_46213_end_mask_0, x = k_411_cast_fp16)[name = tensor<string, []>("op_46213_cast_fp16")];
            tensor<int32, [4]> var_46217_begin_0 = const()[name = tensor<string, []>("op_46217_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_46217_end_0 = const()[name = tensor<string, []>("op_46217_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_46217_end_mask_0 = const()[name = tensor<string, []>("op_46217_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_46217_cast_fp16 = slice_by_index(begin = var_46217_begin_0, end = var_46217_end_0, end_mask = var_46217_end_mask_0, x = k_411_cast_fp16)[name = tensor<string, []>("op_46217_cast_fp16")];
            tensor<int32, [4]> var_46221_begin_0 = const()[name = tensor<string, []>("op_46221_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_46221_end_0 = const()[name = tensor<string, []>("op_46221_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_46221_end_mask_0 = const()[name = tensor<string, []>("op_46221_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_46221_cast_fp16 = slice_by_index(begin = var_46221_begin_0, end = var_46221_end_0, end_mask = var_46221_end_mask_0, x = k_411_cast_fp16)[name = tensor<string, []>("op_46221_cast_fp16")];
            tensor<int32, [4]> var_46225_begin_0 = const()[name = tensor<string, []>("op_46225_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_46225_end_0 = const()[name = tensor<string, []>("op_46225_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_46225_end_mask_0 = const()[name = tensor<string, []>("op_46225_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_46225_cast_fp16 = slice_by_index(begin = var_46225_begin_0, end = var_46225_end_0, end_mask = var_46225_end_mask_0, x = k_411_cast_fp16)[name = tensor<string, []>("op_46225_cast_fp16")];
            tensor<int32, [4]> var_46229_begin_0 = const()[name = tensor<string, []>("op_46229_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_46229_end_0 = const()[name = tensor<string, []>("op_46229_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_46229_end_mask_0 = const()[name = tensor<string, []>("op_46229_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_46229_cast_fp16 = slice_by_index(begin = var_46229_begin_0, end = var_46229_end_0, end_mask = var_46229_end_mask_0, x = k_411_cast_fp16)[name = tensor<string, []>("op_46229_cast_fp16")];
            tensor<int32, [4]> var_46233_begin_0 = const()[name = tensor<string, []>("op_46233_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_46233_end_0 = const()[name = tensor<string, []>("op_46233_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_46233_end_mask_0 = const()[name = tensor<string, []>("op_46233_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_46233_cast_fp16 = slice_by_index(begin = var_46233_begin_0, end = var_46233_end_0, end_mask = var_46233_end_mask_0, x = k_411_cast_fp16)[name = tensor<string, []>("op_46233_cast_fp16")];
            tensor<int32, [4]> var_46237_begin_0 = const()[name = tensor<string, []>("op_46237_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_46237_end_0 = const()[name = tensor<string, []>("op_46237_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_46237_end_mask_0 = const()[name = tensor<string, []>("op_46237_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_46237_cast_fp16 = slice_by_index(begin = var_46237_begin_0, end = var_46237_end_0, end_mask = var_46237_end_mask_0, x = k_411_cast_fp16)[name = tensor<string, []>("op_46237_cast_fp16")];
            tensor<int32, [4]> var_46241_begin_0 = const()[name = tensor<string, []>("op_46241_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_46241_end_0 = const()[name = tensor<string, []>("op_46241_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_46241_end_mask_0 = const()[name = tensor<string, []>("op_46241_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_46241_cast_fp16 = slice_by_index(begin = var_46241_begin_0, end = var_46241_end_0, end_mask = var_46241_end_mask_0, x = k_411_cast_fp16)[name = tensor<string, []>("op_46241_cast_fp16")];
            tensor<int32, [4]> var_46245_begin_0 = const()[name = tensor<string, []>("op_46245_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_46245_end_0 = const()[name = tensor<string, []>("op_46245_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_46245_end_mask_0 = const()[name = tensor<string, []>("op_46245_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_46245_cast_fp16 = slice_by_index(begin = var_46245_begin_0, end = var_46245_end_0, end_mask = var_46245_end_mask_0, x = k_411_cast_fp16)[name = tensor<string, []>("op_46245_cast_fp16")];
            tensor<int32, [4]> var_46249_begin_0 = const()[name = tensor<string, []>("op_46249_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_46249_end_0 = const()[name = tensor<string, []>("op_46249_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_46249_end_mask_0 = const()[name = tensor<string, []>("op_46249_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_46249_cast_fp16 = slice_by_index(begin = var_46249_begin_0, end = var_46249_end_0, end_mask = var_46249_end_mask_0, x = k_411_cast_fp16)[name = tensor<string, []>("op_46249_cast_fp16")];
            tensor<int32, [4]> var_46253_begin_0 = const()[name = tensor<string, []>("op_46253_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_46253_end_0 = const()[name = tensor<string, []>("op_46253_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_46253_end_mask_0 = const()[name = tensor<string, []>("op_46253_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_46253_cast_fp16 = slice_by_index(begin = var_46253_begin_0, end = var_46253_end_0, end_mask = var_46253_end_mask_0, x = k_411_cast_fp16)[name = tensor<string, []>("op_46253_cast_fp16")];
            tensor<int32, [4]> var_46255_begin_0 = const()[name = tensor<string, []>("op_46255_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_46255_end_0 = const()[name = tensor<string, []>("op_46255_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_46255_end_mask_0 = const()[name = tensor<string, []>("op_46255_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46255_cast_fp16 = slice_by_index(begin = var_46255_begin_0, end = var_46255_end_0, end_mask = var_46255_end_mask_0, x = v_205_cast_fp16)[name = tensor<string, []>("op_46255_cast_fp16")];
            tensor<int32, [4]> var_46259_begin_0 = const()[name = tensor<string, []>("op_46259_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_46259_end_0 = const()[name = tensor<string, []>("op_46259_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_46259_end_mask_0 = const()[name = tensor<string, []>("op_46259_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46259_cast_fp16 = slice_by_index(begin = var_46259_begin_0, end = var_46259_end_0, end_mask = var_46259_end_mask_0, x = v_205_cast_fp16)[name = tensor<string, []>("op_46259_cast_fp16")];
            tensor<int32, [4]> var_46263_begin_0 = const()[name = tensor<string, []>("op_46263_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_46263_end_0 = const()[name = tensor<string, []>("op_46263_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_46263_end_mask_0 = const()[name = tensor<string, []>("op_46263_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46263_cast_fp16 = slice_by_index(begin = var_46263_begin_0, end = var_46263_end_0, end_mask = var_46263_end_mask_0, x = v_205_cast_fp16)[name = tensor<string, []>("op_46263_cast_fp16")];
            tensor<int32, [4]> var_46267_begin_0 = const()[name = tensor<string, []>("op_46267_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_46267_end_0 = const()[name = tensor<string, []>("op_46267_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_46267_end_mask_0 = const()[name = tensor<string, []>("op_46267_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46267_cast_fp16 = slice_by_index(begin = var_46267_begin_0, end = var_46267_end_0, end_mask = var_46267_end_mask_0, x = v_205_cast_fp16)[name = tensor<string, []>("op_46267_cast_fp16")];
            tensor<int32, [4]> var_46271_begin_0 = const()[name = tensor<string, []>("op_46271_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_46271_end_0 = const()[name = tensor<string, []>("op_46271_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_46271_end_mask_0 = const()[name = tensor<string, []>("op_46271_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46271_cast_fp16 = slice_by_index(begin = var_46271_begin_0, end = var_46271_end_0, end_mask = var_46271_end_mask_0, x = v_205_cast_fp16)[name = tensor<string, []>("op_46271_cast_fp16")];
            tensor<int32, [4]> var_46275_begin_0 = const()[name = tensor<string, []>("op_46275_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_46275_end_0 = const()[name = tensor<string, []>("op_46275_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_46275_end_mask_0 = const()[name = tensor<string, []>("op_46275_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46275_cast_fp16 = slice_by_index(begin = var_46275_begin_0, end = var_46275_end_0, end_mask = var_46275_end_mask_0, x = v_205_cast_fp16)[name = tensor<string, []>("op_46275_cast_fp16")];
            tensor<int32, [4]> var_46279_begin_0 = const()[name = tensor<string, []>("op_46279_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_46279_end_0 = const()[name = tensor<string, []>("op_46279_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_46279_end_mask_0 = const()[name = tensor<string, []>("op_46279_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46279_cast_fp16 = slice_by_index(begin = var_46279_begin_0, end = var_46279_end_0, end_mask = var_46279_end_mask_0, x = v_205_cast_fp16)[name = tensor<string, []>("op_46279_cast_fp16")];
            tensor<int32, [4]> var_46283_begin_0 = const()[name = tensor<string, []>("op_46283_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_46283_end_0 = const()[name = tensor<string, []>("op_46283_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_46283_end_mask_0 = const()[name = tensor<string, []>("op_46283_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46283_cast_fp16 = slice_by_index(begin = var_46283_begin_0, end = var_46283_end_0, end_mask = var_46283_end_mask_0, x = v_205_cast_fp16)[name = tensor<string, []>("op_46283_cast_fp16")];
            tensor<int32, [4]> var_46287_begin_0 = const()[name = tensor<string, []>("op_46287_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_46287_end_0 = const()[name = tensor<string, []>("op_46287_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_46287_end_mask_0 = const()[name = tensor<string, []>("op_46287_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46287_cast_fp16 = slice_by_index(begin = var_46287_begin_0, end = var_46287_end_0, end_mask = var_46287_end_mask_0, x = v_205_cast_fp16)[name = tensor<string, []>("op_46287_cast_fp16")];
            tensor<int32, [4]> var_46291_begin_0 = const()[name = tensor<string, []>("op_46291_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_46291_end_0 = const()[name = tensor<string, []>("op_46291_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_46291_end_mask_0 = const()[name = tensor<string, []>("op_46291_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46291_cast_fp16 = slice_by_index(begin = var_46291_begin_0, end = var_46291_end_0, end_mask = var_46291_end_mask_0, x = v_205_cast_fp16)[name = tensor<string, []>("op_46291_cast_fp16")];
            tensor<int32, [4]> var_46295_begin_0 = const()[name = tensor<string, []>("op_46295_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_46295_end_0 = const()[name = tensor<string, []>("op_46295_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_46295_end_mask_0 = const()[name = tensor<string, []>("op_46295_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46295_cast_fp16 = slice_by_index(begin = var_46295_begin_0, end = var_46295_end_0, end_mask = var_46295_end_mask_0, x = v_205_cast_fp16)[name = tensor<string, []>("op_46295_cast_fp16")];
            tensor<int32, [4]> var_46299_begin_0 = const()[name = tensor<string, []>("op_46299_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_46299_end_0 = const()[name = tensor<string, []>("op_46299_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_46299_end_mask_0 = const()[name = tensor<string, []>("op_46299_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46299_cast_fp16 = slice_by_index(begin = var_46299_begin_0, end = var_46299_end_0, end_mask = var_46299_end_mask_0, x = v_205_cast_fp16)[name = tensor<string, []>("op_46299_cast_fp16")];
            tensor<int32, [4]> var_46303_begin_0 = const()[name = tensor<string, []>("op_46303_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_46303_end_0 = const()[name = tensor<string, []>("op_46303_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_46303_end_mask_0 = const()[name = tensor<string, []>("op_46303_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46303_cast_fp16 = slice_by_index(begin = var_46303_begin_0, end = var_46303_end_0, end_mask = var_46303_end_mask_0, x = v_205_cast_fp16)[name = tensor<string, []>("op_46303_cast_fp16")];
            tensor<int32, [4]> var_46307_begin_0 = const()[name = tensor<string, []>("op_46307_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_46307_end_0 = const()[name = tensor<string, []>("op_46307_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_46307_end_mask_0 = const()[name = tensor<string, []>("op_46307_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46307_cast_fp16 = slice_by_index(begin = var_46307_begin_0, end = var_46307_end_0, end_mask = var_46307_end_mask_0, x = v_205_cast_fp16)[name = tensor<string, []>("op_46307_cast_fp16")];
            tensor<int32, [4]> var_46311_begin_0 = const()[name = tensor<string, []>("op_46311_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_46311_end_0 = const()[name = tensor<string, []>("op_46311_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_46311_end_mask_0 = const()[name = tensor<string, []>("op_46311_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46311_cast_fp16 = slice_by_index(begin = var_46311_begin_0, end = var_46311_end_0, end_mask = var_46311_end_mask_0, x = v_205_cast_fp16)[name = tensor<string, []>("op_46311_cast_fp16")];
            tensor<int32, [4]> var_46315_begin_0 = const()[name = tensor<string, []>("op_46315_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_46315_end_0 = const()[name = tensor<string, []>("op_46315_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_46315_end_mask_0 = const()[name = tensor<string, []>("op_46315_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46315_cast_fp16 = slice_by_index(begin = var_46315_begin_0, end = var_46315_end_0, end_mask = var_46315_end_mask_0, x = v_205_cast_fp16)[name = tensor<string, []>("op_46315_cast_fp16")];
            tensor<int32, [4]> var_46319_begin_0 = const()[name = tensor<string, []>("op_46319_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_46319_end_0 = const()[name = tensor<string, []>("op_46319_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_46319_end_mask_0 = const()[name = tensor<string, []>("op_46319_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46319_cast_fp16 = slice_by_index(begin = var_46319_begin_0, end = var_46319_end_0, end_mask = var_46319_end_mask_0, x = v_205_cast_fp16)[name = tensor<string, []>("op_46319_cast_fp16")];
            tensor<int32, [4]> var_46323_begin_0 = const()[name = tensor<string, []>("op_46323_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_46323_end_0 = const()[name = tensor<string, []>("op_46323_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_46323_end_mask_0 = const()[name = tensor<string, []>("op_46323_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46323_cast_fp16 = slice_by_index(begin = var_46323_begin_0, end = var_46323_end_0, end_mask = var_46323_end_mask_0, x = v_205_cast_fp16)[name = tensor<string, []>("op_46323_cast_fp16")];
            tensor<int32, [4]> var_46327_begin_0 = const()[name = tensor<string, []>("op_46327_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_46327_end_0 = const()[name = tensor<string, []>("op_46327_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_46327_end_mask_0 = const()[name = tensor<string, []>("op_46327_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46327_cast_fp16 = slice_by_index(begin = var_46327_begin_0, end = var_46327_end_0, end_mask = var_46327_end_mask_0, x = v_205_cast_fp16)[name = tensor<string, []>("op_46327_cast_fp16")];
            tensor<int32, [4]> var_46331_begin_0 = const()[name = tensor<string, []>("op_46331_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_46331_end_0 = const()[name = tensor<string, []>("op_46331_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_46331_end_mask_0 = const()[name = tensor<string, []>("op_46331_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46331_cast_fp16 = slice_by_index(begin = var_46331_begin_0, end = var_46331_end_0, end_mask = var_46331_end_mask_0, x = v_205_cast_fp16)[name = tensor<string, []>("op_46331_cast_fp16")];
            tensor<string, []> var_46335_equation_0 = const()[name = tensor<string, []>("op_46335_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46335_cast_fp16 = einsum(equation = var_46335_equation_0, values = (var_46177_cast_fp16, var_46094_cast_fp16))[name = tensor<string, []>("op_46335_cast_fp16")];
            tensor<fp16, []> var_46336_to_fp16 = const()[name = tensor<string, []>("op_46336_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3921_cast_fp16 = mul(x = var_46335_cast_fp16, y = var_46336_to_fp16)[name = tensor<string, []>("aw_3921_cast_fp16")];
            tensor<string, []> var_46339_equation_0 = const()[name = tensor<string, []>("op_46339_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46339_cast_fp16 = einsum(equation = var_46339_equation_0, values = (var_46181_cast_fp16, var_46098_cast_fp16))[name = tensor<string, []>("op_46339_cast_fp16")];
            tensor<fp16, []> var_46340_to_fp16 = const()[name = tensor<string, []>("op_46340_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3923_cast_fp16 = mul(x = var_46339_cast_fp16, y = var_46340_to_fp16)[name = tensor<string, []>("aw_3923_cast_fp16")];
            tensor<string, []> var_46343_equation_0 = const()[name = tensor<string, []>("op_46343_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46343_cast_fp16 = einsum(equation = var_46343_equation_0, values = (var_46185_cast_fp16, var_46102_cast_fp16))[name = tensor<string, []>("op_46343_cast_fp16")];
            tensor<fp16, []> var_46344_to_fp16 = const()[name = tensor<string, []>("op_46344_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3925_cast_fp16 = mul(x = var_46343_cast_fp16, y = var_46344_to_fp16)[name = tensor<string, []>("aw_3925_cast_fp16")];
            tensor<string, []> var_46347_equation_0 = const()[name = tensor<string, []>("op_46347_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46347_cast_fp16 = einsum(equation = var_46347_equation_0, values = (var_46189_cast_fp16, var_46106_cast_fp16))[name = tensor<string, []>("op_46347_cast_fp16")];
            tensor<fp16, []> var_46348_to_fp16 = const()[name = tensor<string, []>("op_46348_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3927_cast_fp16 = mul(x = var_46347_cast_fp16, y = var_46348_to_fp16)[name = tensor<string, []>("aw_3927_cast_fp16")];
            tensor<string, []> var_46351_equation_0 = const()[name = tensor<string, []>("op_46351_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46351_cast_fp16 = einsum(equation = var_46351_equation_0, values = (var_46193_cast_fp16, var_46110_cast_fp16))[name = tensor<string, []>("op_46351_cast_fp16")];
            tensor<fp16, []> var_46352_to_fp16 = const()[name = tensor<string, []>("op_46352_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3929_cast_fp16 = mul(x = var_46351_cast_fp16, y = var_46352_to_fp16)[name = tensor<string, []>("aw_3929_cast_fp16")];
            tensor<string, []> var_46355_equation_0 = const()[name = tensor<string, []>("op_46355_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46355_cast_fp16 = einsum(equation = var_46355_equation_0, values = (var_46197_cast_fp16, var_46114_cast_fp16))[name = tensor<string, []>("op_46355_cast_fp16")];
            tensor<fp16, []> var_46356_to_fp16 = const()[name = tensor<string, []>("op_46356_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3931_cast_fp16 = mul(x = var_46355_cast_fp16, y = var_46356_to_fp16)[name = tensor<string, []>("aw_3931_cast_fp16")];
            tensor<string, []> var_46359_equation_0 = const()[name = tensor<string, []>("op_46359_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46359_cast_fp16 = einsum(equation = var_46359_equation_0, values = (var_46201_cast_fp16, var_46118_cast_fp16))[name = tensor<string, []>("op_46359_cast_fp16")];
            tensor<fp16, []> var_46360_to_fp16 = const()[name = tensor<string, []>("op_46360_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3933_cast_fp16 = mul(x = var_46359_cast_fp16, y = var_46360_to_fp16)[name = tensor<string, []>("aw_3933_cast_fp16")];
            tensor<string, []> var_46363_equation_0 = const()[name = tensor<string, []>("op_46363_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46363_cast_fp16 = einsum(equation = var_46363_equation_0, values = (var_46205_cast_fp16, var_46122_cast_fp16))[name = tensor<string, []>("op_46363_cast_fp16")];
            tensor<fp16, []> var_46364_to_fp16 = const()[name = tensor<string, []>("op_46364_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3935_cast_fp16 = mul(x = var_46363_cast_fp16, y = var_46364_to_fp16)[name = tensor<string, []>("aw_3935_cast_fp16")];
            tensor<string, []> var_46367_equation_0 = const()[name = tensor<string, []>("op_46367_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46367_cast_fp16 = einsum(equation = var_46367_equation_0, values = (var_46209_cast_fp16, var_46126_cast_fp16))[name = tensor<string, []>("op_46367_cast_fp16")];
            tensor<fp16, []> var_46368_to_fp16 = const()[name = tensor<string, []>("op_46368_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3937_cast_fp16 = mul(x = var_46367_cast_fp16, y = var_46368_to_fp16)[name = tensor<string, []>("aw_3937_cast_fp16")];
            tensor<string, []> var_46371_equation_0 = const()[name = tensor<string, []>("op_46371_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46371_cast_fp16 = einsum(equation = var_46371_equation_0, values = (var_46213_cast_fp16, var_46130_cast_fp16))[name = tensor<string, []>("op_46371_cast_fp16")];
            tensor<fp16, []> var_46372_to_fp16 = const()[name = tensor<string, []>("op_46372_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3939_cast_fp16 = mul(x = var_46371_cast_fp16, y = var_46372_to_fp16)[name = tensor<string, []>("aw_3939_cast_fp16")];
            tensor<string, []> var_46375_equation_0 = const()[name = tensor<string, []>("op_46375_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46375_cast_fp16 = einsum(equation = var_46375_equation_0, values = (var_46217_cast_fp16, var_46134_cast_fp16))[name = tensor<string, []>("op_46375_cast_fp16")];
            tensor<fp16, []> var_46376_to_fp16 = const()[name = tensor<string, []>("op_46376_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3941_cast_fp16 = mul(x = var_46375_cast_fp16, y = var_46376_to_fp16)[name = tensor<string, []>("aw_3941_cast_fp16")];
            tensor<string, []> var_46379_equation_0 = const()[name = tensor<string, []>("op_46379_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46379_cast_fp16 = einsum(equation = var_46379_equation_0, values = (var_46221_cast_fp16, var_46138_cast_fp16))[name = tensor<string, []>("op_46379_cast_fp16")];
            tensor<fp16, []> var_46380_to_fp16 = const()[name = tensor<string, []>("op_46380_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3943_cast_fp16 = mul(x = var_46379_cast_fp16, y = var_46380_to_fp16)[name = tensor<string, []>("aw_3943_cast_fp16")];
            tensor<string, []> var_46383_equation_0 = const()[name = tensor<string, []>("op_46383_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46383_cast_fp16 = einsum(equation = var_46383_equation_0, values = (var_46225_cast_fp16, var_46142_cast_fp16))[name = tensor<string, []>("op_46383_cast_fp16")];
            tensor<fp16, []> var_46384_to_fp16 = const()[name = tensor<string, []>("op_46384_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3945_cast_fp16 = mul(x = var_46383_cast_fp16, y = var_46384_to_fp16)[name = tensor<string, []>("aw_3945_cast_fp16")];
            tensor<string, []> var_46387_equation_0 = const()[name = tensor<string, []>("op_46387_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46387_cast_fp16 = einsum(equation = var_46387_equation_0, values = (var_46229_cast_fp16, var_46146_cast_fp16))[name = tensor<string, []>("op_46387_cast_fp16")];
            tensor<fp16, []> var_46388_to_fp16 = const()[name = tensor<string, []>("op_46388_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3947_cast_fp16 = mul(x = var_46387_cast_fp16, y = var_46388_to_fp16)[name = tensor<string, []>("aw_3947_cast_fp16")];
            tensor<string, []> var_46391_equation_0 = const()[name = tensor<string, []>("op_46391_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46391_cast_fp16 = einsum(equation = var_46391_equation_0, values = (var_46233_cast_fp16, var_46150_cast_fp16))[name = tensor<string, []>("op_46391_cast_fp16")];
            tensor<fp16, []> var_46392_to_fp16 = const()[name = tensor<string, []>("op_46392_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3949_cast_fp16 = mul(x = var_46391_cast_fp16, y = var_46392_to_fp16)[name = tensor<string, []>("aw_3949_cast_fp16")];
            tensor<string, []> var_46395_equation_0 = const()[name = tensor<string, []>("op_46395_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46395_cast_fp16 = einsum(equation = var_46395_equation_0, values = (var_46237_cast_fp16, var_46154_cast_fp16))[name = tensor<string, []>("op_46395_cast_fp16")];
            tensor<fp16, []> var_46396_to_fp16 = const()[name = tensor<string, []>("op_46396_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3951_cast_fp16 = mul(x = var_46395_cast_fp16, y = var_46396_to_fp16)[name = tensor<string, []>("aw_3951_cast_fp16")];
            tensor<string, []> var_46399_equation_0 = const()[name = tensor<string, []>("op_46399_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46399_cast_fp16 = einsum(equation = var_46399_equation_0, values = (var_46241_cast_fp16, var_46158_cast_fp16))[name = tensor<string, []>("op_46399_cast_fp16")];
            tensor<fp16, []> var_46400_to_fp16 = const()[name = tensor<string, []>("op_46400_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3953_cast_fp16 = mul(x = var_46399_cast_fp16, y = var_46400_to_fp16)[name = tensor<string, []>("aw_3953_cast_fp16")];
            tensor<string, []> var_46403_equation_0 = const()[name = tensor<string, []>("op_46403_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46403_cast_fp16 = einsum(equation = var_46403_equation_0, values = (var_46245_cast_fp16, var_46162_cast_fp16))[name = tensor<string, []>("op_46403_cast_fp16")];
            tensor<fp16, []> var_46404_to_fp16 = const()[name = tensor<string, []>("op_46404_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3955_cast_fp16 = mul(x = var_46403_cast_fp16, y = var_46404_to_fp16)[name = tensor<string, []>("aw_3955_cast_fp16")];
            tensor<string, []> var_46407_equation_0 = const()[name = tensor<string, []>("op_46407_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46407_cast_fp16 = einsum(equation = var_46407_equation_0, values = (var_46249_cast_fp16, var_46166_cast_fp16))[name = tensor<string, []>("op_46407_cast_fp16")];
            tensor<fp16, []> var_46408_to_fp16 = const()[name = tensor<string, []>("op_46408_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3957_cast_fp16 = mul(x = var_46407_cast_fp16, y = var_46408_to_fp16)[name = tensor<string, []>("aw_3957_cast_fp16")];
            tensor<string, []> var_46411_equation_0 = const()[name = tensor<string, []>("op_46411_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46411_cast_fp16 = einsum(equation = var_46411_equation_0, values = (var_46253_cast_fp16, var_46170_cast_fp16))[name = tensor<string, []>("op_46411_cast_fp16")];
            tensor<fp16, []> var_46412_to_fp16 = const()[name = tensor<string, []>("op_46412_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3959_cast_fp16 = mul(x = var_46411_cast_fp16, y = var_46412_to_fp16)[name = tensor<string, []>("aw_3959_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46414_cast_fp16 = softmax(axis = var_30355, x = aw_3921_cast_fp16)[name = tensor<string, []>("op_46414_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46415_cast_fp16 = softmax(axis = var_30355, x = aw_3923_cast_fp16)[name = tensor<string, []>("op_46415_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46416_cast_fp16 = softmax(axis = var_30355, x = aw_3925_cast_fp16)[name = tensor<string, []>("op_46416_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46417_cast_fp16 = softmax(axis = var_30355, x = aw_3927_cast_fp16)[name = tensor<string, []>("op_46417_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46418_cast_fp16 = softmax(axis = var_30355, x = aw_3929_cast_fp16)[name = tensor<string, []>("op_46418_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46419_cast_fp16 = softmax(axis = var_30355, x = aw_3931_cast_fp16)[name = tensor<string, []>("op_46419_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46420_cast_fp16 = softmax(axis = var_30355, x = aw_3933_cast_fp16)[name = tensor<string, []>("op_46420_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46421_cast_fp16 = softmax(axis = var_30355, x = aw_3935_cast_fp16)[name = tensor<string, []>("op_46421_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46422_cast_fp16 = softmax(axis = var_30355, x = aw_3937_cast_fp16)[name = tensor<string, []>("op_46422_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46423_cast_fp16 = softmax(axis = var_30355, x = aw_3939_cast_fp16)[name = tensor<string, []>("op_46423_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46424_cast_fp16 = softmax(axis = var_30355, x = aw_3941_cast_fp16)[name = tensor<string, []>("op_46424_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46425_cast_fp16 = softmax(axis = var_30355, x = aw_3943_cast_fp16)[name = tensor<string, []>("op_46425_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46426_cast_fp16 = softmax(axis = var_30355, x = aw_3945_cast_fp16)[name = tensor<string, []>("op_46426_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46427_cast_fp16 = softmax(axis = var_30355, x = aw_3947_cast_fp16)[name = tensor<string, []>("op_46427_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46428_cast_fp16 = softmax(axis = var_30355, x = aw_3949_cast_fp16)[name = tensor<string, []>("op_46428_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46429_cast_fp16 = softmax(axis = var_30355, x = aw_3951_cast_fp16)[name = tensor<string, []>("op_46429_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46430_cast_fp16 = softmax(axis = var_30355, x = aw_3953_cast_fp16)[name = tensor<string, []>("op_46430_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46431_cast_fp16 = softmax(axis = var_30355, x = aw_3955_cast_fp16)[name = tensor<string, []>("op_46431_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46432_cast_fp16 = softmax(axis = var_30355, x = aw_3957_cast_fp16)[name = tensor<string, []>("op_46432_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46433_cast_fp16 = softmax(axis = var_30355, x = aw_3959_cast_fp16)[name = tensor<string, []>("op_46433_cast_fp16")];
            tensor<string, []> var_46435_equation_0 = const()[name = tensor<string, []>("op_46435_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46435_cast_fp16 = einsum(equation = var_46435_equation_0, values = (var_46255_cast_fp16, var_46414_cast_fp16))[name = tensor<string, []>("op_46435_cast_fp16")];
            tensor<string, []> var_46437_equation_0 = const()[name = tensor<string, []>("op_46437_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46437_cast_fp16 = einsum(equation = var_46437_equation_0, values = (var_46259_cast_fp16, var_46415_cast_fp16))[name = tensor<string, []>("op_46437_cast_fp16")];
            tensor<string, []> var_46439_equation_0 = const()[name = tensor<string, []>("op_46439_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46439_cast_fp16 = einsum(equation = var_46439_equation_0, values = (var_46263_cast_fp16, var_46416_cast_fp16))[name = tensor<string, []>("op_46439_cast_fp16")];
            tensor<string, []> var_46441_equation_0 = const()[name = tensor<string, []>("op_46441_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46441_cast_fp16 = einsum(equation = var_46441_equation_0, values = (var_46267_cast_fp16, var_46417_cast_fp16))[name = tensor<string, []>("op_46441_cast_fp16")];
            tensor<string, []> var_46443_equation_0 = const()[name = tensor<string, []>("op_46443_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46443_cast_fp16 = einsum(equation = var_46443_equation_0, values = (var_46271_cast_fp16, var_46418_cast_fp16))[name = tensor<string, []>("op_46443_cast_fp16")];
            tensor<string, []> var_46445_equation_0 = const()[name = tensor<string, []>("op_46445_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46445_cast_fp16 = einsum(equation = var_46445_equation_0, values = (var_46275_cast_fp16, var_46419_cast_fp16))[name = tensor<string, []>("op_46445_cast_fp16")];
            tensor<string, []> var_46447_equation_0 = const()[name = tensor<string, []>("op_46447_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46447_cast_fp16 = einsum(equation = var_46447_equation_0, values = (var_46279_cast_fp16, var_46420_cast_fp16))[name = tensor<string, []>("op_46447_cast_fp16")];
            tensor<string, []> var_46449_equation_0 = const()[name = tensor<string, []>("op_46449_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46449_cast_fp16 = einsum(equation = var_46449_equation_0, values = (var_46283_cast_fp16, var_46421_cast_fp16))[name = tensor<string, []>("op_46449_cast_fp16")];
            tensor<string, []> var_46451_equation_0 = const()[name = tensor<string, []>("op_46451_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46451_cast_fp16 = einsum(equation = var_46451_equation_0, values = (var_46287_cast_fp16, var_46422_cast_fp16))[name = tensor<string, []>("op_46451_cast_fp16")];
            tensor<string, []> var_46453_equation_0 = const()[name = tensor<string, []>("op_46453_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46453_cast_fp16 = einsum(equation = var_46453_equation_0, values = (var_46291_cast_fp16, var_46423_cast_fp16))[name = tensor<string, []>("op_46453_cast_fp16")];
            tensor<string, []> var_46455_equation_0 = const()[name = tensor<string, []>("op_46455_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46455_cast_fp16 = einsum(equation = var_46455_equation_0, values = (var_46295_cast_fp16, var_46424_cast_fp16))[name = tensor<string, []>("op_46455_cast_fp16")];
            tensor<string, []> var_46457_equation_0 = const()[name = tensor<string, []>("op_46457_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46457_cast_fp16 = einsum(equation = var_46457_equation_0, values = (var_46299_cast_fp16, var_46425_cast_fp16))[name = tensor<string, []>("op_46457_cast_fp16")];
            tensor<string, []> var_46459_equation_0 = const()[name = tensor<string, []>("op_46459_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46459_cast_fp16 = einsum(equation = var_46459_equation_0, values = (var_46303_cast_fp16, var_46426_cast_fp16))[name = tensor<string, []>("op_46459_cast_fp16")];
            tensor<string, []> var_46461_equation_0 = const()[name = tensor<string, []>("op_46461_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46461_cast_fp16 = einsum(equation = var_46461_equation_0, values = (var_46307_cast_fp16, var_46427_cast_fp16))[name = tensor<string, []>("op_46461_cast_fp16")];
            tensor<string, []> var_46463_equation_0 = const()[name = tensor<string, []>("op_46463_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46463_cast_fp16 = einsum(equation = var_46463_equation_0, values = (var_46311_cast_fp16, var_46428_cast_fp16))[name = tensor<string, []>("op_46463_cast_fp16")];
            tensor<string, []> var_46465_equation_0 = const()[name = tensor<string, []>("op_46465_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46465_cast_fp16 = einsum(equation = var_46465_equation_0, values = (var_46315_cast_fp16, var_46429_cast_fp16))[name = tensor<string, []>("op_46465_cast_fp16")];
            tensor<string, []> var_46467_equation_0 = const()[name = tensor<string, []>("op_46467_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46467_cast_fp16 = einsum(equation = var_46467_equation_0, values = (var_46319_cast_fp16, var_46430_cast_fp16))[name = tensor<string, []>("op_46467_cast_fp16")];
            tensor<string, []> var_46469_equation_0 = const()[name = tensor<string, []>("op_46469_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46469_cast_fp16 = einsum(equation = var_46469_equation_0, values = (var_46323_cast_fp16, var_46431_cast_fp16))[name = tensor<string, []>("op_46469_cast_fp16")];
            tensor<string, []> var_46471_equation_0 = const()[name = tensor<string, []>("op_46471_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46471_cast_fp16 = einsum(equation = var_46471_equation_0, values = (var_46327_cast_fp16, var_46432_cast_fp16))[name = tensor<string, []>("op_46471_cast_fp16")];
            tensor<string, []> var_46473_equation_0 = const()[name = tensor<string, []>("op_46473_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46473_cast_fp16 = einsum(equation = var_46473_equation_0, values = (var_46331_cast_fp16, var_46433_cast_fp16))[name = tensor<string, []>("op_46473_cast_fp16")];
            tensor<bool, []> input_593_interleave_0 = const()[name = tensor<string, []>("input_593_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_593_cast_fp16 = concat(axis = var_30355, interleave = input_593_interleave_0, values = (var_46435_cast_fp16, var_46437_cast_fp16, var_46439_cast_fp16, var_46441_cast_fp16, var_46443_cast_fp16, var_46445_cast_fp16, var_46447_cast_fp16, var_46449_cast_fp16, var_46451_cast_fp16, var_46453_cast_fp16, var_46455_cast_fp16, var_46457_cast_fp16, var_46459_cast_fp16, var_46461_cast_fp16, var_46463_cast_fp16, var_46465_cast_fp16, var_46467_cast_fp16, var_46469_cast_fp16, var_46471_cast_fp16, var_46473_cast_fp16))[name = tensor<string, []>("input_593_cast_fp16")];
            tensor<string, []> var_46483_pad_type_0 = const()[name = tensor<string, []>("op_46483_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_46483_strides_0 = const()[name = tensor<string, []>("op_46483_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_46483_pad_0 = const()[name = tensor<string, []>("op_46483_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_46483_dilations_0 = const()[name = tensor<string, []>("op_46483_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_46483_groups_0 = const()[name = tensor<string, []>("op_46483_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_7_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(490423104))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(491651968))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_7_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_7_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_7_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(491652160)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_46483_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_7_attn1_to_out_0_bias_to_fp16, dilations = var_46483_dilations_0, groups = var_46483_groups_0, pad = var_46483_pad_0, pad_type = var_46483_pad_type_0, strides = var_46483_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_7_attn1_to_out_0_weight_to_fp16_palettized, x = input_593_cast_fp16)[name = tensor<string, []>("op_46483_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_309_cast_fp16 = add(x = var_46483_cast_fp16, y = inputs_307_cast_fp16)[name = tensor<string, []>("inputs_309_cast_fp16")];
            tensor<int32, [1]> hidden_states_405_axes_0 = const()[name = tensor<string, []>("hidden_states_405_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_405_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_405_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(491654784)))];
            tensor<fp16, [1280]> hidden_states_405_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_405_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(491657408)))];
            tensor<fp16, []> var_46493_to_fp16 = const()[name = tensor<string, []>("op_46493_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_405_cast_fp16 = layer_norm(axes = hidden_states_405_axes_0, beta = hidden_states_405_beta_0_to_fp16, epsilon = var_46493_to_fp16, gamma = hidden_states_405_gamma_0_to_fp16, x = inputs_309_cast_fp16)[name = tensor<string, []>("hidden_states_405_cast_fp16")];
            tensor<string, []> q_207_pad_type_0 = const()[name = tensor<string, []>("q_207_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_207_strides_0 = const()[name = tensor<string, []>("q_207_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_207_pad_0 = const()[name = tensor<string, []>("q_207_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_207_dilations_0 = const()[name = tensor<string, []>("q_207_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_207_groups_0 = const()[name = tensor<string, []>("q_207_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_7_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(491660032))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(492888896))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_7_attn2_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_207_cast_fp16 = conv(dilations = q_207_dilations_0, groups = q_207_groups_0, pad = q_207_pad_0, pad_type = q_207_pad_type_0, strides = q_207_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_7_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_405_cast_fp16)[name = tensor<string, []>("q_207_cast_fp16")];
            tensor<string, []> k_413_pad_type_0 = const()[name = tensor<string, []>("k_413_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_413_strides_0 = const()[name = tensor<string, []>("k_413_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_413_pad_0 = const()[name = tensor<string, []>("k_413_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_413_dilations_0 = const()[name = tensor<string, []>("k_413_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_413_groups_0 = const()[name = tensor<string, []>("k_413_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_7_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(492889088))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(494855232))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_7_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> k_413_cast_fp16 = conv(dilations = k_413_dilations_0, groups = k_413_groups_0, pad = k_413_pad_0, pad_type = k_413_pad_type_0, strides = k_413_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_7_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("k_413_cast_fp16")];
            tensor<string, []> v_207_pad_type_0 = const()[name = tensor<string, []>("v_207_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_207_strides_0 = const()[name = tensor<string, []>("v_207_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_207_pad_0 = const()[name = tensor<string, []>("v_207_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_207_dilations_0 = const()[name = tensor<string, []>("v_207_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_207_groups_0 = const()[name = tensor<string, []>("v_207_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_7_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(494855424))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(496821568))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_7_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> v_207_cast_fp16 = conv(dilations = v_207_dilations_0, groups = v_207_groups_0, pad = v_207_pad_0, pad_type = v_207_pad_type_0, strides = v_207_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_7_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("v_207_cast_fp16")];
            tensor<int32, [4]> var_46526_begin_0 = const()[name = tensor<string, []>("op_46526_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_46526_end_0 = const()[name = tensor<string, []>("op_46526_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_46526_end_mask_0 = const()[name = tensor<string, []>("op_46526_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46526_cast_fp16 = slice_by_index(begin = var_46526_begin_0, end = var_46526_end_0, end_mask = var_46526_end_mask_0, x = q_207_cast_fp16)[name = tensor<string, []>("op_46526_cast_fp16")];
            tensor<int32, [4]> var_46530_begin_0 = const()[name = tensor<string, []>("op_46530_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_46530_end_0 = const()[name = tensor<string, []>("op_46530_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_46530_end_mask_0 = const()[name = tensor<string, []>("op_46530_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46530_cast_fp16 = slice_by_index(begin = var_46530_begin_0, end = var_46530_end_0, end_mask = var_46530_end_mask_0, x = q_207_cast_fp16)[name = tensor<string, []>("op_46530_cast_fp16")];
            tensor<int32, [4]> var_46534_begin_0 = const()[name = tensor<string, []>("op_46534_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_46534_end_0 = const()[name = tensor<string, []>("op_46534_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_46534_end_mask_0 = const()[name = tensor<string, []>("op_46534_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46534_cast_fp16 = slice_by_index(begin = var_46534_begin_0, end = var_46534_end_0, end_mask = var_46534_end_mask_0, x = q_207_cast_fp16)[name = tensor<string, []>("op_46534_cast_fp16")];
            tensor<int32, [4]> var_46538_begin_0 = const()[name = tensor<string, []>("op_46538_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_46538_end_0 = const()[name = tensor<string, []>("op_46538_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_46538_end_mask_0 = const()[name = tensor<string, []>("op_46538_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46538_cast_fp16 = slice_by_index(begin = var_46538_begin_0, end = var_46538_end_0, end_mask = var_46538_end_mask_0, x = q_207_cast_fp16)[name = tensor<string, []>("op_46538_cast_fp16")];
            tensor<int32, [4]> var_46542_begin_0 = const()[name = tensor<string, []>("op_46542_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_46542_end_0 = const()[name = tensor<string, []>("op_46542_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_46542_end_mask_0 = const()[name = tensor<string, []>("op_46542_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46542_cast_fp16 = slice_by_index(begin = var_46542_begin_0, end = var_46542_end_0, end_mask = var_46542_end_mask_0, x = q_207_cast_fp16)[name = tensor<string, []>("op_46542_cast_fp16")];
            tensor<int32, [4]> var_46546_begin_0 = const()[name = tensor<string, []>("op_46546_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_46546_end_0 = const()[name = tensor<string, []>("op_46546_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_46546_end_mask_0 = const()[name = tensor<string, []>("op_46546_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46546_cast_fp16 = slice_by_index(begin = var_46546_begin_0, end = var_46546_end_0, end_mask = var_46546_end_mask_0, x = q_207_cast_fp16)[name = tensor<string, []>("op_46546_cast_fp16")];
            tensor<int32, [4]> var_46550_begin_0 = const()[name = tensor<string, []>("op_46550_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_46550_end_0 = const()[name = tensor<string, []>("op_46550_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_46550_end_mask_0 = const()[name = tensor<string, []>("op_46550_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46550_cast_fp16 = slice_by_index(begin = var_46550_begin_0, end = var_46550_end_0, end_mask = var_46550_end_mask_0, x = q_207_cast_fp16)[name = tensor<string, []>("op_46550_cast_fp16")];
            tensor<int32, [4]> var_46554_begin_0 = const()[name = tensor<string, []>("op_46554_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_46554_end_0 = const()[name = tensor<string, []>("op_46554_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_46554_end_mask_0 = const()[name = tensor<string, []>("op_46554_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46554_cast_fp16 = slice_by_index(begin = var_46554_begin_0, end = var_46554_end_0, end_mask = var_46554_end_mask_0, x = q_207_cast_fp16)[name = tensor<string, []>("op_46554_cast_fp16")];
            tensor<int32, [4]> var_46558_begin_0 = const()[name = tensor<string, []>("op_46558_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_46558_end_0 = const()[name = tensor<string, []>("op_46558_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_46558_end_mask_0 = const()[name = tensor<string, []>("op_46558_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46558_cast_fp16 = slice_by_index(begin = var_46558_begin_0, end = var_46558_end_0, end_mask = var_46558_end_mask_0, x = q_207_cast_fp16)[name = tensor<string, []>("op_46558_cast_fp16")];
            tensor<int32, [4]> var_46562_begin_0 = const()[name = tensor<string, []>("op_46562_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_46562_end_0 = const()[name = tensor<string, []>("op_46562_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_46562_end_mask_0 = const()[name = tensor<string, []>("op_46562_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46562_cast_fp16 = slice_by_index(begin = var_46562_begin_0, end = var_46562_end_0, end_mask = var_46562_end_mask_0, x = q_207_cast_fp16)[name = tensor<string, []>("op_46562_cast_fp16")];
            tensor<int32, [4]> var_46566_begin_0 = const()[name = tensor<string, []>("op_46566_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_46566_end_0 = const()[name = tensor<string, []>("op_46566_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_46566_end_mask_0 = const()[name = tensor<string, []>("op_46566_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46566_cast_fp16 = slice_by_index(begin = var_46566_begin_0, end = var_46566_end_0, end_mask = var_46566_end_mask_0, x = q_207_cast_fp16)[name = tensor<string, []>("op_46566_cast_fp16")];
            tensor<int32, [4]> var_46570_begin_0 = const()[name = tensor<string, []>("op_46570_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_46570_end_0 = const()[name = tensor<string, []>("op_46570_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_46570_end_mask_0 = const()[name = tensor<string, []>("op_46570_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46570_cast_fp16 = slice_by_index(begin = var_46570_begin_0, end = var_46570_end_0, end_mask = var_46570_end_mask_0, x = q_207_cast_fp16)[name = tensor<string, []>("op_46570_cast_fp16")];
            tensor<int32, [4]> var_46574_begin_0 = const()[name = tensor<string, []>("op_46574_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_46574_end_0 = const()[name = tensor<string, []>("op_46574_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_46574_end_mask_0 = const()[name = tensor<string, []>("op_46574_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46574_cast_fp16 = slice_by_index(begin = var_46574_begin_0, end = var_46574_end_0, end_mask = var_46574_end_mask_0, x = q_207_cast_fp16)[name = tensor<string, []>("op_46574_cast_fp16")];
            tensor<int32, [4]> var_46578_begin_0 = const()[name = tensor<string, []>("op_46578_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_46578_end_0 = const()[name = tensor<string, []>("op_46578_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_46578_end_mask_0 = const()[name = tensor<string, []>("op_46578_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46578_cast_fp16 = slice_by_index(begin = var_46578_begin_0, end = var_46578_end_0, end_mask = var_46578_end_mask_0, x = q_207_cast_fp16)[name = tensor<string, []>("op_46578_cast_fp16")];
            tensor<int32, [4]> var_46582_begin_0 = const()[name = tensor<string, []>("op_46582_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_46582_end_0 = const()[name = tensor<string, []>("op_46582_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_46582_end_mask_0 = const()[name = tensor<string, []>("op_46582_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46582_cast_fp16 = slice_by_index(begin = var_46582_begin_0, end = var_46582_end_0, end_mask = var_46582_end_mask_0, x = q_207_cast_fp16)[name = tensor<string, []>("op_46582_cast_fp16")];
            tensor<int32, [4]> var_46586_begin_0 = const()[name = tensor<string, []>("op_46586_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_46586_end_0 = const()[name = tensor<string, []>("op_46586_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_46586_end_mask_0 = const()[name = tensor<string, []>("op_46586_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46586_cast_fp16 = slice_by_index(begin = var_46586_begin_0, end = var_46586_end_0, end_mask = var_46586_end_mask_0, x = q_207_cast_fp16)[name = tensor<string, []>("op_46586_cast_fp16")];
            tensor<int32, [4]> var_46590_begin_0 = const()[name = tensor<string, []>("op_46590_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_46590_end_0 = const()[name = tensor<string, []>("op_46590_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_46590_end_mask_0 = const()[name = tensor<string, []>("op_46590_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46590_cast_fp16 = slice_by_index(begin = var_46590_begin_0, end = var_46590_end_0, end_mask = var_46590_end_mask_0, x = q_207_cast_fp16)[name = tensor<string, []>("op_46590_cast_fp16")];
            tensor<int32, [4]> var_46594_begin_0 = const()[name = tensor<string, []>("op_46594_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_46594_end_0 = const()[name = tensor<string, []>("op_46594_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_46594_end_mask_0 = const()[name = tensor<string, []>("op_46594_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46594_cast_fp16 = slice_by_index(begin = var_46594_begin_0, end = var_46594_end_0, end_mask = var_46594_end_mask_0, x = q_207_cast_fp16)[name = tensor<string, []>("op_46594_cast_fp16")];
            tensor<int32, [4]> var_46598_begin_0 = const()[name = tensor<string, []>("op_46598_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_46598_end_0 = const()[name = tensor<string, []>("op_46598_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_46598_end_mask_0 = const()[name = tensor<string, []>("op_46598_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46598_cast_fp16 = slice_by_index(begin = var_46598_begin_0, end = var_46598_end_0, end_mask = var_46598_end_mask_0, x = q_207_cast_fp16)[name = tensor<string, []>("op_46598_cast_fp16")];
            tensor<int32, [4]> var_46602_begin_0 = const()[name = tensor<string, []>("op_46602_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_46602_end_0 = const()[name = tensor<string, []>("op_46602_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_46602_end_mask_0 = const()[name = tensor<string, []>("op_46602_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46602_cast_fp16 = slice_by_index(begin = var_46602_begin_0, end = var_46602_end_0, end_mask = var_46602_end_mask_0, x = q_207_cast_fp16)[name = tensor<string, []>("op_46602_cast_fp16")];
            tensor<int32, [4]> k_415_perm_0 = const()[name = tensor<string, []>("k_415_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_46609_begin_0 = const()[name = tensor<string, []>("op_46609_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_46609_end_0 = const()[name = tensor<string, []>("op_46609_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_46609_end_mask_0 = const()[name = tensor<string, []>("op_46609_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> k_415_cast_fp16 = transpose(perm = k_415_perm_0, x = k_413_cast_fp16)[name = tensor<string, []>("transpose_36")];
            tensor<fp16, [2, 77, 1, 64]> var_46609_cast_fp16 = slice_by_index(begin = var_46609_begin_0, end = var_46609_end_0, end_mask = var_46609_end_mask_0, x = k_415_cast_fp16)[name = tensor<string, []>("op_46609_cast_fp16")];
            tensor<int32, [4]> var_46613_begin_0 = const()[name = tensor<string, []>("op_46613_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_46613_end_0 = const()[name = tensor<string, []>("op_46613_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_46613_end_mask_0 = const()[name = tensor<string, []>("op_46613_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_46613_cast_fp16 = slice_by_index(begin = var_46613_begin_0, end = var_46613_end_0, end_mask = var_46613_end_mask_0, x = k_415_cast_fp16)[name = tensor<string, []>("op_46613_cast_fp16")];
            tensor<int32, [4]> var_46617_begin_0 = const()[name = tensor<string, []>("op_46617_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_46617_end_0 = const()[name = tensor<string, []>("op_46617_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_46617_end_mask_0 = const()[name = tensor<string, []>("op_46617_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_46617_cast_fp16 = slice_by_index(begin = var_46617_begin_0, end = var_46617_end_0, end_mask = var_46617_end_mask_0, x = k_415_cast_fp16)[name = tensor<string, []>("op_46617_cast_fp16")];
            tensor<int32, [4]> var_46621_begin_0 = const()[name = tensor<string, []>("op_46621_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_46621_end_0 = const()[name = tensor<string, []>("op_46621_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_46621_end_mask_0 = const()[name = tensor<string, []>("op_46621_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_46621_cast_fp16 = slice_by_index(begin = var_46621_begin_0, end = var_46621_end_0, end_mask = var_46621_end_mask_0, x = k_415_cast_fp16)[name = tensor<string, []>("op_46621_cast_fp16")];
            tensor<int32, [4]> var_46625_begin_0 = const()[name = tensor<string, []>("op_46625_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_46625_end_0 = const()[name = tensor<string, []>("op_46625_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_46625_end_mask_0 = const()[name = tensor<string, []>("op_46625_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_46625_cast_fp16 = slice_by_index(begin = var_46625_begin_0, end = var_46625_end_0, end_mask = var_46625_end_mask_0, x = k_415_cast_fp16)[name = tensor<string, []>("op_46625_cast_fp16")];
            tensor<int32, [4]> var_46629_begin_0 = const()[name = tensor<string, []>("op_46629_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_46629_end_0 = const()[name = tensor<string, []>("op_46629_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_46629_end_mask_0 = const()[name = tensor<string, []>("op_46629_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_46629_cast_fp16 = slice_by_index(begin = var_46629_begin_0, end = var_46629_end_0, end_mask = var_46629_end_mask_0, x = k_415_cast_fp16)[name = tensor<string, []>("op_46629_cast_fp16")];
            tensor<int32, [4]> var_46633_begin_0 = const()[name = tensor<string, []>("op_46633_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_46633_end_0 = const()[name = tensor<string, []>("op_46633_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_46633_end_mask_0 = const()[name = tensor<string, []>("op_46633_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_46633_cast_fp16 = slice_by_index(begin = var_46633_begin_0, end = var_46633_end_0, end_mask = var_46633_end_mask_0, x = k_415_cast_fp16)[name = tensor<string, []>("op_46633_cast_fp16")];
            tensor<int32, [4]> var_46637_begin_0 = const()[name = tensor<string, []>("op_46637_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_46637_end_0 = const()[name = tensor<string, []>("op_46637_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_46637_end_mask_0 = const()[name = tensor<string, []>("op_46637_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_46637_cast_fp16 = slice_by_index(begin = var_46637_begin_0, end = var_46637_end_0, end_mask = var_46637_end_mask_0, x = k_415_cast_fp16)[name = tensor<string, []>("op_46637_cast_fp16")];
            tensor<int32, [4]> var_46641_begin_0 = const()[name = tensor<string, []>("op_46641_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_46641_end_0 = const()[name = tensor<string, []>("op_46641_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_46641_end_mask_0 = const()[name = tensor<string, []>("op_46641_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_46641_cast_fp16 = slice_by_index(begin = var_46641_begin_0, end = var_46641_end_0, end_mask = var_46641_end_mask_0, x = k_415_cast_fp16)[name = tensor<string, []>("op_46641_cast_fp16")];
            tensor<int32, [4]> var_46645_begin_0 = const()[name = tensor<string, []>("op_46645_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_46645_end_0 = const()[name = tensor<string, []>("op_46645_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_46645_end_mask_0 = const()[name = tensor<string, []>("op_46645_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_46645_cast_fp16 = slice_by_index(begin = var_46645_begin_0, end = var_46645_end_0, end_mask = var_46645_end_mask_0, x = k_415_cast_fp16)[name = tensor<string, []>("op_46645_cast_fp16")];
            tensor<int32, [4]> var_46649_begin_0 = const()[name = tensor<string, []>("op_46649_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_46649_end_0 = const()[name = tensor<string, []>("op_46649_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_46649_end_mask_0 = const()[name = tensor<string, []>("op_46649_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_46649_cast_fp16 = slice_by_index(begin = var_46649_begin_0, end = var_46649_end_0, end_mask = var_46649_end_mask_0, x = k_415_cast_fp16)[name = tensor<string, []>("op_46649_cast_fp16")];
            tensor<int32, [4]> var_46653_begin_0 = const()[name = tensor<string, []>("op_46653_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_46653_end_0 = const()[name = tensor<string, []>("op_46653_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_46653_end_mask_0 = const()[name = tensor<string, []>("op_46653_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_46653_cast_fp16 = slice_by_index(begin = var_46653_begin_0, end = var_46653_end_0, end_mask = var_46653_end_mask_0, x = k_415_cast_fp16)[name = tensor<string, []>("op_46653_cast_fp16")];
            tensor<int32, [4]> var_46657_begin_0 = const()[name = tensor<string, []>("op_46657_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_46657_end_0 = const()[name = tensor<string, []>("op_46657_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_46657_end_mask_0 = const()[name = tensor<string, []>("op_46657_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_46657_cast_fp16 = slice_by_index(begin = var_46657_begin_0, end = var_46657_end_0, end_mask = var_46657_end_mask_0, x = k_415_cast_fp16)[name = tensor<string, []>("op_46657_cast_fp16")];
            tensor<int32, [4]> var_46661_begin_0 = const()[name = tensor<string, []>("op_46661_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_46661_end_0 = const()[name = tensor<string, []>("op_46661_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_46661_end_mask_0 = const()[name = tensor<string, []>("op_46661_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_46661_cast_fp16 = slice_by_index(begin = var_46661_begin_0, end = var_46661_end_0, end_mask = var_46661_end_mask_0, x = k_415_cast_fp16)[name = tensor<string, []>("op_46661_cast_fp16")];
            tensor<int32, [4]> var_46665_begin_0 = const()[name = tensor<string, []>("op_46665_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_46665_end_0 = const()[name = tensor<string, []>("op_46665_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_46665_end_mask_0 = const()[name = tensor<string, []>("op_46665_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_46665_cast_fp16 = slice_by_index(begin = var_46665_begin_0, end = var_46665_end_0, end_mask = var_46665_end_mask_0, x = k_415_cast_fp16)[name = tensor<string, []>("op_46665_cast_fp16")];
            tensor<int32, [4]> var_46669_begin_0 = const()[name = tensor<string, []>("op_46669_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_46669_end_0 = const()[name = tensor<string, []>("op_46669_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_46669_end_mask_0 = const()[name = tensor<string, []>("op_46669_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_46669_cast_fp16 = slice_by_index(begin = var_46669_begin_0, end = var_46669_end_0, end_mask = var_46669_end_mask_0, x = k_415_cast_fp16)[name = tensor<string, []>("op_46669_cast_fp16")];
            tensor<int32, [4]> var_46673_begin_0 = const()[name = tensor<string, []>("op_46673_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_46673_end_0 = const()[name = tensor<string, []>("op_46673_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_46673_end_mask_0 = const()[name = tensor<string, []>("op_46673_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_46673_cast_fp16 = slice_by_index(begin = var_46673_begin_0, end = var_46673_end_0, end_mask = var_46673_end_mask_0, x = k_415_cast_fp16)[name = tensor<string, []>("op_46673_cast_fp16")];
            tensor<int32, [4]> var_46677_begin_0 = const()[name = tensor<string, []>("op_46677_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_46677_end_0 = const()[name = tensor<string, []>("op_46677_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_46677_end_mask_0 = const()[name = tensor<string, []>("op_46677_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_46677_cast_fp16 = slice_by_index(begin = var_46677_begin_0, end = var_46677_end_0, end_mask = var_46677_end_mask_0, x = k_415_cast_fp16)[name = tensor<string, []>("op_46677_cast_fp16")];
            tensor<int32, [4]> var_46681_begin_0 = const()[name = tensor<string, []>("op_46681_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_46681_end_0 = const()[name = tensor<string, []>("op_46681_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_46681_end_mask_0 = const()[name = tensor<string, []>("op_46681_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_46681_cast_fp16 = slice_by_index(begin = var_46681_begin_0, end = var_46681_end_0, end_mask = var_46681_end_mask_0, x = k_415_cast_fp16)[name = tensor<string, []>("op_46681_cast_fp16")];
            tensor<int32, [4]> var_46685_begin_0 = const()[name = tensor<string, []>("op_46685_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_46685_end_0 = const()[name = tensor<string, []>("op_46685_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_46685_end_mask_0 = const()[name = tensor<string, []>("op_46685_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_46685_cast_fp16 = slice_by_index(begin = var_46685_begin_0, end = var_46685_end_0, end_mask = var_46685_end_mask_0, x = k_415_cast_fp16)[name = tensor<string, []>("op_46685_cast_fp16")];
            tensor<int32, [4]> var_46687_begin_0 = const()[name = tensor<string, []>("op_46687_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_46687_end_0 = const()[name = tensor<string, []>("op_46687_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_46687_end_mask_0 = const()[name = tensor<string, []>("op_46687_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_46687_cast_fp16 = slice_by_index(begin = var_46687_begin_0, end = var_46687_end_0, end_mask = var_46687_end_mask_0, x = v_207_cast_fp16)[name = tensor<string, []>("op_46687_cast_fp16")];
            tensor<int32, [4]> var_46691_begin_0 = const()[name = tensor<string, []>("op_46691_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_46691_end_0 = const()[name = tensor<string, []>("op_46691_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_46691_end_mask_0 = const()[name = tensor<string, []>("op_46691_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_46691_cast_fp16 = slice_by_index(begin = var_46691_begin_0, end = var_46691_end_0, end_mask = var_46691_end_mask_0, x = v_207_cast_fp16)[name = tensor<string, []>("op_46691_cast_fp16")];
            tensor<int32, [4]> var_46695_begin_0 = const()[name = tensor<string, []>("op_46695_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_46695_end_0 = const()[name = tensor<string, []>("op_46695_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_46695_end_mask_0 = const()[name = tensor<string, []>("op_46695_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_46695_cast_fp16 = slice_by_index(begin = var_46695_begin_0, end = var_46695_end_0, end_mask = var_46695_end_mask_0, x = v_207_cast_fp16)[name = tensor<string, []>("op_46695_cast_fp16")];
            tensor<int32, [4]> var_46699_begin_0 = const()[name = tensor<string, []>("op_46699_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_46699_end_0 = const()[name = tensor<string, []>("op_46699_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_46699_end_mask_0 = const()[name = tensor<string, []>("op_46699_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_46699_cast_fp16 = slice_by_index(begin = var_46699_begin_0, end = var_46699_end_0, end_mask = var_46699_end_mask_0, x = v_207_cast_fp16)[name = tensor<string, []>("op_46699_cast_fp16")];
            tensor<int32, [4]> var_46703_begin_0 = const()[name = tensor<string, []>("op_46703_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_46703_end_0 = const()[name = tensor<string, []>("op_46703_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_46703_end_mask_0 = const()[name = tensor<string, []>("op_46703_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_46703_cast_fp16 = slice_by_index(begin = var_46703_begin_0, end = var_46703_end_0, end_mask = var_46703_end_mask_0, x = v_207_cast_fp16)[name = tensor<string, []>("op_46703_cast_fp16")];
            tensor<int32, [4]> var_46707_begin_0 = const()[name = tensor<string, []>("op_46707_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_46707_end_0 = const()[name = tensor<string, []>("op_46707_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_46707_end_mask_0 = const()[name = tensor<string, []>("op_46707_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_46707_cast_fp16 = slice_by_index(begin = var_46707_begin_0, end = var_46707_end_0, end_mask = var_46707_end_mask_0, x = v_207_cast_fp16)[name = tensor<string, []>("op_46707_cast_fp16")];
            tensor<int32, [4]> var_46711_begin_0 = const()[name = tensor<string, []>("op_46711_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_46711_end_0 = const()[name = tensor<string, []>("op_46711_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_46711_end_mask_0 = const()[name = tensor<string, []>("op_46711_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_46711_cast_fp16 = slice_by_index(begin = var_46711_begin_0, end = var_46711_end_0, end_mask = var_46711_end_mask_0, x = v_207_cast_fp16)[name = tensor<string, []>("op_46711_cast_fp16")];
            tensor<int32, [4]> var_46715_begin_0 = const()[name = tensor<string, []>("op_46715_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_46715_end_0 = const()[name = tensor<string, []>("op_46715_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_46715_end_mask_0 = const()[name = tensor<string, []>("op_46715_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_46715_cast_fp16 = slice_by_index(begin = var_46715_begin_0, end = var_46715_end_0, end_mask = var_46715_end_mask_0, x = v_207_cast_fp16)[name = tensor<string, []>("op_46715_cast_fp16")];
            tensor<int32, [4]> var_46719_begin_0 = const()[name = tensor<string, []>("op_46719_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_46719_end_0 = const()[name = tensor<string, []>("op_46719_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_46719_end_mask_0 = const()[name = tensor<string, []>("op_46719_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_46719_cast_fp16 = slice_by_index(begin = var_46719_begin_0, end = var_46719_end_0, end_mask = var_46719_end_mask_0, x = v_207_cast_fp16)[name = tensor<string, []>("op_46719_cast_fp16")];
            tensor<int32, [4]> var_46723_begin_0 = const()[name = tensor<string, []>("op_46723_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_46723_end_0 = const()[name = tensor<string, []>("op_46723_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_46723_end_mask_0 = const()[name = tensor<string, []>("op_46723_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_46723_cast_fp16 = slice_by_index(begin = var_46723_begin_0, end = var_46723_end_0, end_mask = var_46723_end_mask_0, x = v_207_cast_fp16)[name = tensor<string, []>("op_46723_cast_fp16")];
            tensor<int32, [4]> var_46727_begin_0 = const()[name = tensor<string, []>("op_46727_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_46727_end_0 = const()[name = tensor<string, []>("op_46727_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_46727_end_mask_0 = const()[name = tensor<string, []>("op_46727_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_46727_cast_fp16 = slice_by_index(begin = var_46727_begin_0, end = var_46727_end_0, end_mask = var_46727_end_mask_0, x = v_207_cast_fp16)[name = tensor<string, []>("op_46727_cast_fp16")];
            tensor<int32, [4]> var_46731_begin_0 = const()[name = tensor<string, []>("op_46731_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_46731_end_0 = const()[name = tensor<string, []>("op_46731_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_46731_end_mask_0 = const()[name = tensor<string, []>("op_46731_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_46731_cast_fp16 = slice_by_index(begin = var_46731_begin_0, end = var_46731_end_0, end_mask = var_46731_end_mask_0, x = v_207_cast_fp16)[name = tensor<string, []>("op_46731_cast_fp16")];
            tensor<int32, [4]> var_46735_begin_0 = const()[name = tensor<string, []>("op_46735_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_46735_end_0 = const()[name = tensor<string, []>("op_46735_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_46735_end_mask_0 = const()[name = tensor<string, []>("op_46735_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_46735_cast_fp16 = slice_by_index(begin = var_46735_begin_0, end = var_46735_end_0, end_mask = var_46735_end_mask_0, x = v_207_cast_fp16)[name = tensor<string, []>("op_46735_cast_fp16")];
            tensor<int32, [4]> var_46739_begin_0 = const()[name = tensor<string, []>("op_46739_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_46739_end_0 = const()[name = tensor<string, []>("op_46739_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_46739_end_mask_0 = const()[name = tensor<string, []>("op_46739_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_46739_cast_fp16 = slice_by_index(begin = var_46739_begin_0, end = var_46739_end_0, end_mask = var_46739_end_mask_0, x = v_207_cast_fp16)[name = tensor<string, []>("op_46739_cast_fp16")];
            tensor<int32, [4]> var_46743_begin_0 = const()[name = tensor<string, []>("op_46743_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_46743_end_0 = const()[name = tensor<string, []>("op_46743_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_46743_end_mask_0 = const()[name = tensor<string, []>("op_46743_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_46743_cast_fp16 = slice_by_index(begin = var_46743_begin_0, end = var_46743_end_0, end_mask = var_46743_end_mask_0, x = v_207_cast_fp16)[name = tensor<string, []>("op_46743_cast_fp16")];
            tensor<int32, [4]> var_46747_begin_0 = const()[name = tensor<string, []>("op_46747_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_46747_end_0 = const()[name = tensor<string, []>("op_46747_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_46747_end_mask_0 = const()[name = tensor<string, []>("op_46747_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_46747_cast_fp16 = slice_by_index(begin = var_46747_begin_0, end = var_46747_end_0, end_mask = var_46747_end_mask_0, x = v_207_cast_fp16)[name = tensor<string, []>("op_46747_cast_fp16")];
            tensor<int32, [4]> var_46751_begin_0 = const()[name = tensor<string, []>("op_46751_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_46751_end_0 = const()[name = tensor<string, []>("op_46751_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_46751_end_mask_0 = const()[name = tensor<string, []>("op_46751_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_46751_cast_fp16 = slice_by_index(begin = var_46751_begin_0, end = var_46751_end_0, end_mask = var_46751_end_mask_0, x = v_207_cast_fp16)[name = tensor<string, []>("op_46751_cast_fp16")];
            tensor<int32, [4]> var_46755_begin_0 = const()[name = tensor<string, []>("op_46755_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_46755_end_0 = const()[name = tensor<string, []>("op_46755_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_46755_end_mask_0 = const()[name = tensor<string, []>("op_46755_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_46755_cast_fp16 = slice_by_index(begin = var_46755_begin_0, end = var_46755_end_0, end_mask = var_46755_end_mask_0, x = v_207_cast_fp16)[name = tensor<string, []>("op_46755_cast_fp16")];
            tensor<int32, [4]> var_46759_begin_0 = const()[name = tensor<string, []>("op_46759_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_46759_end_0 = const()[name = tensor<string, []>("op_46759_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_46759_end_mask_0 = const()[name = tensor<string, []>("op_46759_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_46759_cast_fp16 = slice_by_index(begin = var_46759_begin_0, end = var_46759_end_0, end_mask = var_46759_end_mask_0, x = v_207_cast_fp16)[name = tensor<string, []>("op_46759_cast_fp16")];
            tensor<int32, [4]> var_46763_begin_0 = const()[name = tensor<string, []>("op_46763_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_46763_end_0 = const()[name = tensor<string, []>("op_46763_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_46763_end_mask_0 = const()[name = tensor<string, []>("op_46763_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_46763_cast_fp16 = slice_by_index(begin = var_46763_begin_0, end = var_46763_end_0, end_mask = var_46763_end_mask_0, x = v_207_cast_fp16)[name = tensor<string, []>("op_46763_cast_fp16")];
            tensor<string, []> var_46767_equation_0 = const()[name = tensor<string, []>("op_46767_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_46767_cast_fp16 = einsum(equation = var_46767_equation_0, values = (var_46609_cast_fp16, var_46526_cast_fp16))[name = tensor<string, []>("op_46767_cast_fp16")];
            tensor<fp16, []> var_46768_to_fp16 = const()[name = tensor<string, []>("op_46768_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3961_cast_fp16 = mul(x = var_46767_cast_fp16, y = var_46768_to_fp16)[name = tensor<string, []>("aw_3961_cast_fp16")];
            tensor<string, []> var_46771_equation_0 = const()[name = tensor<string, []>("op_46771_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_46771_cast_fp16 = einsum(equation = var_46771_equation_0, values = (var_46613_cast_fp16, var_46530_cast_fp16))[name = tensor<string, []>("op_46771_cast_fp16")];
            tensor<fp16, []> var_46772_to_fp16 = const()[name = tensor<string, []>("op_46772_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3963_cast_fp16 = mul(x = var_46771_cast_fp16, y = var_46772_to_fp16)[name = tensor<string, []>("aw_3963_cast_fp16")];
            tensor<string, []> var_46775_equation_0 = const()[name = tensor<string, []>("op_46775_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_46775_cast_fp16 = einsum(equation = var_46775_equation_0, values = (var_46617_cast_fp16, var_46534_cast_fp16))[name = tensor<string, []>("op_46775_cast_fp16")];
            tensor<fp16, []> var_46776_to_fp16 = const()[name = tensor<string, []>("op_46776_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3965_cast_fp16 = mul(x = var_46775_cast_fp16, y = var_46776_to_fp16)[name = tensor<string, []>("aw_3965_cast_fp16")];
            tensor<string, []> var_46779_equation_0 = const()[name = tensor<string, []>("op_46779_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_46779_cast_fp16 = einsum(equation = var_46779_equation_0, values = (var_46621_cast_fp16, var_46538_cast_fp16))[name = tensor<string, []>("op_46779_cast_fp16")];
            tensor<fp16, []> var_46780_to_fp16 = const()[name = tensor<string, []>("op_46780_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3967_cast_fp16 = mul(x = var_46779_cast_fp16, y = var_46780_to_fp16)[name = tensor<string, []>("aw_3967_cast_fp16")];
            tensor<string, []> var_46783_equation_0 = const()[name = tensor<string, []>("op_46783_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_46783_cast_fp16 = einsum(equation = var_46783_equation_0, values = (var_46625_cast_fp16, var_46542_cast_fp16))[name = tensor<string, []>("op_46783_cast_fp16")];
            tensor<fp16, []> var_46784_to_fp16 = const()[name = tensor<string, []>("op_46784_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3969_cast_fp16 = mul(x = var_46783_cast_fp16, y = var_46784_to_fp16)[name = tensor<string, []>("aw_3969_cast_fp16")];
            tensor<string, []> var_46787_equation_0 = const()[name = tensor<string, []>("op_46787_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_46787_cast_fp16 = einsum(equation = var_46787_equation_0, values = (var_46629_cast_fp16, var_46546_cast_fp16))[name = tensor<string, []>("op_46787_cast_fp16")];
            tensor<fp16, []> var_46788_to_fp16 = const()[name = tensor<string, []>("op_46788_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3971_cast_fp16 = mul(x = var_46787_cast_fp16, y = var_46788_to_fp16)[name = tensor<string, []>("aw_3971_cast_fp16")];
            tensor<string, []> var_46791_equation_0 = const()[name = tensor<string, []>("op_46791_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_46791_cast_fp16 = einsum(equation = var_46791_equation_0, values = (var_46633_cast_fp16, var_46550_cast_fp16))[name = tensor<string, []>("op_46791_cast_fp16")];
            tensor<fp16, []> var_46792_to_fp16 = const()[name = tensor<string, []>("op_46792_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3973_cast_fp16 = mul(x = var_46791_cast_fp16, y = var_46792_to_fp16)[name = tensor<string, []>("aw_3973_cast_fp16")];
            tensor<string, []> var_46795_equation_0 = const()[name = tensor<string, []>("op_46795_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_46795_cast_fp16 = einsum(equation = var_46795_equation_0, values = (var_46637_cast_fp16, var_46554_cast_fp16))[name = tensor<string, []>("op_46795_cast_fp16")];
            tensor<fp16, []> var_46796_to_fp16 = const()[name = tensor<string, []>("op_46796_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3975_cast_fp16 = mul(x = var_46795_cast_fp16, y = var_46796_to_fp16)[name = tensor<string, []>("aw_3975_cast_fp16")];
            tensor<string, []> var_46799_equation_0 = const()[name = tensor<string, []>("op_46799_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_46799_cast_fp16 = einsum(equation = var_46799_equation_0, values = (var_46641_cast_fp16, var_46558_cast_fp16))[name = tensor<string, []>("op_46799_cast_fp16")];
            tensor<fp16, []> var_46800_to_fp16 = const()[name = tensor<string, []>("op_46800_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3977_cast_fp16 = mul(x = var_46799_cast_fp16, y = var_46800_to_fp16)[name = tensor<string, []>("aw_3977_cast_fp16")];
            tensor<string, []> var_46803_equation_0 = const()[name = tensor<string, []>("op_46803_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_46803_cast_fp16 = einsum(equation = var_46803_equation_0, values = (var_46645_cast_fp16, var_46562_cast_fp16))[name = tensor<string, []>("op_46803_cast_fp16")];
            tensor<fp16, []> var_46804_to_fp16 = const()[name = tensor<string, []>("op_46804_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3979_cast_fp16 = mul(x = var_46803_cast_fp16, y = var_46804_to_fp16)[name = tensor<string, []>("aw_3979_cast_fp16")];
            tensor<string, []> var_46807_equation_0 = const()[name = tensor<string, []>("op_46807_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_46807_cast_fp16 = einsum(equation = var_46807_equation_0, values = (var_46649_cast_fp16, var_46566_cast_fp16))[name = tensor<string, []>("op_46807_cast_fp16")];
            tensor<fp16, []> var_46808_to_fp16 = const()[name = tensor<string, []>("op_46808_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3981_cast_fp16 = mul(x = var_46807_cast_fp16, y = var_46808_to_fp16)[name = tensor<string, []>("aw_3981_cast_fp16")];
            tensor<string, []> var_46811_equation_0 = const()[name = tensor<string, []>("op_46811_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_46811_cast_fp16 = einsum(equation = var_46811_equation_0, values = (var_46653_cast_fp16, var_46570_cast_fp16))[name = tensor<string, []>("op_46811_cast_fp16")];
            tensor<fp16, []> var_46812_to_fp16 = const()[name = tensor<string, []>("op_46812_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3983_cast_fp16 = mul(x = var_46811_cast_fp16, y = var_46812_to_fp16)[name = tensor<string, []>("aw_3983_cast_fp16")];
            tensor<string, []> var_46815_equation_0 = const()[name = tensor<string, []>("op_46815_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_46815_cast_fp16 = einsum(equation = var_46815_equation_0, values = (var_46657_cast_fp16, var_46574_cast_fp16))[name = tensor<string, []>("op_46815_cast_fp16")];
            tensor<fp16, []> var_46816_to_fp16 = const()[name = tensor<string, []>("op_46816_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3985_cast_fp16 = mul(x = var_46815_cast_fp16, y = var_46816_to_fp16)[name = tensor<string, []>("aw_3985_cast_fp16")];
            tensor<string, []> var_46819_equation_0 = const()[name = tensor<string, []>("op_46819_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_46819_cast_fp16 = einsum(equation = var_46819_equation_0, values = (var_46661_cast_fp16, var_46578_cast_fp16))[name = tensor<string, []>("op_46819_cast_fp16")];
            tensor<fp16, []> var_46820_to_fp16 = const()[name = tensor<string, []>("op_46820_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3987_cast_fp16 = mul(x = var_46819_cast_fp16, y = var_46820_to_fp16)[name = tensor<string, []>("aw_3987_cast_fp16")];
            tensor<string, []> var_46823_equation_0 = const()[name = tensor<string, []>("op_46823_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_46823_cast_fp16 = einsum(equation = var_46823_equation_0, values = (var_46665_cast_fp16, var_46582_cast_fp16))[name = tensor<string, []>("op_46823_cast_fp16")];
            tensor<fp16, []> var_46824_to_fp16 = const()[name = tensor<string, []>("op_46824_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3989_cast_fp16 = mul(x = var_46823_cast_fp16, y = var_46824_to_fp16)[name = tensor<string, []>("aw_3989_cast_fp16")];
            tensor<string, []> var_46827_equation_0 = const()[name = tensor<string, []>("op_46827_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_46827_cast_fp16 = einsum(equation = var_46827_equation_0, values = (var_46669_cast_fp16, var_46586_cast_fp16))[name = tensor<string, []>("op_46827_cast_fp16")];
            tensor<fp16, []> var_46828_to_fp16 = const()[name = tensor<string, []>("op_46828_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3991_cast_fp16 = mul(x = var_46827_cast_fp16, y = var_46828_to_fp16)[name = tensor<string, []>("aw_3991_cast_fp16")];
            tensor<string, []> var_46831_equation_0 = const()[name = tensor<string, []>("op_46831_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_46831_cast_fp16 = einsum(equation = var_46831_equation_0, values = (var_46673_cast_fp16, var_46590_cast_fp16))[name = tensor<string, []>("op_46831_cast_fp16")];
            tensor<fp16, []> var_46832_to_fp16 = const()[name = tensor<string, []>("op_46832_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3993_cast_fp16 = mul(x = var_46831_cast_fp16, y = var_46832_to_fp16)[name = tensor<string, []>("aw_3993_cast_fp16")];
            tensor<string, []> var_46835_equation_0 = const()[name = tensor<string, []>("op_46835_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_46835_cast_fp16 = einsum(equation = var_46835_equation_0, values = (var_46677_cast_fp16, var_46594_cast_fp16))[name = tensor<string, []>("op_46835_cast_fp16")];
            tensor<fp16, []> var_46836_to_fp16 = const()[name = tensor<string, []>("op_46836_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3995_cast_fp16 = mul(x = var_46835_cast_fp16, y = var_46836_to_fp16)[name = tensor<string, []>("aw_3995_cast_fp16")];
            tensor<string, []> var_46839_equation_0 = const()[name = tensor<string, []>("op_46839_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_46839_cast_fp16 = einsum(equation = var_46839_equation_0, values = (var_46681_cast_fp16, var_46598_cast_fp16))[name = tensor<string, []>("op_46839_cast_fp16")];
            tensor<fp16, []> var_46840_to_fp16 = const()[name = tensor<string, []>("op_46840_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3997_cast_fp16 = mul(x = var_46839_cast_fp16, y = var_46840_to_fp16)[name = tensor<string, []>("aw_3997_cast_fp16")];
            tensor<string, []> var_46843_equation_0 = const()[name = tensor<string, []>("op_46843_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_46843_cast_fp16 = einsum(equation = var_46843_equation_0, values = (var_46685_cast_fp16, var_46602_cast_fp16))[name = tensor<string, []>("op_46843_cast_fp16")];
            tensor<fp16, []> var_46844_to_fp16 = const()[name = tensor<string, []>("op_46844_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3999_cast_fp16 = mul(x = var_46843_cast_fp16, y = var_46844_to_fp16)[name = tensor<string, []>("aw_3999_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_46846_cast_fp16 = softmax(axis = var_30355, x = aw_3961_cast_fp16)[name = tensor<string, []>("op_46846_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_46847_cast_fp16 = softmax(axis = var_30355, x = aw_3963_cast_fp16)[name = tensor<string, []>("op_46847_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_46848_cast_fp16 = softmax(axis = var_30355, x = aw_3965_cast_fp16)[name = tensor<string, []>("op_46848_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_46849_cast_fp16 = softmax(axis = var_30355, x = aw_3967_cast_fp16)[name = tensor<string, []>("op_46849_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_46850_cast_fp16 = softmax(axis = var_30355, x = aw_3969_cast_fp16)[name = tensor<string, []>("op_46850_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_46851_cast_fp16 = softmax(axis = var_30355, x = aw_3971_cast_fp16)[name = tensor<string, []>("op_46851_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_46852_cast_fp16 = softmax(axis = var_30355, x = aw_3973_cast_fp16)[name = tensor<string, []>("op_46852_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_46853_cast_fp16 = softmax(axis = var_30355, x = aw_3975_cast_fp16)[name = tensor<string, []>("op_46853_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_46854_cast_fp16 = softmax(axis = var_30355, x = aw_3977_cast_fp16)[name = tensor<string, []>("op_46854_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_46855_cast_fp16 = softmax(axis = var_30355, x = aw_3979_cast_fp16)[name = tensor<string, []>("op_46855_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_46856_cast_fp16 = softmax(axis = var_30355, x = aw_3981_cast_fp16)[name = tensor<string, []>("op_46856_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_46857_cast_fp16 = softmax(axis = var_30355, x = aw_3983_cast_fp16)[name = tensor<string, []>("op_46857_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_46858_cast_fp16 = softmax(axis = var_30355, x = aw_3985_cast_fp16)[name = tensor<string, []>("op_46858_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_46859_cast_fp16 = softmax(axis = var_30355, x = aw_3987_cast_fp16)[name = tensor<string, []>("op_46859_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_46860_cast_fp16 = softmax(axis = var_30355, x = aw_3989_cast_fp16)[name = tensor<string, []>("op_46860_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_46861_cast_fp16 = softmax(axis = var_30355, x = aw_3991_cast_fp16)[name = tensor<string, []>("op_46861_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_46862_cast_fp16 = softmax(axis = var_30355, x = aw_3993_cast_fp16)[name = tensor<string, []>("op_46862_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_46863_cast_fp16 = softmax(axis = var_30355, x = aw_3995_cast_fp16)[name = tensor<string, []>("op_46863_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_46864_cast_fp16 = softmax(axis = var_30355, x = aw_3997_cast_fp16)[name = tensor<string, []>("op_46864_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_46865_cast_fp16 = softmax(axis = var_30355, x = aw_3999_cast_fp16)[name = tensor<string, []>("op_46865_cast_fp16")];
            tensor<string, []> var_46867_equation_0 = const()[name = tensor<string, []>("op_46867_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46867_cast_fp16 = einsum(equation = var_46867_equation_0, values = (var_46687_cast_fp16, var_46846_cast_fp16))[name = tensor<string, []>("op_46867_cast_fp16")];
            tensor<string, []> var_46869_equation_0 = const()[name = tensor<string, []>("op_46869_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46869_cast_fp16 = einsum(equation = var_46869_equation_0, values = (var_46691_cast_fp16, var_46847_cast_fp16))[name = tensor<string, []>("op_46869_cast_fp16")];
            tensor<string, []> var_46871_equation_0 = const()[name = tensor<string, []>("op_46871_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46871_cast_fp16 = einsum(equation = var_46871_equation_0, values = (var_46695_cast_fp16, var_46848_cast_fp16))[name = tensor<string, []>("op_46871_cast_fp16")];
            tensor<string, []> var_46873_equation_0 = const()[name = tensor<string, []>("op_46873_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46873_cast_fp16 = einsum(equation = var_46873_equation_0, values = (var_46699_cast_fp16, var_46849_cast_fp16))[name = tensor<string, []>("op_46873_cast_fp16")];
            tensor<string, []> var_46875_equation_0 = const()[name = tensor<string, []>("op_46875_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46875_cast_fp16 = einsum(equation = var_46875_equation_0, values = (var_46703_cast_fp16, var_46850_cast_fp16))[name = tensor<string, []>("op_46875_cast_fp16")];
            tensor<string, []> var_46877_equation_0 = const()[name = tensor<string, []>("op_46877_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46877_cast_fp16 = einsum(equation = var_46877_equation_0, values = (var_46707_cast_fp16, var_46851_cast_fp16))[name = tensor<string, []>("op_46877_cast_fp16")];
            tensor<string, []> var_46879_equation_0 = const()[name = tensor<string, []>("op_46879_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46879_cast_fp16 = einsum(equation = var_46879_equation_0, values = (var_46711_cast_fp16, var_46852_cast_fp16))[name = tensor<string, []>("op_46879_cast_fp16")];
            tensor<string, []> var_46881_equation_0 = const()[name = tensor<string, []>("op_46881_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46881_cast_fp16 = einsum(equation = var_46881_equation_0, values = (var_46715_cast_fp16, var_46853_cast_fp16))[name = tensor<string, []>("op_46881_cast_fp16")];
            tensor<string, []> var_46883_equation_0 = const()[name = tensor<string, []>("op_46883_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46883_cast_fp16 = einsum(equation = var_46883_equation_0, values = (var_46719_cast_fp16, var_46854_cast_fp16))[name = tensor<string, []>("op_46883_cast_fp16")];
            tensor<string, []> var_46885_equation_0 = const()[name = tensor<string, []>("op_46885_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46885_cast_fp16 = einsum(equation = var_46885_equation_0, values = (var_46723_cast_fp16, var_46855_cast_fp16))[name = tensor<string, []>("op_46885_cast_fp16")];
            tensor<string, []> var_46887_equation_0 = const()[name = tensor<string, []>("op_46887_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46887_cast_fp16 = einsum(equation = var_46887_equation_0, values = (var_46727_cast_fp16, var_46856_cast_fp16))[name = tensor<string, []>("op_46887_cast_fp16")];
            tensor<string, []> var_46889_equation_0 = const()[name = tensor<string, []>("op_46889_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46889_cast_fp16 = einsum(equation = var_46889_equation_0, values = (var_46731_cast_fp16, var_46857_cast_fp16))[name = tensor<string, []>("op_46889_cast_fp16")];
            tensor<string, []> var_46891_equation_0 = const()[name = tensor<string, []>("op_46891_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46891_cast_fp16 = einsum(equation = var_46891_equation_0, values = (var_46735_cast_fp16, var_46858_cast_fp16))[name = tensor<string, []>("op_46891_cast_fp16")];
            tensor<string, []> var_46893_equation_0 = const()[name = tensor<string, []>("op_46893_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46893_cast_fp16 = einsum(equation = var_46893_equation_0, values = (var_46739_cast_fp16, var_46859_cast_fp16))[name = tensor<string, []>("op_46893_cast_fp16")];
            tensor<string, []> var_46895_equation_0 = const()[name = tensor<string, []>("op_46895_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46895_cast_fp16 = einsum(equation = var_46895_equation_0, values = (var_46743_cast_fp16, var_46860_cast_fp16))[name = tensor<string, []>("op_46895_cast_fp16")];
            tensor<string, []> var_46897_equation_0 = const()[name = tensor<string, []>("op_46897_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46897_cast_fp16 = einsum(equation = var_46897_equation_0, values = (var_46747_cast_fp16, var_46861_cast_fp16))[name = tensor<string, []>("op_46897_cast_fp16")];
            tensor<string, []> var_46899_equation_0 = const()[name = tensor<string, []>("op_46899_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46899_cast_fp16 = einsum(equation = var_46899_equation_0, values = (var_46751_cast_fp16, var_46862_cast_fp16))[name = tensor<string, []>("op_46899_cast_fp16")];
            tensor<string, []> var_46901_equation_0 = const()[name = tensor<string, []>("op_46901_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46901_cast_fp16 = einsum(equation = var_46901_equation_0, values = (var_46755_cast_fp16, var_46863_cast_fp16))[name = tensor<string, []>("op_46901_cast_fp16")];
            tensor<string, []> var_46903_equation_0 = const()[name = tensor<string, []>("op_46903_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46903_cast_fp16 = einsum(equation = var_46903_equation_0, values = (var_46759_cast_fp16, var_46864_cast_fp16))[name = tensor<string, []>("op_46903_cast_fp16")];
            tensor<string, []> var_46905_equation_0 = const()[name = tensor<string, []>("op_46905_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46905_cast_fp16 = einsum(equation = var_46905_equation_0, values = (var_46763_cast_fp16, var_46865_cast_fp16))[name = tensor<string, []>("op_46905_cast_fp16")];
            tensor<bool, []> input_595_interleave_0 = const()[name = tensor<string, []>("input_595_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_595_cast_fp16 = concat(axis = var_30355, interleave = input_595_interleave_0, values = (var_46867_cast_fp16, var_46869_cast_fp16, var_46871_cast_fp16, var_46873_cast_fp16, var_46875_cast_fp16, var_46877_cast_fp16, var_46879_cast_fp16, var_46881_cast_fp16, var_46883_cast_fp16, var_46885_cast_fp16, var_46887_cast_fp16, var_46889_cast_fp16, var_46891_cast_fp16, var_46893_cast_fp16, var_46895_cast_fp16, var_46897_cast_fp16, var_46899_cast_fp16, var_46901_cast_fp16, var_46903_cast_fp16, var_46905_cast_fp16))[name = tensor<string, []>("input_595_cast_fp16")];
            tensor<string, []> var_46915_pad_type_0 = const()[name = tensor<string, []>("op_46915_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_46915_strides_0 = const()[name = tensor<string, []>("op_46915_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_46915_pad_0 = const()[name = tensor<string, []>("op_46915_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_46915_dilations_0 = const()[name = tensor<string, []>("op_46915_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_46915_groups_0 = const()[name = tensor<string, []>("op_46915_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_7_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(496821760))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(498050624))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_7_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_7_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_7_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(498050816)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_46915_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_7_attn2_to_out_0_bias_to_fp16, dilations = var_46915_dilations_0, groups = var_46915_groups_0, pad = var_46915_pad_0, pad_type = var_46915_pad_type_0, strides = var_46915_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_7_attn2_to_out_0_weight_to_fp16_palettized, x = input_595_cast_fp16)[name = tensor<string, []>("op_46915_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_311_cast_fp16 = add(x = var_46915_cast_fp16, y = inputs_309_cast_fp16)[name = tensor<string, []>("inputs_311_cast_fp16")];
            tensor<int32, [1]> input_597_axes_0 = const()[name = tensor<string, []>("input_597_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> input_597_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_597_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(498053440)))];
            tensor<fp16, [1280]> input_597_beta_0_to_fp16 = const()[name = tensor<string, []>("input_597_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(498056064)))];
            tensor<fp16, []> var_46925_to_fp16 = const()[name = tensor<string, []>("op_46925_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> input_597_cast_fp16 = layer_norm(axes = input_597_axes_0, beta = input_597_beta_0_to_fp16, epsilon = var_46925_to_fp16, gamma = input_597_gamma_0_to_fp16, x = inputs_311_cast_fp16)[name = tensor<string, []>("input_597_cast_fp16")];
            tensor<string, []> var_46945_pad_type_0 = const()[name = tensor<string, []>("op_46945_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_46945_strides_0 = const()[name = tensor<string, []>("op_46945_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_46945_pad_0 = const()[name = tensor<string, []>("op_46945_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_46945_dilations_0 = const()[name = tensor<string, []>("op_46945_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_46945_groups_0 = const()[name = tensor<string, []>("op_46945_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_7_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [9830400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(498058688))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(507889152))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_7_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([10240, 1280, 1, 1])];
            tensor<fp16, [10240]> up_blocks_0_attentions_1_transformer_blocks_7_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_7_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(507889344)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_46945_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_7_ff_net_0_proj_bias_to_fp16, dilations = var_46945_dilations_0, groups = var_46945_groups_0, pad = var_46945_pad_0, pad_type = var_46945_pad_type_0, strides = var_46945_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_7_ff_net_0_proj_weight_to_fp16_palettized, x = input_597_cast_fp16)[name = tensor<string, []>("op_46945_cast_fp16")];
            tensor<int32, [2]> var_46946_split_sizes_0 = const()[name = tensor<string, []>("op_46946_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_46946_axis_0 = const()[name = tensor<string, []>("op_46946_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_46946_cast_fp16_0, tensor<fp16, [2, 5120, 1, 1024]> var_46946_cast_fp16_1 = split(axis = var_46946_axis_0, split_sizes = var_46946_split_sizes_0, x = var_46945_cast_fp16)[name = tensor<string, []>("op_46946_cast_fp16")];
            tensor<string, []> var_46948_mode_0 = const()[name = tensor<string, []>("op_46948_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_46948_cast_fp16 = gelu(mode = var_46948_mode_0, x = var_46946_cast_fp16_1)[name = tensor<string, []>("op_46948_cast_fp16")];
            tensor<fp16, [2, 5120, 1, 1024]> input_599_cast_fp16 = mul(x = var_46946_cast_fp16_0, y = var_46948_cast_fp16)[name = tensor<string, []>("input_599_cast_fp16")];
            tensor<string, []> var_46956_pad_type_0 = const()[name = tensor<string, []>("op_46956_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_46956_strides_0 = const()[name = tensor<string, []>("op_46956_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_46956_pad_0 = const()[name = tensor<string, []>("op_46956_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_46956_dilations_0 = const()[name = tensor<string, []>("op_46956_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_46956_groups_0 = const()[name = tensor<string, []>("op_46956_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_7_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [4915200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(507909888))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(512825152))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_7_ff_net_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 5120, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_7_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_7_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(512825344)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_46956_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_7_ff_net_2_bias_to_fp16, dilations = var_46956_dilations_0, groups = var_46956_groups_0, pad = var_46956_pad_0, pad_type = var_46956_pad_type_0, strides = var_46956_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_7_ff_net_2_weight_to_fp16_palettized, x = input_599_cast_fp16)[name = tensor<string, []>("op_46956_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_313_cast_fp16 = add(x = var_46956_cast_fp16, y = inputs_311_cast_fp16)[name = tensor<string, []>("inputs_313_cast_fp16")];
            tensor<int32, [1]> hidden_states_409_axes_0 = const()[name = tensor<string, []>("hidden_states_409_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_409_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_409_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(512827968)))];
            tensor<fp16, [1280]> hidden_states_409_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_409_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(512830592)))];
            tensor<fp16, []> var_46972_to_fp16 = const()[name = tensor<string, []>("op_46972_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_409_cast_fp16 = layer_norm(axes = hidden_states_409_axes_0, beta = hidden_states_409_beta_0_to_fp16, epsilon = var_46972_to_fp16, gamma = hidden_states_409_gamma_0_to_fp16, x = inputs_313_cast_fp16)[name = tensor<string, []>("hidden_states_409_cast_fp16")];
            tensor<string, []> q_209_pad_type_0 = const()[name = tensor<string, []>("q_209_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_209_strides_0 = const()[name = tensor<string, []>("q_209_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_209_pad_0 = const()[name = tensor<string, []>("q_209_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_209_dilations_0 = const()[name = tensor<string, []>("q_209_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_209_groups_0 = const()[name = tensor<string, []>("q_209_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_8_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(512833216))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(514062080))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_8_attn1_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_209_cast_fp16 = conv(dilations = q_209_dilations_0, groups = q_209_groups_0, pad = q_209_pad_0, pad_type = q_209_pad_type_0, strides = q_209_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_8_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_409_cast_fp16)[name = tensor<string, []>("q_209_cast_fp16")];
            tensor<string, []> k_417_pad_type_0 = const()[name = tensor<string, []>("k_417_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_417_strides_0 = const()[name = tensor<string, []>("k_417_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_417_pad_0 = const()[name = tensor<string, []>("k_417_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_417_dilations_0 = const()[name = tensor<string, []>("k_417_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_417_groups_0 = const()[name = tensor<string, []>("k_417_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_8_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(514062272))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(515291136))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_8_attn1_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> k_417_cast_fp16 = conv(dilations = k_417_dilations_0, groups = k_417_groups_0, pad = k_417_pad_0, pad_type = k_417_pad_type_0, strides = k_417_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_8_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_409_cast_fp16)[name = tensor<string, []>("k_417_cast_fp16")];
            tensor<string, []> v_209_pad_type_0 = const()[name = tensor<string, []>("v_209_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_209_strides_0 = const()[name = tensor<string, []>("v_209_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_209_pad_0 = const()[name = tensor<string, []>("v_209_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_209_dilations_0 = const()[name = tensor<string, []>("v_209_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_209_groups_0 = const()[name = tensor<string, []>("v_209_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_8_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(515291328))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(516520192))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_8_attn1_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> v_209_cast_fp16 = conv(dilations = v_209_dilations_0, groups = v_209_groups_0, pad = v_209_pad_0, pad_type = v_209_pad_type_0, strides = v_209_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_8_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_409_cast_fp16)[name = tensor<string, []>("v_209_cast_fp16")];
            tensor<int32, [4]> var_47005_begin_0 = const()[name = tensor<string, []>("op_47005_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_47005_end_0 = const()[name = tensor<string, []>("op_47005_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_47005_end_mask_0 = const()[name = tensor<string, []>("op_47005_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47005_cast_fp16 = slice_by_index(begin = var_47005_begin_0, end = var_47005_end_0, end_mask = var_47005_end_mask_0, x = q_209_cast_fp16)[name = tensor<string, []>("op_47005_cast_fp16")];
            tensor<int32, [4]> var_47009_begin_0 = const()[name = tensor<string, []>("op_47009_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_47009_end_0 = const()[name = tensor<string, []>("op_47009_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_47009_end_mask_0 = const()[name = tensor<string, []>("op_47009_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47009_cast_fp16 = slice_by_index(begin = var_47009_begin_0, end = var_47009_end_0, end_mask = var_47009_end_mask_0, x = q_209_cast_fp16)[name = tensor<string, []>("op_47009_cast_fp16")];
            tensor<int32, [4]> var_47013_begin_0 = const()[name = tensor<string, []>("op_47013_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_47013_end_0 = const()[name = tensor<string, []>("op_47013_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_47013_end_mask_0 = const()[name = tensor<string, []>("op_47013_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47013_cast_fp16 = slice_by_index(begin = var_47013_begin_0, end = var_47013_end_0, end_mask = var_47013_end_mask_0, x = q_209_cast_fp16)[name = tensor<string, []>("op_47013_cast_fp16")];
            tensor<int32, [4]> var_47017_begin_0 = const()[name = tensor<string, []>("op_47017_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_47017_end_0 = const()[name = tensor<string, []>("op_47017_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_47017_end_mask_0 = const()[name = tensor<string, []>("op_47017_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47017_cast_fp16 = slice_by_index(begin = var_47017_begin_0, end = var_47017_end_0, end_mask = var_47017_end_mask_0, x = q_209_cast_fp16)[name = tensor<string, []>("op_47017_cast_fp16")];
            tensor<int32, [4]> var_47021_begin_0 = const()[name = tensor<string, []>("op_47021_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_47021_end_0 = const()[name = tensor<string, []>("op_47021_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_47021_end_mask_0 = const()[name = tensor<string, []>("op_47021_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47021_cast_fp16 = slice_by_index(begin = var_47021_begin_0, end = var_47021_end_0, end_mask = var_47021_end_mask_0, x = q_209_cast_fp16)[name = tensor<string, []>("op_47021_cast_fp16")];
            tensor<int32, [4]> var_47025_begin_0 = const()[name = tensor<string, []>("op_47025_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_47025_end_0 = const()[name = tensor<string, []>("op_47025_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_47025_end_mask_0 = const()[name = tensor<string, []>("op_47025_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47025_cast_fp16 = slice_by_index(begin = var_47025_begin_0, end = var_47025_end_0, end_mask = var_47025_end_mask_0, x = q_209_cast_fp16)[name = tensor<string, []>("op_47025_cast_fp16")];
            tensor<int32, [4]> var_47029_begin_0 = const()[name = tensor<string, []>("op_47029_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_47029_end_0 = const()[name = tensor<string, []>("op_47029_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_47029_end_mask_0 = const()[name = tensor<string, []>("op_47029_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47029_cast_fp16 = slice_by_index(begin = var_47029_begin_0, end = var_47029_end_0, end_mask = var_47029_end_mask_0, x = q_209_cast_fp16)[name = tensor<string, []>("op_47029_cast_fp16")];
            tensor<int32, [4]> var_47033_begin_0 = const()[name = tensor<string, []>("op_47033_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_47033_end_0 = const()[name = tensor<string, []>("op_47033_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_47033_end_mask_0 = const()[name = tensor<string, []>("op_47033_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47033_cast_fp16 = slice_by_index(begin = var_47033_begin_0, end = var_47033_end_0, end_mask = var_47033_end_mask_0, x = q_209_cast_fp16)[name = tensor<string, []>("op_47033_cast_fp16")];
            tensor<int32, [4]> var_47037_begin_0 = const()[name = tensor<string, []>("op_47037_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_47037_end_0 = const()[name = tensor<string, []>("op_47037_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_47037_end_mask_0 = const()[name = tensor<string, []>("op_47037_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47037_cast_fp16 = slice_by_index(begin = var_47037_begin_0, end = var_47037_end_0, end_mask = var_47037_end_mask_0, x = q_209_cast_fp16)[name = tensor<string, []>("op_47037_cast_fp16")];
            tensor<int32, [4]> var_47041_begin_0 = const()[name = tensor<string, []>("op_47041_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_47041_end_0 = const()[name = tensor<string, []>("op_47041_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_47041_end_mask_0 = const()[name = tensor<string, []>("op_47041_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47041_cast_fp16 = slice_by_index(begin = var_47041_begin_0, end = var_47041_end_0, end_mask = var_47041_end_mask_0, x = q_209_cast_fp16)[name = tensor<string, []>("op_47041_cast_fp16")];
            tensor<int32, [4]> var_47045_begin_0 = const()[name = tensor<string, []>("op_47045_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_47045_end_0 = const()[name = tensor<string, []>("op_47045_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_47045_end_mask_0 = const()[name = tensor<string, []>("op_47045_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47045_cast_fp16 = slice_by_index(begin = var_47045_begin_0, end = var_47045_end_0, end_mask = var_47045_end_mask_0, x = q_209_cast_fp16)[name = tensor<string, []>("op_47045_cast_fp16")];
            tensor<int32, [4]> var_47049_begin_0 = const()[name = tensor<string, []>("op_47049_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_47049_end_0 = const()[name = tensor<string, []>("op_47049_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_47049_end_mask_0 = const()[name = tensor<string, []>("op_47049_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47049_cast_fp16 = slice_by_index(begin = var_47049_begin_0, end = var_47049_end_0, end_mask = var_47049_end_mask_0, x = q_209_cast_fp16)[name = tensor<string, []>("op_47049_cast_fp16")];
            tensor<int32, [4]> var_47053_begin_0 = const()[name = tensor<string, []>("op_47053_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_47053_end_0 = const()[name = tensor<string, []>("op_47053_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_47053_end_mask_0 = const()[name = tensor<string, []>("op_47053_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47053_cast_fp16 = slice_by_index(begin = var_47053_begin_0, end = var_47053_end_0, end_mask = var_47053_end_mask_0, x = q_209_cast_fp16)[name = tensor<string, []>("op_47053_cast_fp16")];
            tensor<int32, [4]> var_47057_begin_0 = const()[name = tensor<string, []>("op_47057_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_47057_end_0 = const()[name = tensor<string, []>("op_47057_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_47057_end_mask_0 = const()[name = tensor<string, []>("op_47057_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47057_cast_fp16 = slice_by_index(begin = var_47057_begin_0, end = var_47057_end_0, end_mask = var_47057_end_mask_0, x = q_209_cast_fp16)[name = tensor<string, []>("op_47057_cast_fp16")];
            tensor<int32, [4]> var_47061_begin_0 = const()[name = tensor<string, []>("op_47061_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_47061_end_0 = const()[name = tensor<string, []>("op_47061_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_47061_end_mask_0 = const()[name = tensor<string, []>("op_47061_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47061_cast_fp16 = slice_by_index(begin = var_47061_begin_0, end = var_47061_end_0, end_mask = var_47061_end_mask_0, x = q_209_cast_fp16)[name = tensor<string, []>("op_47061_cast_fp16")];
            tensor<int32, [4]> var_47065_begin_0 = const()[name = tensor<string, []>("op_47065_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_47065_end_0 = const()[name = tensor<string, []>("op_47065_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_47065_end_mask_0 = const()[name = tensor<string, []>("op_47065_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47065_cast_fp16 = slice_by_index(begin = var_47065_begin_0, end = var_47065_end_0, end_mask = var_47065_end_mask_0, x = q_209_cast_fp16)[name = tensor<string, []>("op_47065_cast_fp16")];
            tensor<int32, [4]> var_47069_begin_0 = const()[name = tensor<string, []>("op_47069_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_47069_end_0 = const()[name = tensor<string, []>("op_47069_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_47069_end_mask_0 = const()[name = tensor<string, []>("op_47069_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47069_cast_fp16 = slice_by_index(begin = var_47069_begin_0, end = var_47069_end_0, end_mask = var_47069_end_mask_0, x = q_209_cast_fp16)[name = tensor<string, []>("op_47069_cast_fp16")];
            tensor<int32, [4]> var_47073_begin_0 = const()[name = tensor<string, []>("op_47073_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_47073_end_0 = const()[name = tensor<string, []>("op_47073_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_47073_end_mask_0 = const()[name = tensor<string, []>("op_47073_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47073_cast_fp16 = slice_by_index(begin = var_47073_begin_0, end = var_47073_end_0, end_mask = var_47073_end_mask_0, x = q_209_cast_fp16)[name = tensor<string, []>("op_47073_cast_fp16")];
            tensor<int32, [4]> var_47077_begin_0 = const()[name = tensor<string, []>("op_47077_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_47077_end_0 = const()[name = tensor<string, []>("op_47077_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_47077_end_mask_0 = const()[name = tensor<string, []>("op_47077_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47077_cast_fp16 = slice_by_index(begin = var_47077_begin_0, end = var_47077_end_0, end_mask = var_47077_end_mask_0, x = q_209_cast_fp16)[name = tensor<string, []>("op_47077_cast_fp16")];
            tensor<int32, [4]> var_47081_begin_0 = const()[name = tensor<string, []>("op_47081_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_47081_end_0 = const()[name = tensor<string, []>("op_47081_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_47081_end_mask_0 = const()[name = tensor<string, []>("op_47081_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47081_cast_fp16 = slice_by_index(begin = var_47081_begin_0, end = var_47081_end_0, end_mask = var_47081_end_mask_0, x = q_209_cast_fp16)[name = tensor<string, []>("op_47081_cast_fp16")];
            tensor<int32, [4]> k_419_perm_0 = const()[name = tensor<string, []>("k_419_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_47088_begin_0 = const()[name = tensor<string, []>("op_47088_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_47088_end_0 = const()[name = tensor<string, []>("op_47088_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_47088_end_mask_0 = const()[name = tensor<string, []>("op_47088_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> k_419_cast_fp16 = transpose(perm = k_419_perm_0, x = k_417_cast_fp16)[name = tensor<string, []>("transpose_35")];
            tensor<fp16, [2, 1024, 1, 64]> var_47088_cast_fp16 = slice_by_index(begin = var_47088_begin_0, end = var_47088_end_0, end_mask = var_47088_end_mask_0, x = k_419_cast_fp16)[name = tensor<string, []>("op_47088_cast_fp16")];
            tensor<int32, [4]> var_47092_begin_0 = const()[name = tensor<string, []>("op_47092_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_47092_end_0 = const()[name = tensor<string, []>("op_47092_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_47092_end_mask_0 = const()[name = tensor<string, []>("op_47092_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_47092_cast_fp16 = slice_by_index(begin = var_47092_begin_0, end = var_47092_end_0, end_mask = var_47092_end_mask_0, x = k_419_cast_fp16)[name = tensor<string, []>("op_47092_cast_fp16")];
            tensor<int32, [4]> var_47096_begin_0 = const()[name = tensor<string, []>("op_47096_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_47096_end_0 = const()[name = tensor<string, []>("op_47096_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_47096_end_mask_0 = const()[name = tensor<string, []>("op_47096_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_47096_cast_fp16 = slice_by_index(begin = var_47096_begin_0, end = var_47096_end_0, end_mask = var_47096_end_mask_0, x = k_419_cast_fp16)[name = tensor<string, []>("op_47096_cast_fp16")];
            tensor<int32, [4]> var_47100_begin_0 = const()[name = tensor<string, []>("op_47100_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_47100_end_0 = const()[name = tensor<string, []>("op_47100_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_47100_end_mask_0 = const()[name = tensor<string, []>("op_47100_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_47100_cast_fp16 = slice_by_index(begin = var_47100_begin_0, end = var_47100_end_0, end_mask = var_47100_end_mask_0, x = k_419_cast_fp16)[name = tensor<string, []>("op_47100_cast_fp16")];
            tensor<int32, [4]> var_47104_begin_0 = const()[name = tensor<string, []>("op_47104_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_47104_end_0 = const()[name = tensor<string, []>("op_47104_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_47104_end_mask_0 = const()[name = tensor<string, []>("op_47104_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_47104_cast_fp16 = slice_by_index(begin = var_47104_begin_0, end = var_47104_end_0, end_mask = var_47104_end_mask_0, x = k_419_cast_fp16)[name = tensor<string, []>("op_47104_cast_fp16")];
            tensor<int32, [4]> var_47108_begin_0 = const()[name = tensor<string, []>("op_47108_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_47108_end_0 = const()[name = tensor<string, []>("op_47108_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_47108_end_mask_0 = const()[name = tensor<string, []>("op_47108_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_47108_cast_fp16 = slice_by_index(begin = var_47108_begin_0, end = var_47108_end_0, end_mask = var_47108_end_mask_0, x = k_419_cast_fp16)[name = tensor<string, []>("op_47108_cast_fp16")];
            tensor<int32, [4]> var_47112_begin_0 = const()[name = tensor<string, []>("op_47112_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_47112_end_0 = const()[name = tensor<string, []>("op_47112_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_47112_end_mask_0 = const()[name = tensor<string, []>("op_47112_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_47112_cast_fp16 = slice_by_index(begin = var_47112_begin_0, end = var_47112_end_0, end_mask = var_47112_end_mask_0, x = k_419_cast_fp16)[name = tensor<string, []>("op_47112_cast_fp16")];
            tensor<int32, [4]> var_47116_begin_0 = const()[name = tensor<string, []>("op_47116_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_47116_end_0 = const()[name = tensor<string, []>("op_47116_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_47116_end_mask_0 = const()[name = tensor<string, []>("op_47116_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_47116_cast_fp16 = slice_by_index(begin = var_47116_begin_0, end = var_47116_end_0, end_mask = var_47116_end_mask_0, x = k_419_cast_fp16)[name = tensor<string, []>("op_47116_cast_fp16")];
            tensor<int32, [4]> var_47120_begin_0 = const()[name = tensor<string, []>("op_47120_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_47120_end_0 = const()[name = tensor<string, []>("op_47120_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_47120_end_mask_0 = const()[name = tensor<string, []>("op_47120_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_47120_cast_fp16 = slice_by_index(begin = var_47120_begin_0, end = var_47120_end_0, end_mask = var_47120_end_mask_0, x = k_419_cast_fp16)[name = tensor<string, []>("op_47120_cast_fp16")];
            tensor<int32, [4]> var_47124_begin_0 = const()[name = tensor<string, []>("op_47124_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_47124_end_0 = const()[name = tensor<string, []>("op_47124_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_47124_end_mask_0 = const()[name = tensor<string, []>("op_47124_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_47124_cast_fp16 = slice_by_index(begin = var_47124_begin_0, end = var_47124_end_0, end_mask = var_47124_end_mask_0, x = k_419_cast_fp16)[name = tensor<string, []>("op_47124_cast_fp16")];
            tensor<int32, [4]> var_47128_begin_0 = const()[name = tensor<string, []>("op_47128_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_47128_end_0 = const()[name = tensor<string, []>("op_47128_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_47128_end_mask_0 = const()[name = tensor<string, []>("op_47128_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_47128_cast_fp16 = slice_by_index(begin = var_47128_begin_0, end = var_47128_end_0, end_mask = var_47128_end_mask_0, x = k_419_cast_fp16)[name = tensor<string, []>("op_47128_cast_fp16")];
            tensor<int32, [4]> var_47132_begin_0 = const()[name = tensor<string, []>("op_47132_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_47132_end_0 = const()[name = tensor<string, []>("op_47132_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_47132_end_mask_0 = const()[name = tensor<string, []>("op_47132_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_47132_cast_fp16 = slice_by_index(begin = var_47132_begin_0, end = var_47132_end_0, end_mask = var_47132_end_mask_0, x = k_419_cast_fp16)[name = tensor<string, []>("op_47132_cast_fp16")];
            tensor<int32, [4]> var_47136_begin_0 = const()[name = tensor<string, []>("op_47136_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_47136_end_0 = const()[name = tensor<string, []>("op_47136_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_47136_end_mask_0 = const()[name = tensor<string, []>("op_47136_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_47136_cast_fp16 = slice_by_index(begin = var_47136_begin_0, end = var_47136_end_0, end_mask = var_47136_end_mask_0, x = k_419_cast_fp16)[name = tensor<string, []>("op_47136_cast_fp16")];
            tensor<int32, [4]> var_47140_begin_0 = const()[name = tensor<string, []>("op_47140_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_47140_end_0 = const()[name = tensor<string, []>("op_47140_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_47140_end_mask_0 = const()[name = tensor<string, []>("op_47140_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_47140_cast_fp16 = slice_by_index(begin = var_47140_begin_0, end = var_47140_end_0, end_mask = var_47140_end_mask_0, x = k_419_cast_fp16)[name = tensor<string, []>("op_47140_cast_fp16")];
            tensor<int32, [4]> var_47144_begin_0 = const()[name = tensor<string, []>("op_47144_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_47144_end_0 = const()[name = tensor<string, []>("op_47144_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_47144_end_mask_0 = const()[name = tensor<string, []>("op_47144_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_47144_cast_fp16 = slice_by_index(begin = var_47144_begin_0, end = var_47144_end_0, end_mask = var_47144_end_mask_0, x = k_419_cast_fp16)[name = tensor<string, []>("op_47144_cast_fp16")];
            tensor<int32, [4]> var_47148_begin_0 = const()[name = tensor<string, []>("op_47148_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_47148_end_0 = const()[name = tensor<string, []>("op_47148_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_47148_end_mask_0 = const()[name = tensor<string, []>("op_47148_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_47148_cast_fp16 = slice_by_index(begin = var_47148_begin_0, end = var_47148_end_0, end_mask = var_47148_end_mask_0, x = k_419_cast_fp16)[name = tensor<string, []>("op_47148_cast_fp16")];
            tensor<int32, [4]> var_47152_begin_0 = const()[name = tensor<string, []>("op_47152_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_47152_end_0 = const()[name = tensor<string, []>("op_47152_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_47152_end_mask_0 = const()[name = tensor<string, []>("op_47152_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_47152_cast_fp16 = slice_by_index(begin = var_47152_begin_0, end = var_47152_end_0, end_mask = var_47152_end_mask_0, x = k_419_cast_fp16)[name = tensor<string, []>("op_47152_cast_fp16")];
            tensor<int32, [4]> var_47156_begin_0 = const()[name = tensor<string, []>("op_47156_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_47156_end_0 = const()[name = tensor<string, []>("op_47156_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_47156_end_mask_0 = const()[name = tensor<string, []>("op_47156_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_47156_cast_fp16 = slice_by_index(begin = var_47156_begin_0, end = var_47156_end_0, end_mask = var_47156_end_mask_0, x = k_419_cast_fp16)[name = tensor<string, []>("op_47156_cast_fp16")];
            tensor<int32, [4]> var_47160_begin_0 = const()[name = tensor<string, []>("op_47160_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_47160_end_0 = const()[name = tensor<string, []>("op_47160_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_47160_end_mask_0 = const()[name = tensor<string, []>("op_47160_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_47160_cast_fp16 = slice_by_index(begin = var_47160_begin_0, end = var_47160_end_0, end_mask = var_47160_end_mask_0, x = k_419_cast_fp16)[name = tensor<string, []>("op_47160_cast_fp16")];
            tensor<int32, [4]> var_47164_begin_0 = const()[name = tensor<string, []>("op_47164_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_47164_end_0 = const()[name = tensor<string, []>("op_47164_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_47164_end_mask_0 = const()[name = tensor<string, []>("op_47164_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_47164_cast_fp16 = slice_by_index(begin = var_47164_begin_0, end = var_47164_end_0, end_mask = var_47164_end_mask_0, x = k_419_cast_fp16)[name = tensor<string, []>("op_47164_cast_fp16")];
            tensor<int32, [4]> var_47166_begin_0 = const()[name = tensor<string, []>("op_47166_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_47166_end_0 = const()[name = tensor<string, []>("op_47166_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_47166_end_mask_0 = const()[name = tensor<string, []>("op_47166_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47166_cast_fp16 = slice_by_index(begin = var_47166_begin_0, end = var_47166_end_0, end_mask = var_47166_end_mask_0, x = v_209_cast_fp16)[name = tensor<string, []>("op_47166_cast_fp16")];
            tensor<int32, [4]> var_47170_begin_0 = const()[name = tensor<string, []>("op_47170_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_47170_end_0 = const()[name = tensor<string, []>("op_47170_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_47170_end_mask_0 = const()[name = tensor<string, []>("op_47170_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47170_cast_fp16 = slice_by_index(begin = var_47170_begin_0, end = var_47170_end_0, end_mask = var_47170_end_mask_0, x = v_209_cast_fp16)[name = tensor<string, []>("op_47170_cast_fp16")];
            tensor<int32, [4]> var_47174_begin_0 = const()[name = tensor<string, []>("op_47174_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_47174_end_0 = const()[name = tensor<string, []>("op_47174_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_47174_end_mask_0 = const()[name = tensor<string, []>("op_47174_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47174_cast_fp16 = slice_by_index(begin = var_47174_begin_0, end = var_47174_end_0, end_mask = var_47174_end_mask_0, x = v_209_cast_fp16)[name = tensor<string, []>("op_47174_cast_fp16")];
            tensor<int32, [4]> var_47178_begin_0 = const()[name = tensor<string, []>("op_47178_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_47178_end_0 = const()[name = tensor<string, []>("op_47178_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_47178_end_mask_0 = const()[name = tensor<string, []>("op_47178_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47178_cast_fp16 = slice_by_index(begin = var_47178_begin_0, end = var_47178_end_0, end_mask = var_47178_end_mask_0, x = v_209_cast_fp16)[name = tensor<string, []>("op_47178_cast_fp16")];
            tensor<int32, [4]> var_47182_begin_0 = const()[name = tensor<string, []>("op_47182_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_47182_end_0 = const()[name = tensor<string, []>("op_47182_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_47182_end_mask_0 = const()[name = tensor<string, []>("op_47182_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47182_cast_fp16 = slice_by_index(begin = var_47182_begin_0, end = var_47182_end_0, end_mask = var_47182_end_mask_0, x = v_209_cast_fp16)[name = tensor<string, []>("op_47182_cast_fp16")];
            tensor<int32, [4]> var_47186_begin_0 = const()[name = tensor<string, []>("op_47186_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_47186_end_0 = const()[name = tensor<string, []>("op_47186_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_47186_end_mask_0 = const()[name = tensor<string, []>("op_47186_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47186_cast_fp16 = slice_by_index(begin = var_47186_begin_0, end = var_47186_end_0, end_mask = var_47186_end_mask_0, x = v_209_cast_fp16)[name = tensor<string, []>("op_47186_cast_fp16")];
            tensor<int32, [4]> var_47190_begin_0 = const()[name = tensor<string, []>("op_47190_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_47190_end_0 = const()[name = tensor<string, []>("op_47190_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_47190_end_mask_0 = const()[name = tensor<string, []>("op_47190_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47190_cast_fp16 = slice_by_index(begin = var_47190_begin_0, end = var_47190_end_0, end_mask = var_47190_end_mask_0, x = v_209_cast_fp16)[name = tensor<string, []>("op_47190_cast_fp16")];
            tensor<int32, [4]> var_47194_begin_0 = const()[name = tensor<string, []>("op_47194_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_47194_end_0 = const()[name = tensor<string, []>("op_47194_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_47194_end_mask_0 = const()[name = tensor<string, []>("op_47194_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47194_cast_fp16 = slice_by_index(begin = var_47194_begin_0, end = var_47194_end_0, end_mask = var_47194_end_mask_0, x = v_209_cast_fp16)[name = tensor<string, []>("op_47194_cast_fp16")];
            tensor<int32, [4]> var_47198_begin_0 = const()[name = tensor<string, []>("op_47198_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_47198_end_0 = const()[name = tensor<string, []>("op_47198_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_47198_end_mask_0 = const()[name = tensor<string, []>("op_47198_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47198_cast_fp16 = slice_by_index(begin = var_47198_begin_0, end = var_47198_end_0, end_mask = var_47198_end_mask_0, x = v_209_cast_fp16)[name = tensor<string, []>("op_47198_cast_fp16")];
            tensor<int32, [4]> var_47202_begin_0 = const()[name = tensor<string, []>("op_47202_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_47202_end_0 = const()[name = tensor<string, []>("op_47202_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_47202_end_mask_0 = const()[name = tensor<string, []>("op_47202_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47202_cast_fp16 = slice_by_index(begin = var_47202_begin_0, end = var_47202_end_0, end_mask = var_47202_end_mask_0, x = v_209_cast_fp16)[name = tensor<string, []>("op_47202_cast_fp16")];
            tensor<int32, [4]> var_47206_begin_0 = const()[name = tensor<string, []>("op_47206_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_47206_end_0 = const()[name = tensor<string, []>("op_47206_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_47206_end_mask_0 = const()[name = tensor<string, []>("op_47206_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47206_cast_fp16 = slice_by_index(begin = var_47206_begin_0, end = var_47206_end_0, end_mask = var_47206_end_mask_0, x = v_209_cast_fp16)[name = tensor<string, []>("op_47206_cast_fp16")];
            tensor<int32, [4]> var_47210_begin_0 = const()[name = tensor<string, []>("op_47210_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_47210_end_0 = const()[name = tensor<string, []>("op_47210_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_47210_end_mask_0 = const()[name = tensor<string, []>("op_47210_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47210_cast_fp16 = slice_by_index(begin = var_47210_begin_0, end = var_47210_end_0, end_mask = var_47210_end_mask_0, x = v_209_cast_fp16)[name = tensor<string, []>("op_47210_cast_fp16")];
            tensor<int32, [4]> var_47214_begin_0 = const()[name = tensor<string, []>("op_47214_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_47214_end_0 = const()[name = tensor<string, []>("op_47214_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_47214_end_mask_0 = const()[name = tensor<string, []>("op_47214_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47214_cast_fp16 = slice_by_index(begin = var_47214_begin_0, end = var_47214_end_0, end_mask = var_47214_end_mask_0, x = v_209_cast_fp16)[name = tensor<string, []>("op_47214_cast_fp16")];
            tensor<int32, [4]> var_47218_begin_0 = const()[name = tensor<string, []>("op_47218_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_47218_end_0 = const()[name = tensor<string, []>("op_47218_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_47218_end_mask_0 = const()[name = tensor<string, []>("op_47218_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47218_cast_fp16 = slice_by_index(begin = var_47218_begin_0, end = var_47218_end_0, end_mask = var_47218_end_mask_0, x = v_209_cast_fp16)[name = tensor<string, []>("op_47218_cast_fp16")];
            tensor<int32, [4]> var_47222_begin_0 = const()[name = tensor<string, []>("op_47222_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_47222_end_0 = const()[name = tensor<string, []>("op_47222_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_47222_end_mask_0 = const()[name = tensor<string, []>("op_47222_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47222_cast_fp16 = slice_by_index(begin = var_47222_begin_0, end = var_47222_end_0, end_mask = var_47222_end_mask_0, x = v_209_cast_fp16)[name = tensor<string, []>("op_47222_cast_fp16")];
            tensor<int32, [4]> var_47226_begin_0 = const()[name = tensor<string, []>("op_47226_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_47226_end_0 = const()[name = tensor<string, []>("op_47226_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_47226_end_mask_0 = const()[name = tensor<string, []>("op_47226_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47226_cast_fp16 = slice_by_index(begin = var_47226_begin_0, end = var_47226_end_0, end_mask = var_47226_end_mask_0, x = v_209_cast_fp16)[name = tensor<string, []>("op_47226_cast_fp16")];
            tensor<int32, [4]> var_47230_begin_0 = const()[name = tensor<string, []>("op_47230_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_47230_end_0 = const()[name = tensor<string, []>("op_47230_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_47230_end_mask_0 = const()[name = tensor<string, []>("op_47230_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47230_cast_fp16 = slice_by_index(begin = var_47230_begin_0, end = var_47230_end_0, end_mask = var_47230_end_mask_0, x = v_209_cast_fp16)[name = tensor<string, []>("op_47230_cast_fp16")];
            tensor<int32, [4]> var_47234_begin_0 = const()[name = tensor<string, []>("op_47234_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_47234_end_0 = const()[name = tensor<string, []>("op_47234_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_47234_end_mask_0 = const()[name = tensor<string, []>("op_47234_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47234_cast_fp16 = slice_by_index(begin = var_47234_begin_0, end = var_47234_end_0, end_mask = var_47234_end_mask_0, x = v_209_cast_fp16)[name = tensor<string, []>("op_47234_cast_fp16")];
            tensor<int32, [4]> var_47238_begin_0 = const()[name = tensor<string, []>("op_47238_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_47238_end_0 = const()[name = tensor<string, []>("op_47238_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_47238_end_mask_0 = const()[name = tensor<string, []>("op_47238_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47238_cast_fp16 = slice_by_index(begin = var_47238_begin_0, end = var_47238_end_0, end_mask = var_47238_end_mask_0, x = v_209_cast_fp16)[name = tensor<string, []>("op_47238_cast_fp16")];
            tensor<int32, [4]> var_47242_begin_0 = const()[name = tensor<string, []>("op_47242_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_47242_end_0 = const()[name = tensor<string, []>("op_47242_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_47242_end_mask_0 = const()[name = tensor<string, []>("op_47242_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47242_cast_fp16 = slice_by_index(begin = var_47242_begin_0, end = var_47242_end_0, end_mask = var_47242_end_mask_0, x = v_209_cast_fp16)[name = tensor<string, []>("op_47242_cast_fp16")];
            tensor<string, []> var_47246_equation_0 = const()[name = tensor<string, []>("op_47246_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47246_cast_fp16 = einsum(equation = var_47246_equation_0, values = (var_47088_cast_fp16, var_47005_cast_fp16))[name = tensor<string, []>("op_47246_cast_fp16")];
            tensor<fp16, []> var_47247_to_fp16 = const()[name = tensor<string, []>("op_47247_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4001_cast_fp16 = mul(x = var_47246_cast_fp16, y = var_47247_to_fp16)[name = tensor<string, []>("aw_4001_cast_fp16")];
            tensor<string, []> var_47250_equation_0 = const()[name = tensor<string, []>("op_47250_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47250_cast_fp16 = einsum(equation = var_47250_equation_0, values = (var_47092_cast_fp16, var_47009_cast_fp16))[name = tensor<string, []>("op_47250_cast_fp16")];
            tensor<fp16, []> var_47251_to_fp16 = const()[name = tensor<string, []>("op_47251_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4003_cast_fp16 = mul(x = var_47250_cast_fp16, y = var_47251_to_fp16)[name = tensor<string, []>("aw_4003_cast_fp16")];
            tensor<string, []> var_47254_equation_0 = const()[name = tensor<string, []>("op_47254_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47254_cast_fp16 = einsum(equation = var_47254_equation_0, values = (var_47096_cast_fp16, var_47013_cast_fp16))[name = tensor<string, []>("op_47254_cast_fp16")];
            tensor<fp16, []> var_47255_to_fp16 = const()[name = tensor<string, []>("op_47255_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4005_cast_fp16 = mul(x = var_47254_cast_fp16, y = var_47255_to_fp16)[name = tensor<string, []>("aw_4005_cast_fp16")];
            tensor<string, []> var_47258_equation_0 = const()[name = tensor<string, []>("op_47258_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47258_cast_fp16 = einsum(equation = var_47258_equation_0, values = (var_47100_cast_fp16, var_47017_cast_fp16))[name = tensor<string, []>("op_47258_cast_fp16")];
            tensor<fp16, []> var_47259_to_fp16 = const()[name = tensor<string, []>("op_47259_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4007_cast_fp16 = mul(x = var_47258_cast_fp16, y = var_47259_to_fp16)[name = tensor<string, []>("aw_4007_cast_fp16")];
            tensor<string, []> var_47262_equation_0 = const()[name = tensor<string, []>("op_47262_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47262_cast_fp16 = einsum(equation = var_47262_equation_0, values = (var_47104_cast_fp16, var_47021_cast_fp16))[name = tensor<string, []>("op_47262_cast_fp16")];
            tensor<fp16, []> var_47263_to_fp16 = const()[name = tensor<string, []>("op_47263_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4009_cast_fp16 = mul(x = var_47262_cast_fp16, y = var_47263_to_fp16)[name = tensor<string, []>("aw_4009_cast_fp16")];
            tensor<string, []> var_47266_equation_0 = const()[name = tensor<string, []>("op_47266_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47266_cast_fp16 = einsum(equation = var_47266_equation_0, values = (var_47108_cast_fp16, var_47025_cast_fp16))[name = tensor<string, []>("op_47266_cast_fp16")];
            tensor<fp16, []> var_47267_to_fp16 = const()[name = tensor<string, []>("op_47267_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4011_cast_fp16 = mul(x = var_47266_cast_fp16, y = var_47267_to_fp16)[name = tensor<string, []>("aw_4011_cast_fp16")];
            tensor<string, []> var_47270_equation_0 = const()[name = tensor<string, []>("op_47270_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47270_cast_fp16 = einsum(equation = var_47270_equation_0, values = (var_47112_cast_fp16, var_47029_cast_fp16))[name = tensor<string, []>("op_47270_cast_fp16")];
            tensor<fp16, []> var_47271_to_fp16 = const()[name = tensor<string, []>("op_47271_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4013_cast_fp16 = mul(x = var_47270_cast_fp16, y = var_47271_to_fp16)[name = tensor<string, []>("aw_4013_cast_fp16")];
            tensor<string, []> var_47274_equation_0 = const()[name = tensor<string, []>("op_47274_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47274_cast_fp16 = einsum(equation = var_47274_equation_0, values = (var_47116_cast_fp16, var_47033_cast_fp16))[name = tensor<string, []>("op_47274_cast_fp16")];
            tensor<fp16, []> var_47275_to_fp16 = const()[name = tensor<string, []>("op_47275_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4015_cast_fp16 = mul(x = var_47274_cast_fp16, y = var_47275_to_fp16)[name = tensor<string, []>("aw_4015_cast_fp16")];
            tensor<string, []> var_47278_equation_0 = const()[name = tensor<string, []>("op_47278_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47278_cast_fp16 = einsum(equation = var_47278_equation_0, values = (var_47120_cast_fp16, var_47037_cast_fp16))[name = tensor<string, []>("op_47278_cast_fp16")];
            tensor<fp16, []> var_47279_to_fp16 = const()[name = tensor<string, []>("op_47279_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4017_cast_fp16 = mul(x = var_47278_cast_fp16, y = var_47279_to_fp16)[name = tensor<string, []>("aw_4017_cast_fp16")];
            tensor<string, []> var_47282_equation_0 = const()[name = tensor<string, []>("op_47282_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47282_cast_fp16 = einsum(equation = var_47282_equation_0, values = (var_47124_cast_fp16, var_47041_cast_fp16))[name = tensor<string, []>("op_47282_cast_fp16")];
            tensor<fp16, []> var_47283_to_fp16 = const()[name = tensor<string, []>("op_47283_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4019_cast_fp16 = mul(x = var_47282_cast_fp16, y = var_47283_to_fp16)[name = tensor<string, []>("aw_4019_cast_fp16")];
            tensor<string, []> var_47286_equation_0 = const()[name = tensor<string, []>("op_47286_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47286_cast_fp16 = einsum(equation = var_47286_equation_0, values = (var_47128_cast_fp16, var_47045_cast_fp16))[name = tensor<string, []>("op_47286_cast_fp16")];
            tensor<fp16, []> var_47287_to_fp16 = const()[name = tensor<string, []>("op_47287_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4021_cast_fp16 = mul(x = var_47286_cast_fp16, y = var_47287_to_fp16)[name = tensor<string, []>("aw_4021_cast_fp16")];
            tensor<string, []> var_47290_equation_0 = const()[name = tensor<string, []>("op_47290_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47290_cast_fp16 = einsum(equation = var_47290_equation_0, values = (var_47132_cast_fp16, var_47049_cast_fp16))[name = tensor<string, []>("op_47290_cast_fp16")];
            tensor<fp16, []> var_47291_to_fp16 = const()[name = tensor<string, []>("op_47291_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4023_cast_fp16 = mul(x = var_47290_cast_fp16, y = var_47291_to_fp16)[name = tensor<string, []>("aw_4023_cast_fp16")];
            tensor<string, []> var_47294_equation_0 = const()[name = tensor<string, []>("op_47294_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47294_cast_fp16 = einsum(equation = var_47294_equation_0, values = (var_47136_cast_fp16, var_47053_cast_fp16))[name = tensor<string, []>("op_47294_cast_fp16")];
            tensor<fp16, []> var_47295_to_fp16 = const()[name = tensor<string, []>("op_47295_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4025_cast_fp16 = mul(x = var_47294_cast_fp16, y = var_47295_to_fp16)[name = tensor<string, []>("aw_4025_cast_fp16")];
            tensor<string, []> var_47298_equation_0 = const()[name = tensor<string, []>("op_47298_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47298_cast_fp16 = einsum(equation = var_47298_equation_0, values = (var_47140_cast_fp16, var_47057_cast_fp16))[name = tensor<string, []>("op_47298_cast_fp16")];
            tensor<fp16, []> var_47299_to_fp16 = const()[name = tensor<string, []>("op_47299_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4027_cast_fp16 = mul(x = var_47298_cast_fp16, y = var_47299_to_fp16)[name = tensor<string, []>("aw_4027_cast_fp16")];
            tensor<string, []> var_47302_equation_0 = const()[name = tensor<string, []>("op_47302_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47302_cast_fp16 = einsum(equation = var_47302_equation_0, values = (var_47144_cast_fp16, var_47061_cast_fp16))[name = tensor<string, []>("op_47302_cast_fp16")];
            tensor<fp16, []> var_47303_to_fp16 = const()[name = tensor<string, []>("op_47303_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4029_cast_fp16 = mul(x = var_47302_cast_fp16, y = var_47303_to_fp16)[name = tensor<string, []>("aw_4029_cast_fp16")];
            tensor<string, []> var_47306_equation_0 = const()[name = tensor<string, []>("op_47306_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47306_cast_fp16 = einsum(equation = var_47306_equation_0, values = (var_47148_cast_fp16, var_47065_cast_fp16))[name = tensor<string, []>("op_47306_cast_fp16")];
            tensor<fp16, []> var_47307_to_fp16 = const()[name = tensor<string, []>("op_47307_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4031_cast_fp16 = mul(x = var_47306_cast_fp16, y = var_47307_to_fp16)[name = tensor<string, []>("aw_4031_cast_fp16")];
            tensor<string, []> var_47310_equation_0 = const()[name = tensor<string, []>("op_47310_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47310_cast_fp16 = einsum(equation = var_47310_equation_0, values = (var_47152_cast_fp16, var_47069_cast_fp16))[name = tensor<string, []>("op_47310_cast_fp16")];
            tensor<fp16, []> var_47311_to_fp16 = const()[name = tensor<string, []>("op_47311_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4033_cast_fp16 = mul(x = var_47310_cast_fp16, y = var_47311_to_fp16)[name = tensor<string, []>("aw_4033_cast_fp16")];
            tensor<string, []> var_47314_equation_0 = const()[name = tensor<string, []>("op_47314_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47314_cast_fp16 = einsum(equation = var_47314_equation_0, values = (var_47156_cast_fp16, var_47073_cast_fp16))[name = tensor<string, []>("op_47314_cast_fp16")];
            tensor<fp16, []> var_47315_to_fp16 = const()[name = tensor<string, []>("op_47315_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4035_cast_fp16 = mul(x = var_47314_cast_fp16, y = var_47315_to_fp16)[name = tensor<string, []>("aw_4035_cast_fp16")];
            tensor<string, []> var_47318_equation_0 = const()[name = tensor<string, []>("op_47318_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47318_cast_fp16 = einsum(equation = var_47318_equation_0, values = (var_47160_cast_fp16, var_47077_cast_fp16))[name = tensor<string, []>("op_47318_cast_fp16")];
            tensor<fp16, []> var_47319_to_fp16 = const()[name = tensor<string, []>("op_47319_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4037_cast_fp16 = mul(x = var_47318_cast_fp16, y = var_47319_to_fp16)[name = tensor<string, []>("aw_4037_cast_fp16")];
            tensor<string, []> var_47322_equation_0 = const()[name = tensor<string, []>("op_47322_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47322_cast_fp16 = einsum(equation = var_47322_equation_0, values = (var_47164_cast_fp16, var_47081_cast_fp16))[name = tensor<string, []>("op_47322_cast_fp16")];
            tensor<fp16, []> var_47323_to_fp16 = const()[name = tensor<string, []>("op_47323_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4039_cast_fp16 = mul(x = var_47322_cast_fp16, y = var_47323_to_fp16)[name = tensor<string, []>("aw_4039_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47325_cast_fp16 = softmax(axis = var_30355, x = aw_4001_cast_fp16)[name = tensor<string, []>("op_47325_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47326_cast_fp16 = softmax(axis = var_30355, x = aw_4003_cast_fp16)[name = tensor<string, []>("op_47326_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47327_cast_fp16 = softmax(axis = var_30355, x = aw_4005_cast_fp16)[name = tensor<string, []>("op_47327_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47328_cast_fp16 = softmax(axis = var_30355, x = aw_4007_cast_fp16)[name = tensor<string, []>("op_47328_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47329_cast_fp16 = softmax(axis = var_30355, x = aw_4009_cast_fp16)[name = tensor<string, []>("op_47329_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47330_cast_fp16 = softmax(axis = var_30355, x = aw_4011_cast_fp16)[name = tensor<string, []>("op_47330_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47331_cast_fp16 = softmax(axis = var_30355, x = aw_4013_cast_fp16)[name = tensor<string, []>("op_47331_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47332_cast_fp16 = softmax(axis = var_30355, x = aw_4015_cast_fp16)[name = tensor<string, []>("op_47332_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47333_cast_fp16 = softmax(axis = var_30355, x = aw_4017_cast_fp16)[name = tensor<string, []>("op_47333_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47334_cast_fp16 = softmax(axis = var_30355, x = aw_4019_cast_fp16)[name = tensor<string, []>("op_47334_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47335_cast_fp16 = softmax(axis = var_30355, x = aw_4021_cast_fp16)[name = tensor<string, []>("op_47335_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47336_cast_fp16 = softmax(axis = var_30355, x = aw_4023_cast_fp16)[name = tensor<string, []>("op_47336_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47337_cast_fp16 = softmax(axis = var_30355, x = aw_4025_cast_fp16)[name = tensor<string, []>("op_47337_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47338_cast_fp16 = softmax(axis = var_30355, x = aw_4027_cast_fp16)[name = tensor<string, []>("op_47338_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47339_cast_fp16 = softmax(axis = var_30355, x = aw_4029_cast_fp16)[name = tensor<string, []>("op_47339_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47340_cast_fp16 = softmax(axis = var_30355, x = aw_4031_cast_fp16)[name = tensor<string, []>("op_47340_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47341_cast_fp16 = softmax(axis = var_30355, x = aw_4033_cast_fp16)[name = tensor<string, []>("op_47341_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47342_cast_fp16 = softmax(axis = var_30355, x = aw_4035_cast_fp16)[name = tensor<string, []>("op_47342_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47343_cast_fp16 = softmax(axis = var_30355, x = aw_4037_cast_fp16)[name = tensor<string, []>("op_47343_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47344_cast_fp16 = softmax(axis = var_30355, x = aw_4039_cast_fp16)[name = tensor<string, []>("op_47344_cast_fp16")];
            tensor<string, []> var_47346_equation_0 = const()[name = tensor<string, []>("op_47346_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47346_cast_fp16 = einsum(equation = var_47346_equation_0, values = (var_47166_cast_fp16, var_47325_cast_fp16))[name = tensor<string, []>("op_47346_cast_fp16")];
            tensor<string, []> var_47348_equation_0 = const()[name = tensor<string, []>("op_47348_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47348_cast_fp16 = einsum(equation = var_47348_equation_0, values = (var_47170_cast_fp16, var_47326_cast_fp16))[name = tensor<string, []>("op_47348_cast_fp16")];
            tensor<string, []> var_47350_equation_0 = const()[name = tensor<string, []>("op_47350_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47350_cast_fp16 = einsum(equation = var_47350_equation_0, values = (var_47174_cast_fp16, var_47327_cast_fp16))[name = tensor<string, []>("op_47350_cast_fp16")];
            tensor<string, []> var_47352_equation_0 = const()[name = tensor<string, []>("op_47352_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47352_cast_fp16 = einsum(equation = var_47352_equation_0, values = (var_47178_cast_fp16, var_47328_cast_fp16))[name = tensor<string, []>("op_47352_cast_fp16")];
            tensor<string, []> var_47354_equation_0 = const()[name = tensor<string, []>("op_47354_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47354_cast_fp16 = einsum(equation = var_47354_equation_0, values = (var_47182_cast_fp16, var_47329_cast_fp16))[name = tensor<string, []>("op_47354_cast_fp16")];
            tensor<string, []> var_47356_equation_0 = const()[name = tensor<string, []>("op_47356_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47356_cast_fp16 = einsum(equation = var_47356_equation_0, values = (var_47186_cast_fp16, var_47330_cast_fp16))[name = tensor<string, []>("op_47356_cast_fp16")];
            tensor<string, []> var_47358_equation_0 = const()[name = tensor<string, []>("op_47358_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47358_cast_fp16 = einsum(equation = var_47358_equation_0, values = (var_47190_cast_fp16, var_47331_cast_fp16))[name = tensor<string, []>("op_47358_cast_fp16")];
            tensor<string, []> var_47360_equation_0 = const()[name = tensor<string, []>("op_47360_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47360_cast_fp16 = einsum(equation = var_47360_equation_0, values = (var_47194_cast_fp16, var_47332_cast_fp16))[name = tensor<string, []>("op_47360_cast_fp16")];
            tensor<string, []> var_47362_equation_0 = const()[name = tensor<string, []>("op_47362_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47362_cast_fp16 = einsum(equation = var_47362_equation_0, values = (var_47198_cast_fp16, var_47333_cast_fp16))[name = tensor<string, []>("op_47362_cast_fp16")];
            tensor<string, []> var_47364_equation_0 = const()[name = tensor<string, []>("op_47364_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47364_cast_fp16 = einsum(equation = var_47364_equation_0, values = (var_47202_cast_fp16, var_47334_cast_fp16))[name = tensor<string, []>("op_47364_cast_fp16")];
            tensor<string, []> var_47366_equation_0 = const()[name = tensor<string, []>("op_47366_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47366_cast_fp16 = einsum(equation = var_47366_equation_0, values = (var_47206_cast_fp16, var_47335_cast_fp16))[name = tensor<string, []>("op_47366_cast_fp16")];
            tensor<string, []> var_47368_equation_0 = const()[name = tensor<string, []>("op_47368_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47368_cast_fp16 = einsum(equation = var_47368_equation_0, values = (var_47210_cast_fp16, var_47336_cast_fp16))[name = tensor<string, []>("op_47368_cast_fp16")];
            tensor<string, []> var_47370_equation_0 = const()[name = tensor<string, []>("op_47370_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47370_cast_fp16 = einsum(equation = var_47370_equation_0, values = (var_47214_cast_fp16, var_47337_cast_fp16))[name = tensor<string, []>("op_47370_cast_fp16")];
            tensor<string, []> var_47372_equation_0 = const()[name = tensor<string, []>("op_47372_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47372_cast_fp16 = einsum(equation = var_47372_equation_0, values = (var_47218_cast_fp16, var_47338_cast_fp16))[name = tensor<string, []>("op_47372_cast_fp16")];
            tensor<string, []> var_47374_equation_0 = const()[name = tensor<string, []>("op_47374_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47374_cast_fp16 = einsum(equation = var_47374_equation_0, values = (var_47222_cast_fp16, var_47339_cast_fp16))[name = tensor<string, []>("op_47374_cast_fp16")];
            tensor<string, []> var_47376_equation_0 = const()[name = tensor<string, []>("op_47376_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47376_cast_fp16 = einsum(equation = var_47376_equation_0, values = (var_47226_cast_fp16, var_47340_cast_fp16))[name = tensor<string, []>("op_47376_cast_fp16")];
            tensor<string, []> var_47378_equation_0 = const()[name = tensor<string, []>("op_47378_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47378_cast_fp16 = einsum(equation = var_47378_equation_0, values = (var_47230_cast_fp16, var_47341_cast_fp16))[name = tensor<string, []>("op_47378_cast_fp16")];
            tensor<string, []> var_47380_equation_0 = const()[name = tensor<string, []>("op_47380_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47380_cast_fp16 = einsum(equation = var_47380_equation_0, values = (var_47234_cast_fp16, var_47342_cast_fp16))[name = tensor<string, []>("op_47380_cast_fp16")];
            tensor<string, []> var_47382_equation_0 = const()[name = tensor<string, []>("op_47382_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47382_cast_fp16 = einsum(equation = var_47382_equation_0, values = (var_47238_cast_fp16, var_47343_cast_fp16))[name = tensor<string, []>("op_47382_cast_fp16")];
            tensor<string, []> var_47384_equation_0 = const()[name = tensor<string, []>("op_47384_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47384_cast_fp16 = einsum(equation = var_47384_equation_0, values = (var_47242_cast_fp16, var_47344_cast_fp16))[name = tensor<string, []>("op_47384_cast_fp16")];
            tensor<bool, []> input_601_interleave_0 = const()[name = tensor<string, []>("input_601_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_601_cast_fp16 = concat(axis = var_30355, interleave = input_601_interleave_0, values = (var_47346_cast_fp16, var_47348_cast_fp16, var_47350_cast_fp16, var_47352_cast_fp16, var_47354_cast_fp16, var_47356_cast_fp16, var_47358_cast_fp16, var_47360_cast_fp16, var_47362_cast_fp16, var_47364_cast_fp16, var_47366_cast_fp16, var_47368_cast_fp16, var_47370_cast_fp16, var_47372_cast_fp16, var_47374_cast_fp16, var_47376_cast_fp16, var_47378_cast_fp16, var_47380_cast_fp16, var_47382_cast_fp16, var_47384_cast_fp16))[name = tensor<string, []>("input_601_cast_fp16")];
            tensor<string, []> var_47394_pad_type_0 = const()[name = tensor<string, []>("op_47394_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_47394_strides_0 = const()[name = tensor<string, []>("op_47394_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_47394_pad_0 = const()[name = tensor<string, []>("op_47394_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_47394_dilations_0 = const()[name = tensor<string, []>("op_47394_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_47394_groups_0 = const()[name = tensor<string, []>("op_47394_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_8_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(516520384))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(517749248))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_8_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_8_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_8_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(517749440)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_47394_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_8_attn1_to_out_0_bias_to_fp16, dilations = var_47394_dilations_0, groups = var_47394_groups_0, pad = var_47394_pad_0, pad_type = var_47394_pad_type_0, strides = var_47394_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_8_attn1_to_out_0_weight_to_fp16_palettized, x = input_601_cast_fp16)[name = tensor<string, []>("op_47394_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_315_cast_fp16 = add(x = var_47394_cast_fp16, y = inputs_313_cast_fp16)[name = tensor<string, []>("inputs_315_cast_fp16")];
            tensor<int32, [1]> hidden_states_411_axes_0 = const()[name = tensor<string, []>("hidden_states_411_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_411_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_411_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(517752064)))];
            tensor<fp16, [1280]> hidden_states_411_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_411_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(517754688)))];
            tensor<fp16, []> var_47404_to_fp16 = const()[name = tensor<string, []>("op_47404_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_411_cast_fp16 = layer_norm(axes = hidden_states_411_axes_0, beta = hidden_states_411_beta_0_to_fp16, epsilon = var_47404_to_fp16, gamma = hidden_states_411_gamma_0_to_fp16, x = inputs_315_cast_fp16)[name = tensor<string, []>("hidden_states_411_cast_fp16")];
            tensor<string, []> q_211_pad_type_0 = const()[name = tensor<string, []>("q_211_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_211_strides_0 = const()[name = tensor<string, []>("q_211_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_211_pad_0 = const()[name = tensor<string, []>("q_211_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_211_dilations_0 = const()[name = tensor<string, []>("q_211_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_211_groups_0 = const()[name = tensor<string, []>("q_211_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_8_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(517757312))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(518986176))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_8_attn2_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_211_cast_fp16 = conv(dilations = q_211_dilations_0, groups = q_211_groups_0, pad = q_211_pad_0, pad_type = q_211_pad_type_0, strides = q_211_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_8_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_411_cast_fp16)[name = tensor<string, []>("q_211_cast_fp16")];
            tensor<string, []> k_421_pad_type_0 = const()[name = tensor<string, []>("k_421_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_421_strides_0 = const()[name = tensor<string, []>("k_421_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_421_pad_0 = const()[name = tensor<string, []>("k_421_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_421_dilations_0 = const()[name = tensor<string, []>("k_421_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_421_groups_0 = const()[name = tensor<string, []>("k_421_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_8_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(518986368))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(520952512))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_8_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> k_421_cast_fp16 = conv(dilations = k_421_dilations_0, groups = k_421_groups_0, pad = k_421_pad_0, pad_type = k_421_pad_type_0, strides = k_421_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_8_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("k_421_cast_fp16")];
            tensor<string, []> v_211_pad_type_0 = const()[name = tensor<string, []>("v_211_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_211_strides_0 = const()[name = tensor<string, []>("v_211_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_211_pad_0 = const()[name = tensor<string, []>("v_211_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_211_dilations_0 = const()[name = tensor<string, []>("v_211_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_211_groups_0 = const()[name = tensor<string, []>("v_211_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_8_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(520952704))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(522918848))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_8_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> v_211_cast_fp16 = conv(dilations = v_211_dilations_0, groups = v_211_groups_0, pad = v_211_pad_0, pad_type = v_211_pad_type_0, strides = v_211_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_8_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("v_211_cast_fp16")];
            tensor<int32, [4]> var_47437_begin_0 = const()[name = tensor<string, []>("op_47437_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_47437_end_0 = const()[name = tensor<string, []>("op_47437_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_47437_end_mask_0 = const()[name = tensor<string, []>("op_47437_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47437_cast_fp16 = slice_by_index(begin = var_47437_begin_0, end = var_47437_end_0, end_mask = var_47437_end_mask_0, x = q_211_cast_fp16)[name = tensor<string, []>("op_47437_cast_fp16")];
            tensor<int32, [4]> var_47441_begin_0 = const()[name = tensor<string, []>("op_47441_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_47441_end_0 = const()[name = tensor<string, []>("op_47441_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_47441_end_mask_0 = const()[name = tensor<string, []>("op_47441_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47441_cast_fp16 = slice_by_index(begin = var_47441_begin_0, end = var_47441_end_0, end_mask = var_47441_end_mask_0, x = q_211_cast_fp16)[name = tensor<string, []>("op_47441_cast_fp16")];
            tensor<int32, [4]> var_47445_begin_0 = const()[name = tensor<string, []>("op_47445_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_47445_end_0 = const()[name = tensor<string, []>("op_47445_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_47445_end_mask_0 = const()[name = tensor<string, []>("op_47445_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47445_cast_fp16 = slice_by_index(begin = var_47445_begin_0, end = var_47445_end_0, end_mask = var_47445_end_mask_0, x = q_211_cast_fp16)[name = tensor<string, []>("op_47445_cast_fp16")];
            tensor<int32, [4]> var_47449_begin_0 = const()[name = tensor<string, []>("op_47449_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_47449_end_0 = const()[name = tensor<string, []>("op_47449_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_47449_end_mask_0 = const()[name = tensor<string, []>("op_47449_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47449_cast_fp16 = slice_by_index(begin = var_47449_begin_0, end = var_47449_end_0, end_mask = var_47449_end_mask_0, x = q_211_cast_fp16)[name = tensor<string, []>("op_47449_cast_fp16")];
            tensor<int32, [4]> var_47453_begin_0 = const()[name = tensor<string, []>("op_47453_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_47453_end_0 = const()[name = tensor<string, []>("op_47453_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_47453_end_mask_0 = const()[name = tensor<string, []>("op_47453_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47453_cast_fp16 = slice_by_index(begin = var_47453_begin_0, end = var_47453_end_0, end_mask = var_47453_end_mask_0, x = q_211_cast_fp16)[name = tensor<string, []>("op_47453_cast_fp16")];
            tensor<int32, [4]> var_47457_begin_0 = const()[name = tensor<string, []>("op_47457_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_47457_end_0 = const()[name = tensor<string, []>("op_47457_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_47457_end_mask_0 = const()[name = tensor<string, []>("op_47457_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47457_cast_fp16 = slice_by_index(begin = var_47457_begin_0, end = var_47457_end_0, end_mask = var_47457_end_mask_0, x = q_211_cast_fp16)[name = tensor<string, []>("op_47457_cast_fp16")];
            tensor<int32, [4]> var_47461_begin_0 = const()[name = tensor<string, []>("op_47461_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_47461_end_0 = const()[name = tensor<string, []>("op_47461_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_47461_end_mask_0 = const()[name = tensor<string, []>("op_47461_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47461_cast_fp16 = slice_by_index(begin = var_47461_begin_0, end = var_47461_end_0, end_mask = var_47461_end_mask_0, x = q_211_cast_fp16)[name = tensor<string, []>("op_47461_cast_fp16")];
            tensor<int32, [4]> var_47465_begin_0 = const()[name = tensor<string, []>("op_47465_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_47465_end_0 = const()[name = tensor<string, []>("op_47465_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_47465_end_mask_0 = const()[name = tensor<string, []>("op_47465_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47465_cast_fp16 = slice_by_index(begin = var_47465_begin_0, end = var_47465_end_0, end_mask = var_47465_end_mask_0, x = q_211_cast_fp16)[name = tensor<string, []>("op_47465_cast_fp16")];
            tensor<int32, [4]> var_47469_begin_0 = const()[name = tensor<string, []>("op_47469_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_47469_end_0 = const()[name = tensor<string, []>("op_47469_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_47469_end_mask_0 = const()[name = tensor<string, []>("op_47469_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47469_cast_fp16 = slice_by_index(begin = var_47469_begin_0, end = var_47469_end_0, end_mask = var_47469_end_mask_0, x = q_211_cast_fp16)[name = tensor<string, []>("op_47469_cast_fp16")];
            tensor<int32, [4]> var_47473_begin_0 = const()[name = tensor<string, []>("op_47473_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_47473_end_0 = const()[name = tensor<string, []>("op_47473_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_47473_end_mask_0 = const()[name = tensor<string, []>("op_47473_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47473_cast_fp16 = slice_by_index(begin = var_47473_begin_0, end = var_47473_end_0, end_mask = var_47473_end_mask_0, x = q_211_cast_fp16)[name = tensor<string, []>("op_47473_cast_fp16")];
            tensor<int32, [4]> var_47477_begin_0 = const()[name = tensor<string, []>("op_47477_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_47477_end_0 = const()[name = tensor<string, []>("op_47477_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_47477_end_mask_0 = const()[name = tensor<string, []>("op_47477_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47477_cast_fp16 = slice_by_index(begin = var_47477_begin_0, end = var_47477_end_0, end_mask = var_47477_end_mask_0, x = q_211_cast_fp16)[name = tensor<string, []>("op_47477_cast_fp16")];
            tensor<int32, [4]> var_47481_begin_0 = const()[name = tensor<string, []>("op_47481_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_47481_end_0 = const()[name = tensor<string, []>("op_47481_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_47481_end_mask_0 = const()[name = tensor<string, []>("op_47481_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47481_cast_fp16 = slice_by_index(begin = var_47481_begin_0, end = var_47481_end_0, end_mask = var_47481_end_mask_0, x = q_211_cast_fp16)[name = tensor<string, []>("op_47481_cast_fp16")];
            tensor<int32, [4]> var_47485_begin_0 = const()[name = tensor<string, []>("op_47485_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_47485_end_0 = const()[name = tensor<string, []>("op_47485_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_47485_end_mask_0 = const()[name = tensor<string, []>("op_47485_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47485_cast_fp16 = slice_by_index(begin = var_47485_begin_0, end = var_47485_end_0, end_mask = var_47485_end_mask_0, x = q_211_cast_fp16)[name = tensor<string, []>("op_47485_cast_fp16")];
            tensor<int32, [4]> var_47489_begin_0 = const()[name = tensor<string, []>("op_47489_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_47489_end_0 = const()[name = tensor<string, []>("op_47489_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_47489_end_mask_0 = const()[name = tensor<string, []>("op_47489_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47489_cast_fp16 = slice_by_index(begin = var_47489_begin_0, end = var_47489_end_0, end_mask = var_47489_end_mask_0, x = q_211_cast_fp16)[name = tensor<string, []>("op_47489_cast_fp16")];
            tensor<int32, [4]> var_47493_begin_0 = const()[name = tensor<string, []>("op_47493_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_47493_end_0 = const()[name = tensor<string, []>("op_47493_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_47493_end_mask_0 = const()[name = tensor<string, []>("op_47493_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47493_cast_fp16 = slice_by_index(begin = var_47493_begin_0, end = var_47493_end_0, end_mask = var_47493_end_mask_0, x = q_211_cast_fp16)[name = tensor<string, []>("op_47493_cast_fp16")];
            tensor<int32, [4]> var_47497_begin_0 = const()[name = tensor<string, []>("op_47497_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_47497_end_0 = const()[name = tensor<string, []>("op_47497_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_47497_end_mask_0 = const()[name = tensor<string, []>("op_47497_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47497_cast_fp16 = slice_by_index(begin = var_47497_begin_0, end = var_47497_end_0, end_mask = var_47497_end_mask_0, x = q_211_cast_fp16)[name = tensor<string, []>("op_47497_cast_fp16")];
            tensor<int32, [4]> var_47501_begin_0 = const()[name = tensor<string, []>("op_47501_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_47501_end_0 = const()[name = tensor<string, []>("op_47501_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_47501_end_mask_0 = const()[name = tensor<string, []>("op_47501_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47501_cast_fp16 = slice_by_index(begin = var_47501_begin_0, end = var_47501_end_0, end_mask = var_47501_end_mask_0, x = q_211_cast_fp16)[name = tensor<string, []>("op_47501_cast_fp16")];
            tensor<int32, [4]> var_47505_begin_0 = const()[name = tensor<string, []>("op_47505_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_47505_end_0 = const()[name = tensor<string, []>("op_47505_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_47505_end_mask_0 = const()[name = tensor<string, []>("op_47505_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47505_cast_fp16 = slice_by_index(begin = var_47505_begin_0, end = var_47505_end_0, end_mask = var_47505_end_mask_0, x = q_211_cast_fp16)[name = tensor<string, []>("op_47505_cast_fp16")];
            tensor<int32, [4]> var_47509_begin_0 = const()[name = tensor<string, []>("op_47509_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_47509_end_0 = const()[name = tensor<string, []>("op_47509_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_47509_end_mask_0 = const()[name = tensor<string, []>("op_47509_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47509_cast_fp16 = slice_by_index(begin = var_47509_begin_0, end = var_47509_end_0, end_mask = var_47509_end_mask_0, x = q_211_cast_fp16)[name = tensor<string, []>("op_47509_cast_fp16")];
            tensor<int32, [4]> var_47513_begin_0 = const()[name = tensor<string, []>("op_47513_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_47513_end_0 = const()[name = tensor<string, []>("op_47513_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_47513_end_mask_0 = const()[name = tensor<string, []>("op_47513_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47513_cast_fp16 = slice_by_index(begin = var_47513_begin_0, end = var_47513_end_0, end_mask = var_47513_end_mask_0, x = q_211_cast_fp16)[name = tensor<string, []>("op_47513_cast_fp16")];
            tensor<int32, [4]> k_423_perm_0 = const()[name = tensor<string, []>("k_423_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_47520_begin_0 = const()[name = tensor<string, []>("op_47520_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_47520_end_0 = const()[name = tensor<string, []>("op_47520_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_47520_end_mask_0 = const()[name = tensor<string, []>("op_47520_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> k_423_cast_fp16 = transpose(perm = k_423_perm_0, x = k_421_cast_fp16)[name = tensor<string, []>("transpose_34")];
            tensor<fp16, [2, 77, 1, 64]> var_47520_cast_fp16 = slice_by_index(begin = var_47520_begin_0, end = var_47520_end_0, end_mask = var_47520_end_mask_0, x = k_423_cast_fp16)[name = tensor<string, []>("op_47520_cast_fp16")];
            tensor<int32, [4]> var_47524_begin_0 = const()[name = tensor<string, []>("op_47524_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_47524_end_0 = const()[name = tensor<string, []>("op_47524_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_47524_end_mask_0 = const()[name = tensor<string, []>("op_47524_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_47524_cast_fp16 = slice_by_index(begin = var_47524_begin_0, end = var_47524_end_0, end_mask = var_47524_end_mask_0, x = k_423_cast_fp16)[name = tensor<string, []>("op_47524_cast_fp16")];
            tensor<int32, [4]> var_47528_begin_0 = const()[name = tensor<string, []>("op_47528_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_47528_end_0 = const()[name = tensor<string, []>("op_47528_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_47528_end_mask_0 = const()[name = tensor<string, []>("op_47528_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_47528_cast_fp16 = slice_by_index(begin = var_47528_begin_0, end = var_47528_end_0, end_mask = var_47528_end_mask_0, x = k_423_cast_fp16)[name = tensor<string, []>("op_47528_cast_fp16")];
            tensor<int32, [4]> var_47532_begin_0 = const()[name = tensor<string, []>("op_47532_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_47532_end_0 = const()[name = tensor<string, []>("op_47532_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_47532_end_mask_0 = const()[name = tensor<string, []>("op_47532_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_47532_cast_fp16 = slice_by_index(begin = var_47532_begin_0, end = var_47532_end_0, end_mask = var_47532_end_mask_0, x = k_423_cast_fp16)[name = tensor<string, []>("op_47532_cast_fp16")];
            tensor<int32, [4]> var_47536_begin_0 = const()[name = tensor<string, []>("op_47536_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_47536_end_0 = const()[name = tensor<string, []>("op_47536_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_47536_end_mask_0 = const()[name = tensor<string, []>("op_47536_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_47536_cast_fp16 = slice_by_index(begin = var_47536_begin_0, end = var_47536_end_0, end_mask = var_47536_end_mask_0, x = k_423_cast_fp16)[name = tensor<string, []>("op_47536_cast_fp16")];
            tensor<int32, [4]> var_47540_begin_0 = const()[name = tensor<string, []>("op_47540_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_47540_end_0 = const()[name = tensor<string, []>("op_47540_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_47540_end_mask_0 = const()[name = tensor<string, []>("op_47540_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_47540_cast_fp16 = slice_by_index(begin = var_47540_begin_0, end = var_47540_end_0, end_mask = var_47540_end_mask_0, x = k_423_cast_fp16)[name = tensor<string, []>("op_47540_cast_fp16")];
            tensor<int32, [4]> var_47544_begin_0 = const()[name = tensor<string, []>("op_47544_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_47544_end_0 = const()[name = tensor<string, []>("op_47544_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_47544_end_mask_0 = const()[name = tensor<string, []>("op_47544_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_47544_cast_fp16 = slice_by_index(begin = var_47544_begin_0, end = var_47544_end_0, end_mask = var_47544_end_mask_0, x = k_423_cast_fp16)[name = tensor<string, []>("op_47544_cast_fp16")];
            tensor<int32, [4]> var_47548_begin_0 = const()[name = tensor<string, []>("op_47548_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_47548_end_0 = const()[name = tensor<string, []>("op_47548_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_47548_end_mask_0 = const()[name = tensor<string, []>("op_47548_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_47548_cast_fp16 = slice_by_index(begin = var_47548_begin_0, end = var_47548_end_0, end_mask = var_47548_end_mask_0, x = k_423_cast_fp16)[name = tensor<string, []>("op_47548_cast_fp16")];
            tensor<int32, [4]> var_47552_begin_0 = const()[name = tensor<string, []>("op_47552_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_47552_end_0 = const()[name = tensor<string, []>("op_47552_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_47552_end_mask_0 = const()[name = tensor<string, []>("op_47552_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_47552_cast_fp16 = slice_by_index(begin = var_47552_begin_0, end = var_47552_end_0, end_mask = var_47552_end_mask_0, x = k_423_cast_fp16)[name = tensor<string, []>("op_47552_cast_fp16")];
            tensor<int32, [4]> var_47556_begin_0 = const()[name = tensor<string, []>("op_47556_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_47556_end_0 = const()[name = tensor<string, []>("op_47556_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_47556_end_mask_0 = const()[name = tensor<string, []>("op_47556_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_47556_cast_fp16 = slice_by_index(begin = var_47556_begin_0, end = var_47556_end_0, end_mask = var_47556_end_mask_0, x = k_423_cast_fp16)[name = tensor<string, []>("op_47556_cast_fp16")];
            tensor<int32, [4]> var_47560_begin_0 = const()[name = tensor<string, []>("op_47560_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_47560_end_0 = const()[name = tensor<string, []>("op_47560_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_47560_end_mask_0 = const()[name = tensor<string, []>("op_47560_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_47560_cast_fp16 = slice_by_index(begin = var_47560_begin_0, end = var_47560_end_0, end_mask = var_47560_end_mask_0, x = k_423_cast_fp16)[name = tensor<string, []>("op_47560_cast_fp16")];
            tensor<int32, [4]> var_47564_begin_0 = const()[name = tensor<string, []>("op_47564_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_47564_end_0 = const()[name = tensor<string, []>("op_47564_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_47564_end_mask_0 = const()[name = tensor<string, []>("op_47564_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_47564_cast_fp16 = slice_by_index(begin = var_47564_begin_0, end = var_47564_end_0, end_mask = var_47564_end_mask_0, x = k_423_cast_fp16)[name = tensor<string, []>("op_47564_cast_fp16")];
            tensor<int32, [4]> var_47568_begin_0 = const()[name = tensor<string, []>("op_47568_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_47568_end_0 = const()[name = tensor<string, []>("op_47568_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_47568_end_mask_0 = const()[name = tensor<string, []>("op_47568_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_47568_cast_fp16 = slice_by_index(begin = var_47568_begin_0, end = var_47568_end_0, end_mask = var_47568_end_mask_0, x = k_423_cast_fp16)[name = tensor<string, []>("op_47568_cast_fp16")];
            tensor<int32, [4]> var_47572_begin_0 = const()[name = tensor<string, []>("op_47572_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_47572_end_0 = const()[name = tensor<string, []>("op_47572_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_47572_end_mask_0 = const()[name = tensor<string, []>("op_47572_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_47572_cast_fp16 = slice_by_index(begin = var_47572_begin_0, end = var_47572_end_0, end_mask = var_47572_end_mask_0, x = k_423_cast_fp16)[name = tensor<string, []>("op_47572_cast_fp16")];
            tensor<int32, [4]> var_47576_begin_0 = const()[name = tensor<string, []>("op_47576_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_47576_end_0 = const()[name = tensor<string, []>("op_47576_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_47576_end_mask_0 = const()[name = tensor<string, []>("op_47576_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_47576_cast_fp16 = slice_by_index(begin = var_47576_begin_0, end = var_47576_end_0, end_mask = var_47576_end_mask_0, x = k_423_cast_fp16)[name = tensor<string, []>("op_47576_cast_fp16")];
            tensor<int32, [4]> var_47580_begin_0 = const()[name = tensor<string, []>("op_47580_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_47580_end_0 = const()[name = tensor<string, []>("op_47580_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_47580_end_mask_0 = const()[name = tensor<string, []>("op_47580_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_47580_cast_fp16 = slice_by_index(begin = var_47580_begin_0, end = var_47580_end_0, end_mask = var_47580_end_mask_0, x = k_423_cast_fp16)[name = tensor<string, []>("op_47580_cast_fp16")];
            tensor<int32, [4]> var_47584_begin_0 = const()[name = tensor<string, []>("op_47584_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_47584_end_0 = const()[name = tensor<string, []>("op_47584_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_47584_end_mask_0 = const()[name = tensor<string, []>("op_47584_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_47584_cast_fp16 = slice_by_index(begin = var_47584_begin_0, end = var_47584_end_0, end_mask = var_47584_end_mask_0, x = k_423_cast_fp16)[name = tensor<string, []>("op_47584_cast_fp16")];
            tensor<int32, [4]> var_47588_begin_0 = const()[name = tensor<string, []>("op_47588_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_47588_end_0 = const()[name = tensor<string, []>("op_47588_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_47588_end_mask_0 = const()[name = tensor<string, []>("op_47588_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_47588_cast_fp16 = slice_by_index(begin = var_47588_begin_0, end = var_47588_end_0, end_mask = var_47588_end_mask_0, x = k_423_cast_fp16)[name = tensor<string, []>("op_47588_cast_fp16")];
            tensor<int32, [4]> var_47592_begin_0 = const()[name = tensor<string, []>("op_47592_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_47592_end_0 = const()[name = tensor<string, []>("op_47592_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_47592_end_mask_0 = const()[name = tensor<string, []>("op_47592_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_47592_cast_fp16 = slice_by_index(begin = var_47592_begin_0, end = var_47592_end_0, end_mask = var_47592_end_mask_0, x = k_423_cast_fp16)[name = tensor<string, []>("op_47592_cast_fp16")];
            tensor<int32, [4]> var_47596_begin_0 = const()[name = tensor<string, []>("op_47596_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_47596_end_0 = const()[name = tensor<string, []>("op_47596_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_47596_end_mask_0 = const()[name = tensor<string, []>("op_47596_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_47596_cast_fp16 = slice_by_index(begin = var_47596_begin_0, end = var_47596_end_0, end_mask = var_47596_end_mask_0, x = k_423_cast_fp16)[name = tensor<string, []>("op_47596_cast_fp16")];
            tensor<int32, [4]> var_47598_begin_0 = const()[name = tensor<string, []>("op_47598_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_47598_end_0 = const()[name = tensor<string, []>("op_47598_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_47598_end_mask_0 = const()[name = tensor<string, []>("op_47598_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_47598_cast_fp16 = slice_by_index(begin = var_47598_begin_0, end = var_47598_end_0, end_mask = var_47598_end_mask_0, x = v_211_cast_fp16)[name = tensor<string, []>("op_47598_cast_fp16")];
            tensor<int32, [4]> var_47602_begin_0 = const()[name = tensor<string, []>("op_47602_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_47602_end_0 = const()[name = tensor<string, []>("op_47602_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_47602_end_mask_0 = const()[name = tensor<string, []>("op_47602_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_47602_cast_fp16 = slice_by_index(begin = var_47602_begin_0, end = var_47602_end_0, end_mask = var_47602_end_mask_0, x = v_211_cast_fp16)[name = tensor<string, []>("op_47602_cast_fp16")];
            tensor<int32, [4]> var_47606_begin_0 = const()[name = tensor<string, []>("op_47606_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_47606_end_0 = const()[name = tensor<string, []>("op_47606_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_47606_end_mask_0 = const()[name = tensor<string, []>("op_47606_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_47606_cast_fp16 = slice_by_index(begin = var_47606_begin_0, end = var_47606_end_0, end_mask = var_47606_end_mask_0, x = v_211_cast_fp16)[name = tensor<string, []>("op_47606_cast_fp16")];
            tensor<int32, [4]> var_47610_begin_0 = const()[name = tensor<string, []>("op_47610_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_47610_end_0 = const()[name = tensor<string, []>("op_47610_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_47610_end_mask_0 = const()[name = tensor<string, []>("op_47610_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_47610_cast_fp16 = slice_by_index(begin = var_47610_begin_0, end = var_47610_end_0, end_mask = var_47610_end_mask_0, x = v_211_cast_fp16)[name = tensor<string, []>("op_47610_cast_fp16")];
            tensor<int32, [4]> var_47614_begin_0 = const()[name = tensor<string, []>("op_47614_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_47614_end_0 = const()[name = tensor<string, []>("op_47614_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_47614_end_mask_0 = const()[name = tensor<string, []>("op_47614_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_47614_cast_fp16 = slice_by_index(begin = var_47614_begin_0, end = var_47614_end_0, end_mask = var_47614_end_mask_0, x = v_211_cast_fp16)[name = tensor<string, []>("op_47614_cast_fp16")];
            tensor<int32, [4]> var_47618_begin_0 = const()[name = tensor<string, []>("op_47618_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_47618_end_0 = const()[name = tensor<string, []>("op_47618_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_47618_end_mask_0 = const()[name = tensor<string, []>("op_47618_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_47618_cast_fp16 = slice_by_index(begin = var_47618_begin_0, end = var_47618_end_0, end_mask = var_47618_end_mask_0, x = v_211_cast_fp16)[name = tensor<string, []>("op_47618_cast_fp16")];
            tensor<int32, [4]> var_47622_begin_0 = const()[name = tensor<string, []>("op_47622_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_47622_end_0 = const()[name = tensor<string, []>("op_47622_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_47622_end_mask_0 = const()[name = tensor<string, []>("op_47622_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_47622_cast_fp16 = slice_by_index(begin = var_47622_begin_0, end = var_47622_end_0, end_mask = var_47622_end_mask_0, x = v_211_cast_fp16)[name = tensor<string, []>("op_47622_cast_fp16")];
            tensor<int32, [4]> var_47626_begin_0 = const()[name = tensor<string, []>("op_47626_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_47626_end_0 = const()[name = tensor<string, []>("op_47626_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_47626_end_mask_0 = const()[name = tensor<string, []>("op_47626_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_47626_cast_fp16 = slice_by_index(begin = var_47626_begin_0, end = var_47626_end_0, end_mask = var_47626_end_mask_0, x = v_211_cast_fp16)[name = tensor<string, []>("op_47626_cast_fp16")];
            tensor<int32, [4]> var_47630_begin_0 = const()[name = tensor<string, []>("op_47630_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_47630_end_0 = const()[name = tensor<string, []>("op_47630_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_47630_end_mask_0 = const()[name = tensor<string, []>("op_47630_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_47630_cast_fp16 = slice_by_index(begin = var_47630_begin_0, end = var_47630_end_0, end_mask = var_47630_end_mask_0, x = v_211_cast_fp16)[name = tensor<string, []>("op_47630_cast_fp16")];
            tensor<int32, [4]> var_47634_begin_0 = const()[name = tensor<string, []>("op_47634_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_47634_end_0 = const()[name = tensor<string, []>("op_47634_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_47634_end_mask_0 = const()[name = tensor<string, []>("op_47634_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_47634_cast_fp16 = slice_by_index(begin = var_47634_begin_0, end = var_47634_end_0, end_mask = var_47634_end_mask_0, x = v_211_cast_fp16)[name = tensor<string, []>("op_47634_cast_fp16")];
            tensor<int32, [4]> var_47638_begin_0 = const()[name = tensor<string, []>("op_47638_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_47638_end_0 = const()[name = tensor<string, []>("op_47638_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_47638_end_mask_0 = const()[name = tensor<string, []>("op_47638_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_47638_cast_fp16 = slice_by_index(begin = var_47638_begin_0, end = var_47638_end_0, end_mask = var_47638_end_mask_0, x = v_211_cast_fp16)[name = tensor<string, []>("op_47638_cast_fp16")];
            tensor<int32, [4]> var_47642_begin_0 = const()[name = tensor<string, []>("op_47642_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_47642_end_0 = const()[name = tensor<string, []>("op_47642_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_47642_end_mask_0 = const()[name = tensor<string, []>("op_47642_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_47642_cast_fp16 = slice_by_index(begin = var_47642_begin_0, end = var_47642_end_0, end_mask = var_47642_end_mask_0, x = v_211_cast_fp16)[name = tensor<string, []>("op_47642_cast_fp16")];
            tensor<int32, [4]> var_47646_begin_0 = const()[name = tensor<string, []>("op_47646_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_47646_end_0 = const()[name = tensor<string, []>("op_47646_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_47646_end_mask_0 = const()[name = tensor<string, []>("op_47646_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_47646_cast_fp16 = slice_by_index(begin = var_47646_begin_0, end = var_47646_end_0, end_mask = var_47646_end_mask_0, x = v_211_cast_fp16)[name = tensor<string, []>("op_47646_cast_fp16")];
            tensor<int32, [4]> var_47650_begin_0 = const()[name = tensor<string, []>("op_47650_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_47650_end_0 = const()[name = tensor<string, []>("op_47650_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_47650_end_mask_0 = const()[name = tensor<string, []>("op_47650_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_47650_cast_fp16 = slice_by_index(begin = var_47650_begin_0, end = var_47650_end_0, end_mask = var_47650_end_mask_0, x = v_211_cast_fp16)[name = tensor<string, []>("op_47650_cast_fp16")];
            tensor<int32, [4]> var_47654_begin_0 = const()[name = tensor<string, []>("op_47654_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_47654_end_0 = const()[name = tensor<string, []>("op_47654_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_47654_end_mask_0 = const()[name = tensor<string, []>("op_47654_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_47654_cast_fp16 = slice_by_index(begin = var_47654_begin_0, end = var_47654_end_0, end_mask = var_47654_end_mask_0, x = v_211_cast_fp16)[name = tensor<string, []>("op_47654_cast_fp16")];
            tensor<int32, [4]> var_47658_begin_0 = const()[name = tensor<string, []>("op_47658_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_47658_end_0 = const()[name = tensor<string, []>("op_47658_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_47658_end_mask_0 = const()[name = tensor<string, []>("op_47658_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_47658_cast_fp16 = slice_by_index(begin = var_47658_begin_0, end = var_47658_end_0, end_mask = var_47658_end_mask_0, x = v_211_cast_fp16)[name = tensor<string, []>("op_47658_cast_fp16")];
            tensor<int32, [4]> var_47662_begin_0 = const()[name = tensor<string, []>("op_47662_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_47662_end_0 = const()[name = tensor<string, []>("op_47662_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_47662_end_mask_0 = const()[name = tensor<string, []>("op_47662_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_47662_cast_fp16 = slice_by_index(begin = var_47662_begin_0, end = var_47662_end_0, end_mask = var_47662_end_mask_0, x = v_211_cast_fp16)[name = tensor<string, []>("op_47662_cast_fp16")];
            tensor<int32, [4]> var_47666_begin_0 = const()[name = tensor<string, []>("op_47666_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_47666_end_0 = const()[name = tensor<string, []>("op_47666_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_47666_end_mask_0 = const()[name = tensor<string, []>("op_47666_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_47666_cast_fp16 = slice_by_index(begin = var_47666_begin_0, end = var_47666_end_0, end_mask = var_47666_end_mask_0, x = v_211_cast_fp16)[name = tensor<string, []>("op_47666_cast_fp16")];
            tensor<int32, [4]> var_47670_begin_0 = const()[name = tensor<string, []>("op_47670_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_47670_end_0 = const()[name = tensor<string, []>("op_47670_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_47670_end_mask_0 = const()[name = tensor<string, []>("op_47670_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_47670_cast_fp16 = slice_by_index(begin = var_47670_begin_0, end = var_47670_end_0, end_mask = var_47670_end_mask_0, x = v_211_cast_fp16)[name = tensor<string, []>("op_47670_cast_fp16")];
            tensor<int32, [4]> var_47674_begin_0 = const()[name = tensor<string, []>("op_47674_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_47674_end_0 = const()[name = tensor<string, []>("op_47674_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_47674_end_mask_0 = const()[name = tensor<string, []>("op_47674_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_47674_cast_fp16 = slice_by_index(begin = var_47674_begin_0, end = var_47674_end_0, end_mask = var_47674_end_mask_0, x = v_211_cast_fp16)[name = tensor<string, []>("op_47674_cast_fp16")];
            tensor<string, []> var_47678_equation_0 = const()[name = tensor<string, []>("op_47678_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_47678_cast_fp16 = einsum(equation = var_47678_equation_0, values = (var_47520_cast_fp16, var_47437_cast_fp16))[name = tensor<string, []>("op_47678_cast_fp16")];
            tensor<fp16, []> var_47679_to_fp16 = const()[name = tensor<string, []>("op_47679_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4041_cast_fp16 = mul(x = var_47678_cast_fp16, y = var_47679_to_fp16)[name = tensor<string, []>("aw_4041_cast_fp16")];
            tensor<string, []> var_47682_equation_0 = const()[name = tensor<string, []>("op_47682_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_47682_cast_fp16 = einsum(equation = var_47682_equation_0, values = (var_47524_cast_fp16, var_47441_cast_fp16))[name = tensor<string, []>("op_47682_cast_fp16")];
            tensor<fp16, []> var_47683_to_fp16 = const()[name = tensor<string, []>("op_47683_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4043_cast_fp16 = mul(x = var_47682_cast_fp16, y = var_47683_to_fp16)[name = tensor<string, []>("aw_4043_cast_fp16")];
            tensor<string, []> var_47686_equation_0 = const()[name = tensor<string, []>("op_47686_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_47686_cast_fp16 = einsum(equation = var_47686_equation_0, values = (var_47528_cast_fp16, var_47445_cast_fp16))[name = tensor<string, []>("op_47686_cast_fp16")];
            tensor<fp16, []> var_47687_to_fp16 = const()[name = tensor<string, []>("op_47687_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4045_cast_fp16 = mul(x = var_47686_cast_fp16, y = var_47687_to_fp16)[name = tensor<string, []>("aw_4045_cast_fp16")];
            tensor<string, []> var_47690_equation_0 = const()[name = tensor<string, []>("op_47690_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_47690_cast_fp16 = einsum(equation = var_47690_equation_0, values = (var_47532_cast_fp16, var_47449_cast_fp16))[name = tensor<string, []>("op_47690_cast_fp16")];
            tensor<fp16, []> var_47691_to_fp16 = const()[name = tensor<string, []>("op_47691_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4047_cast_fp16 = mul(x = var_47690_cast_fp16, y = var_47691_to_fp16)[name = tensor<string, []>("aw_4047_cast_fp16")];
            tensor<string, []> var_47694_equation_0 = const()[name = tensor<string, []>("op_47694_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_47694_cast_fp16 = einsum(equation = var_47694_equation_0, values = (var_47536_cast_fp16, var_47453_cast_fp16))[name = tensor<string, []>("op_47694_cast_fp16")];
            tensor<fp16, []> var_47695_to_fp16 = const()[name = tensor<string, []>("op_47695_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4049_cast_fp16 = mul(x = var_47694_cast_fp16, y = var_47695_to_fp16)[name = tensor<string, []>("aw_4049_cast_fp16")];
            tensor<string, []> var_47698_equation_0 = const()[name = tensor<string, []>("op_47698_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_47698_cast_fp16 = einsum(equation = var_47698_equation_0, values = (var_47540_cast_fp16, var_47457_cast_fp16))[name = tensor<string, []>("op_47698_cast_fp16")];
            tensor<fp16, []> var_47699_to_fp16 = const()[name = tensor<string, []>("op_47699_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4051_cast_fp16 = mul(x = var_47698_cast_fp16, y = var_47699_to_fp16)[name = tensor<string, []>("aw_4051_cast_fp16")];
            tensor<string, []> var_47702_equation_0 = const()[name = tensor<string, []>("op_47702_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_47702_cast_fp16 = einsum(equation = var_47702_equation_0, values = (var_47544_cast_fp16, var_47461_cast_fp16))[name = tensor<string, []>("op_47702_cast_fp16")];
            tensor<fp16, []> var_47703_to_fp16 = const()[name = tensor<string, []>("op_47703_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4053_cast_fp16 = mul(x = var_47702_cast_fp16, y = var_47703_to_fp16)[name = tensor<string, []>("aw_4053_cast_fp16")];
            tensor<string, []> var_47706_equation_0 = const()[name = tensor<string, []>("op_47706_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_47706_cast_fp16 = einsum(equation = var_47706_equation_0, values = (var_47548_cast_fp16, var_47465_cast_fp16))[name = tensor<string, []>("op_47706_cast_fp16")];
            tensor<fp16, []> var_47707_to_fp16 = const()[name = tensor<string, []>("op_47707_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4055_cast_fp16 = mul(x = var_47706_cast_fp16, y = var_47707_to_fp16)[name = tensor<string, []>("aw_4055_cast_fp16")];
            tensor<string, []> var_47710_equation_0 = const()[name = tensor<string, []>("op_47710_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_47710_cast_fp16 = einsum(equation = var_47710_equation_0, values = (var_47552_cast_fp16, var_47469_cast_fp16))[name = tensor<string, []>("op_47710_cast_fp16")];
            tensor<fp16, []> var_47711_to_fp16 = const()[name = tensor<string, []>("op_47711_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4057_cast_fp16 = mul(x = var_47710_cast_fp16, y = var_47711_to_fp16)[name = tensor<string, []>("aw_4057_cast_fp16")];
            tensor<string, []> var_47714_equation_0 = const()[name = tensor<string, []>("op_47714_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_47714_cast_fp16 = einsum(equation = var_47714_equation_0, values = (var_47556_cast_fp16, var_47473_cast_fp16))[name = tensor<string, []>("op_47714_cast_fp16")];
            tensor<fp16, []> var_47715_to_fp16 = const()[name = tensor<string, []>("op_47715_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4059_cast_fp16 = mul(x = var_47714_cast_fp16, y = var_47715_to_fp16)[name = tensor<string, []>("aw_4059_cast_fp16")];
            tensor<string, []> var_47718_equation_0 = const()[name = tensor<string, []>("op_47718_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_47718_cast_fp16 = einsum(equation = var_47718_equation_0, values = (var_47560_cast_fp16, var_47477_cast_fp16))[name = tensor<string, []>("op_47718_cast_fp16")];
            tensor<fp16, []> var_47719_to_fp16 = const()[name = tensor<string, []>("op_47719_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4061_cast_fp16 = mul(x = var_47718_cast_fp16, y = var_47719_to_fp16)[name = tensor<string, []>("aw_4061_cast_fp16")];
            tensor<string, []> var_47722_equation_0 = const()[name = tensor<string, []>("op_47722_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_47722_cast_fp16 = einsum(equation = var_47722_equation_0, values = (var_47564_cast_fp16, var_47481_cast_fp16))[name = tensor<string, []>("op_47722_cast_fp16")];
            tensor<fp16, []> var_47723_to_fp16 = const()[name = tensor<string, []>("op_47723_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4063_cast_fp16 = mul(x = var_47722_cast_fp16, y = var_47723_to_fp16)[name = tensor<string, []>("aw_4063_cast_fp16")];
            tensor<string, []> var_47726_equation_0 = const()[name = tensor<string, []>("op_47726_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_47726_cast_fp16 = einsum(equation = var_47726_equation_0, values = (var_47568_cast_fp16, var_47485_cast_fp16))[name = tensor<string, []>("op_47726_cast_fp16")];
            tensor<fp16, []> var_47727_to_fp16 = const()[name = tensor<string, []>("op_47727_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4065_cast_fp16 = mul(x = var_47726_cast_fp16, y = var_47727_to_fp16)[name = tensor<string, []>("aw_4065_cast_fp16")];
            tensor<string, []> var_47730_equation_0 = const()[name = tensor<string, []>("op_47730_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_47730_cast_fp16 = einsum(equation = var_47730_equation_0, values = (var_47572_cast_fp16, var_47489_cast_fp16))[name = tensor<string, []>("op_47730_cast_fp16")];
            tensor<fp16, []> var_47731_to_fp16 = const()[name = tensor<string, []>("op_47731_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4067_cast_fp16 = mul(x = var_47730_cast_fp16, y = var_47731_to_fp16)[name = tensor<string, []>("aw_4067_cast_fp16")];
            tensor<string, []> var_47734_equation_0 = const()[name = tensor<string, []>("op_47734_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_47734_cast_fp16 = einsum(equation = var_47734_equation_0, values = (var_47576_cast_fp16, var_47493_cast_fp16))[name = tensor<string, []>("op_47734_cast_fp16")];
            tensor<fp16, []> var_47735_to_fp16 = const()[name = tensor<string, []>("op_47735_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4069_cast_fp16 = mul(x = var_47734_cast_fp16, y = var_47735_to_fp16)[name = tensor<string, []>("aw_4069_cast_fp16")];
            tensor<string, []> var_47738_equation_0 = const()[name = tensor<string, []>("op_47738_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_47738_cast_fp16 = einsum(equation = var_47738_equation_0, values = (var_47580_cast_fp16, var_47497_cast_fp16))[name = tensor<string, []>("op_47738_cast_fp16")];
            tensor<fp16, []> var_47739_to_fp16 = const()[name = tensor<string, []>("op_47739_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4071_cast_fp16 = mul(x = var_47738_cast_fp16, y = var_47739_to_fp16)[name = tensor<string, []>("aw_4071_cast_fp16")];
            tensor<string, []> var_47742_equation_0 = const()[name = tensor<string, []>("op_47742_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_47742_cast_fp16 = einsum(equation = var_47742_equation_0, values = (var_47584_cast_fp16, var_47501_cast_fp16))[name = tensor<string, []>("op_47742_cast_fp16")];
            tensor<fp16, []> var_47743_to_fp16 = const()[name = tensor<string, []>("op_47743_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4073_cast_fp16 = mul(x = var_47742_cast_fp16, y = var_47743_to_fp16)[name = tensor<string, []>("aw_4073_cast_fp16")];
            tensor<string, []> var_47746_equation_0 = const()[name = tensor<string, []>("op_47746_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_47746_cast_fp16 = einsum(equation = var_47746_equation_0, values = (var_47588_cast_fp16, var_47505_cast_fp16))[name = tensor<string, []>("op_47746_cast_fp16")];
            tensor<fp16, []> var_47747_to_fp16 = const()[name = tensor<string, []>("op_47747_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4075_cast_fp16 = mul(x = var_47746_cast_fp16, y = var_47747_to_fp16)[name = tensor<string, []>("aw_4075_cast_fp16")];
            tensor<string, []> var_47750_equation_0 = const()[name = tensor<string, []>("op_47750_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_47750_cast_fp16 = einsum(equation = var_47750_equation_0, values = (var_47592_cast_fp16, var_47509_cast_fp16))[name = tensor<string, []>("op_47750_cast_fp16")];
            tensor<fp16, []> var_47751_to_fp16 = const()[name = tensor<string, []>("op_47751_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4077_cast_fp16 = mul(x = var_47750_cast_fp16, y = var_47751_to_fp16)[name = tensor<string, []>("aw_4077_cast_fp16")];
            tensor<string, []> var_47754_equation_0 = const()[name = tensor<string, []>("op_47754_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_47754_cast_fp16 = einsum(equation = var_47754_equation_0, values = (var_47596_cast_fp16, var_47513_cast_fp16))[name = tensor<string, []>("op_47754_cast_fp16")];
            tensor<fp16, []> var_47755_to_fp16 = const()[name = tensor<string, []>("op_47755_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4079_cast_fp16 = mul(x = var_47754_cast_fp16, y = var_47755_to_fp16)[name = tensor<string, []>("aw_4079_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_47757_cast_fp16 = softmax(axis = var_30355, x = aw_4041_cast_fp16)[name = tensor<string, []>("op_47757_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_47758_cast_fp16 = softmax(axis = var_30355, x = aw_4043_cast_fp16)[name = tensor<string, []>("op_47758_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_47759_cast_fp16 = softmax(axis = var_30355, x = aw_4045_cast_fp16)[name = tensor<string, []>("op_47759_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_47760_cast_fp16 = softmax(axis = var_30355, x = aw_4047_cast_fp16)[name = tensor<string, []>("op_47760_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_47761_cast_fp16 = softmax(axis = var_30355, x = aw_4049_cast_fp16)[name = tensor<string, []>("op_47761_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_47762_cast_fp16 = softmax(axis = var_30355, x = aw_4051_cast_fp16)[name = tensor<string, []>("op_47762_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_47763_cast_fp16 = softmax(axis = var_30355, x = aw_4053_cast_fp16)[name = tensor<string, []>("op_47763_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_47764_cast_fp16 = softmax(axis = var_30355, x = aw_4055_cast_fp16)[name = tensor<string, []>("op_47764_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_47765_cast_fp16 = softmax(axis = var_30355, x = aw_4057_cast_fp16)[name = tensor<string, []>("op_47765_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_47766_cast_fp16 = softmax(axis = var_30355, x = aw_4059_cast_fp16)[name = tensor<string, []>("op_47766_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_47767_cast_fp16 = softmax(axis = var_30355, x = aw_4061_cast_fp16)[name = tensor<string, []>("op_47767_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_47768_cast_fp16 = softmax(axis = var_30355, x = aw_4063_cast_fp16)[name = tensor<string, []>("op_47768_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_47769_cast_fp16 = softmax(axis = var_30355, x = aw_4065_cast_fp16)[name = tensor<string, []>("op_47769_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_47770_cast_fp16 = softmax(axis = var_30355, x = aw_4067_cast_fp16)[name = tensor<string, []>("op_47770_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_47771_cast_fp16 = softmax(axis = var_30355, x = aw_4069_cast_fp16)[name = tensor<string, []>("op_47771_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_47772_cast_fp16 = softmax(axis = var_30355, x = aw_4071_cast_fp16)[name = tensor<string, []>("op_47772_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_47773_cast_fp16 = softmax(axis = var_30355, x = aw_4073_cast_fp16)[name = tensor<string, []>("op_47773_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_47774_cast_fp16 = softmax(axis = var_30355, x = aw_4075_cast_fp16)[name = tensor<string, []>("op_47774_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_47775_cast_fp16 = softmax(axis = var_30355, x = aw_4077_cast_fp16)[name = tensor<string, []>("op_47775_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_47776_cast_fp16 = softmax(axis = var_30355, x = aw_4079_cast_fp16)[name = tensor<string, []>("op_47776_cast_fp16")];
            tensor<string, []> var_47778_equation_0 = const()[name = tensor<string, []>("op_47778_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47778_cast_fp16 = einsum(equation = var_47778_equation_0, values = (var_47598_cast_fp16, var_47757_cast_fp16))[name = tensor<string, []>("op_47778_cast_fp16")];
            tensor<string, []> var_47780_equation_0 = const()[name = tensor<string, []>("op_47780_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47780_cast_fp16 = einsum(equation = var_47780_equation_0, values = (var_47602_cast_fp16, var_47758_cast_fp16))[name = tensor<string, []>("op_47780_cast_fp16")];
            tensor<string, []> var_47782_equation_0 = const()[name = tensor<string, []>("op_47782_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47782_cast_fp16 = einsum(equation = var_47782_equation_0, values = (var_47606_cast_fp16, var_47759_cast_fp16))[name = tensor<string, []>("op_47782_cast_fp16")];
            tensor<string, []> var_47784_equation_0 = const()[name = tensor<string, []>("op_47784_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47784_cast_fp16 = einsum(equation = var_47784_equation_0, values = (var_47610_cast_fp16, var_47760_cast_fp16))[name = tensor<string, []>("op_47784_cast_fp16")];
            tensor<string, []> var_47786_equation_0 = const()[name = tensor<string, []>("op_47786_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47786_cast_fp16 = einsum(equation = var_47786_equation_0, values = (var_47614_cast_fp16, var_47761_cast_fp16))[name = tensor<string, []>("op_47786_cast_fp16")];
            tensor<string, []> var_47788_equation_0 = const()[name = tensor<string, []>("op_47788_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47788_cast_fp16 = einsum(equation = var_47788_equation_0, values = (var_47618_cast_fp16, var_47762_cast_fp16))[name = tensor<string, []>("op_47788_cast_fp16")];
            tensor<string, []> var_47790_equation_0 = const()[name = tensor<string, []>("op_47790_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47790_cast_fp16 = einsum(equation = var_47790_equation_0, values = (var_47622_cast_fp16, var_47763_cast_fp16))[name = tensor<string, []>("op_47790_cast_fp16")];
            tensor<string, []> var_47792_equation_0 = const()[name = tensor<string, []>("op_47792_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47792_cast_fp16 = einsum(equation = var_47792_equation_0, values = (var_47626_cast_fp16, var_47764_cast_fp16))[name = tensor<string, []>("op_47792_cast_fp16")];
            tensor<string, []> var_47794_equation_0 = const()[name = tensor<string, []>("op_47794_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47794_cast_fp16 = einsum(equation = var_47794_equation_0, values = (var_47630_cast_fp16, var_47765_cast_fp16))[name = tensor<string, []>("op_47794_cast_fp16")];
            tensor<string, []> var_47796_equation_0 = const()[name = tensor<string, []>("op_47796_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47796_cast_fp16 = einsum(equation = var_47796_equation_0, values = (var_47634_cast_fp16, var_47766_cast_fp16))[name = tensor<string, []>("op_47796_cast_fp16")];
            tensor<string, []> var_47798_equation_0 = const()[name = tensor<string, []>("op_47798_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47798_cast_fp16 = einsum(equation = var_47798_equation_0, values = (var_47638_cast_fp16, var_47767_cast_fp16))[name = tensor<string, []>("op_47798_cast_fp16")];
            tensor<string, []> var_47800_equation_0 = const()[name = tensor<string, []>("op_47800_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47800_cast_fp16 = einsum(equation = var_47800_equation_0, values = (var_47642_cast_fp16, var_47768_cast_fp16))[name = tensor<string, []>("op_47800_cast_fp16")];
            tensor<string, []> var_47802_equation_0 = const()[name = tensor<string, []>("op_47802_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47802_cast_fp16 = einsum(equation = var_47802_equation_0, values = (var_47646_cast_fp16, var_47769_cast_fp16))[name = tensor<string, []>("op_47802_cast_fp16")];
            tensor<string, []> var_47804_equation_0 = const()[name = tensor<string, []>("op_47804_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47804_cast_fp16 = einsum(equation = var_47804_equation_0, values = (var_47650_cast_fp16, var_47770_cast_fp16))[name = tensor<string, []>("op_47804_cast_fp16")];
            tensor<string, []> var_47806_equation_0 = const()[name = tensor<string, []>("op_47806_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47806_cast_fp16 = einsum(equation = var_47806_equation_0, values = (var_47654_cast_fp16, var_47771_cast_fp16))[name = tensor<string, []>("op_47806_cast_fp16")];
            tensor<string, []> var_47808_equation_0 = const()[name = tensor<string, []>("op_47808_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47808_cast_fp16 = einsum(equation = var_47808_equation_0, values = (var_47658_cast_fp16, var_47772_cast_fp16))[name = tensor<string, []>("op_47808_cast_fp16")];
            tensor<string, []> var_47810_equation_0 = const()[name = tensor<string, []>("op_47810_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47810_cast_fp16 = einsum(equation = var_47810_equation_0, values = (var_47662_cast_fp16, var_47773_cast_fp16))[name = tensor<string, []>("op_47810_cast_fp16")];
            tensor<string, []> var_47812_equation_0 = const()[name = tensor<string, []>("op_47812_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47812_cast_fp16 = einsum(equation = var_47812_equation_0, values = (var_47666_cast_fp16, var_47774_cast_fp16))[name = tensor<string, []>("op_47812_cast_fp16")];
            tensor<string, []> var_47814_equation_0 = const()[name = tensor<string, []>("op_47814_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47814_cast_fp16 = einsum(equation = var_47814_equation_0, values = (var_47670_cast_fp16, var_47775_cast_fp16))[name = tensor<string, []>("op_47814_cast_fp16")];
            tensor<string, []> var_47816_equation_0 = const()[name = tensor<string, []>("op_47816_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47816_cast_fp16 = einsum(equation = var_47816_equation_0, values = (var_47674_cast_fp16, var_47776_cast_fp16))[name = tensor<string, []>("op_47816_cast_fp16")];
            tensor<bool, []> input_603_interleave_0 = const()[name = tensor<string, []>("input_603_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_603_cast_fp16 = concat(axis = var_30355, interleave = input_603_interleave_0, values = (var_47778_cast_fp16, var_47780_cast_fp16, var_47782_cast_fp16, var_47784_cast_fp16, var_47786_cast_fp16, var_47788_cast_fp16, var_47790_cast_fp16, var_47792_cast_fp16, var_47794_cast_fp16, var_47796_cast_fp16, var_47798_cast_fp16, var_47800_cast_fp16, var_47802_cast_fp16, var_47804_cast_fp16, var_47806_cast_fp16, var_47808_cast_fp16, var_47810_cast_fp16, var_47812_cast_fp16, var_47814_cast_fp16, var_47816_cast_fp16))[name = tensor<string, []>("input_603_cast_fp16")];
            tensor<string, []> var_47826_pad_type_0 = const()[name = tensor<string, []>("op_47826_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_47826_strides_0 = const()[name = tensor<string, []>("op_47826_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_47826_pad_0 = const()[name = tensor<string, []>("op_47826_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_47826_dilations_0 = const()[name = tensor<string, []>("op_47826_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_47826_groups_0 = const()[name = tensor<string, []>("op_47826_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_8_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(522919040))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(524147904))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_8_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_8_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_8_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(524148096)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_47826_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_8_attn2_to_out_0_bias_to_fp16, dilations = var_47826_dilations_0, groups = var_47826_groups_0, pad = var_47826_pad_0, pad_type = var_47826_pad_type_0, strides = var_47826_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_8_attn2_to_out_0_weight_to_fp16_palettized, x = input_603_cast_fp16)[name = tensor<string, []>("op_47826_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_317_cast_fp16 = add(x = var_47826_cast_fp16, y = inputs_315_cast_fp16)[name = tensor<string, []>("inputs_317_cast_fp16")];
            tensor<int32, [1]> input_605_axes_0 = const()[name = tensor<string, []>("input_605_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> input_605_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_605_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(524150720)))];
            tensor<fp16, [1280]> input_605_beta_0_to_fp16 = const()[name = tensor<string, []>("input_605_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(524153344)))];
            tensor<fp16, []> var_47836_to_fp16 = const()[name = tensor<string, []>("op_47836_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> input_605_cast_fp16 = layer_norm(axes = input_605_axes_0, beta = input_605_beta_0_to_fp16, epsilon = var_47836_to_fp16, gamma = input_605_gamma_0_to_fp16, x = inputs_317_cast_fp16)[name = tensor<string, []>("input_605_cast_fp16")];
            tensor<string, []> var_47856_pad_type_0 = const()[name = tensor<string, []>("op_47856_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_47856_strides_0 = const()[name = tensor<string, []>("op_47856_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_47856_pad_0 = const()[name = tensor<string, []>("op_47856_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_47856_dilations_0 = const()[name = tensor<string, []>("op_47856_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_47856_groups_0 = const()[name = tensor<string, []>("op_47856_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_8_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [9830400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(524155968))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(533986432))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_8_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([10240, 1280, 1, 1])];
            tensor<fp16, [10240]> up_blocks_0_attentions_1_transformer_blocks_8_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_8_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(533986624)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_47856_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_8_ff_net_0_proj_bias_to_fp16, dilations = var_47856_dilations_0, groups = var_47856_groups_0, pad = var_47856_pad_0, pad_type = var_47856_pad_type_0, strides = var_47856_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_8_ff_net_0_proj_weight_to_fp16_palettized, x = input_605_cast_fp16)[name = tensor<string, []>("op_47856_cast_fp16")];
            tensor<int32, [2]> var_47857_split_sizes_0 = const()[name = tensor<string, []>("op_47857_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_47857_axis_0 = const()[name = tensor<string, []>("op_47857_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_47857_cast_fp16_0, tensor<fp16, [2, 5120, 1, 1024]> var_47857_cast_fp16_1 = split(axis = var_47857_axis_0, split_sizes = var_47857_split_sizes_0, x = var_47856_cast_fp16)[name = tensor<string, []>("op_47857_cast_fp16")];
            tensor<string, []> var_47859_mode_0 = const()[name = tensor<string, []>("op_47859_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_47859_cast_fp16 = gelu(mode = var_47859_mode_0, x = var_47857_cast_fp16_1)[name = tensor<string, []>("op_47859_cast_fp16")];
            tensor<fp16, [2, 5120, 1, 1024]> input_607_cast_fp16 = mul(x = var_47857_cast_fp16_0, y = var_47859_cast_fp16)[name = tensor<string, []>("input_607_cast_fp16")];
            tensor<string, []> var_47867_pad_type_0 = const()[name = tensor<string, []>("op_47867_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_47867_strides_0 = const()[name = tensor<string, []>("op_47867_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_47867_pad_0 = const()[name = tensor<string, []>("op_47867_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_47867_dilations_0 = const()[name = tensor<string, []>("op_47867_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_47867_groups_0 = const()[name = tensor<string, []>("op_47867_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_8_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [4915200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(534007168))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(538922432))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_8_ff_net_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 5120, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_8_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_8_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(538922624)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_47867_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_8_ff_net_2_bias_to_fp16, dilations = var_47867_dilations_0, groups = var_47867_groups_0, pad = var_47867_pad_0, pad_type = var_47867_pad_type_0, strides = var_47867_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_8_ff_net_2_weight_to_fp16_palettized, x = input_607_cast_fp16)[name = tensor<string, []>("op_47867_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_319_cast_fp16 = add(x = var_47867_cast_fp16, y = inputs_317_cast_fp16)[name = tensor<string, []>("inputs_319_cast_fp16")];
            tensor<int32, [1]> hidden_states_415_axes_0 = const()[name = tensor<string, []>("hidden_states_415_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_415_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_415_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(538925248)))];
            tensor<fp16, [1280]> hidden_states_415_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_415_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(538927872)))];
            tensor<fp16, []> var_47883_to_fp16 = const()[name = tensor<string, []>("op_47883_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_415_cast_fp16 = layer_norm(axes = hidden_states_415_axes_0, beta = hidden_states_415_beta_0_to_fp16, epsilon = var_47883_to_fp16, gamma = hidden_states_415_gamma_0_to_fp16, x = inputs_319_cast_fp16)[name = tensor<string, []>("hidden_states_415_cast_fp16")];
            tensor<string, []> q_213_pad_type_0 = const()[name = tensor<string, []>("q_213_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_213_strides_0 = const()[name = tensor<string, []>("q_213_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_213_pad_0 = const()[name = tensor<string, []>("q_213_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_213_dilations_0 = const()[name = tensor<string, []>("q_213_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_213_groups_0 = const()[name = tensor<string, []>("q_213_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_9_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(538930496))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(540159360))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_9_attn1_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_213_cast_fp16 = conv(dilations = q_213_dilations_0, groups = q_213_groups_0, pad = q_213_pad_0, pad_type = q_213_pad_type_0, strides = q_213_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_9_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_415_cast_fp16)[name = tensor<string, []>("q_213_cast_fp16")];
            tensor<string, []> k_425_pad_type_0 = const()[name = tensor<string, []>("k_425_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_425_strides_0 = const()[name = tensor<string, []>("k_425_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_425_pad_0 = const()[name = tensor<string, []>("k_425_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_425_dilations_0 = const()[name = tensor<string, []>("k_425_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_425_groups_0 = const()[name = tensor<string, []>("k_425_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_9_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(540159552))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(541388416))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_9_attn1_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> k_425_cast_fp16 = conv(dilations = k_425_dilations_0, groups = k_425_groups_0, pad = k_425_pad_0, pad_type = k_425_pad_type_0, strides = k_425_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_9_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_415_cast_fp16)[name = tensor<string, []>("k_425_cast_fp16")];
            tensor<string, []> v_213_pad_type_0 = const()[name = tensor<string, []>("v_213_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_213_strides_0 = const()[name = tensor<string, []>("v_213_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_213_pad_0 = const()[name = tensor<string, []>("v_213_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_213_dilations_0 = const()[name = tensor<string, []>("v_213_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_213_groups_0 = const()[name = tensor<string, []>("v_213_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_9_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(541388608))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(542617472))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_9_attn1_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> v_213_cast_fp16 = conv(dilations = v_213_dilations_0, groups = v_213_groups_0, pad = v_213_pad_0, pad_type = v_213_pad_type_0, strides = v_213_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_9_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_415_cast_fp16)[name = tensor<string, []>("v_213_cast_fp16")];
            tensor<int32, [4]> var_47916_begin_0 = const()[name = tensor<string, []>("op_47916_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_47916_end_0 = const()[name = tensor<string, []>("op_47916_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_47916_end_mask_0 = const()[name = tensor<string, []>("op_47916_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47916_cast_fp16 = slice_by_index(begin = var_47916_begin_0, end = var_47916_end_0, end_mask = var_47916_end_mask_0, x = q_213_cast_fp16)[name = tensor<string, []>("op_47916_cast_fp16")];
            tensor<int32, [4]> var_47920_begin_0 = const()[name = tensor<string, []>("op_47920_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_47920_end_0 = const()[name = tensor<string, []>("op_47920_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_47920_end_mask_0 = const()[name = tensor<string, []>("op_47920_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47920_cast_fp16 = slice_by_index(begin = var_47920_begin_0, end = var_47920_end_0, end_mask = var_47920_end_mask_0, x = q_213_cast_fp16)[name = tensor<string, []>("op_47920_cast_fp16")];
            tensor<int32, [4]> var_47924_begin_0 = const()[name = tensor<string, []>("op_47924_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_47924_end_0 = const()[name = tensor<string, []>("op_47924_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_47924_end_mask_0 = const()[name = tensor<string, []>("op_47924_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47924_cast_fp16 = slice_by_index(begin = var_47924_begin_0, end = var_47924_end_0, end_mask = var_47924_end_mask_0, x = q_213_cast_fp16)[name = tensor<string, []>("op_47924_cast_fp16")];
            tensor<int32, [4]> var_47928_begin_0 = const()[name = tensor<string, []>("op_47928_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_47928_end_0 = const()[name = tensor<string, []>("op_47928_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_47928_end_mask_0 = const()[name = tensor<string, []>("op_47928_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47928_cast_fp16 = slice_by_index(begin = var_47928_begin_0, end = var_47928_end_0, end_mask = var_47928_end_mask_0, x = q_213_cast_fp16)[name = tensor<string, []>("op_47928_cast_fp16")];
            tensor<int32, [4]> var_47932_begin_0 = const()[name = tensor<string, []>("op_47932_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_47932_end_0 = const()[name = tensor<string, []>("op_47932_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_47932_end_mask_0 = const()[name = tensor<string, []>("op_47932_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47932_cast_fp16 = slice_by_index(begin = var_47932_begin_0, end = var_47932_end_0, end_mask = var_47932_end_mask_0, x = q_213_cast_fp16)[name = tensor<string, []>("op_47932_cast_fp16")];
            tensor<int32, [4]> var_47936_begin_0 = const()[name = tensor<string, []>("op_47936_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_47936_end_0 = const()[name = tensor<string, []>("op_47936_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_47936_end_mask_0 = const()[name = tensor<string, []>("op_47936_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47936_cast_fp16 = slice_by_index(begin = var_47936_begin_0, end = var_47936_end_0, end_mask = var_47936_end_mask_0, x = q_213_cast_fp16)[name = tensor<string, []>("op_47936_cast_fp16")];
            tensor<int32, [4]> var_47940_begin_0 = const()[name = tensor<string, []>("op_47940_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_47940_end_0 = const()[name = tensor<string, []>("op_47940_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_47940_end_mask_0 = const()[name = tensor<string, []>("op_47940_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47940_cast_fp16 = slice_by_index(begin = var_47940_begin_0, end = var_47940_end_0, end_mask = var_47940_end_mask_0, x = q_213_cast_fp16)[name = tensor<string, []>("op_47940_cast_fp16")];
            tensor<int32, [4]> var_47944_begin_0 = const()[name = tensor<string, []>("op_47944_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_47944_end_0 = const()[name = tensor<string, []>("op_47944_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_47944_end_mask_0 = const()[name = tensor<string, []>("op_47944_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47944_cast_fp16 = slice_by_index(begin = var_47944_begin_0, end = var_47944_end_0, end_mask = var_47944_end_mask_0, x = q_213_cast_fp16)[name = tensor<string, []>("op_47944_cast_fp16")];
            tensor<int32, [4]> var_47948_begin_0 = const()[name = tensor<string, []>("op_47948_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_47948_end_0 = const()[name = tensor<string, []>("op_47948_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_47948_end_mask_0 = const()[name = tensor<string, []>("op_47948_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47948_cast_fp16 = slice_by_index(begin = var_47948_begin_0, end = var_47948_end_0, end_mask = var_47948_end_mask_0, x = q_213_cast_fp16)[name = tensor<string, []>("op_47948_cast_fp16")];
            tensor<int32, [4]> var_47952_begin_0 = const()[name = tensor<string, []>("op_47952_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_47952_end_0 = const()[name = tensor<string, []>("op_47952_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_47952_end_mask_0 = const()[name = tensor<string, []>("op_47952_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47952_cast_fp16 = slice_by_index(begin = var_47952_begin_0, end = var_47952_end_0, end_mask = var_47952_end_mask_0, x = q_213_cast_fp16)[name = tensor<string, []>("op_47952_cast_fp16")];
            tensor<int32, [4]> var_47956_begin_0 = const()[name = tensor<string, []>("op_47956_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_47956_end_0 = const()[name = tensor<string, []>("op_47956_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_47956_end_mask_0 = const()[name = tensor<string, []>("op_47956_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47956_cast_fp16 = slice_by_index(begin = var_47956_begin_0, end = var_47956_end_0, end_mask = var_47956_end_mask_0, x = q_213_cast_fp16)[name = tensor<string, []>("op_47956_cast_fp16")];
            tensor<int32, [4]> var_47960_begin_0 = const()[name = tensor<string, []>("op_47960_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_47960_end_0 = const()[name = tensor<string, []>("op_47960_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_47960_end_mask_0 = const()[name = tensor<string, []>("op_47960_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47960_cast_fp16 = slice_by_index(begin = var_47960_begin_0, end = var_47960_end_0, end_mask = var_47960_end_mask_0, x = q_213_cast_fp16)[name = tensor<string, []>("op_47960_cast_fp16")];
            tensor<int32, [4]> var_47964_begin_0 = const()[name = tensor<string, []>("op_47964_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_47964_end_0 = const()[name = tensor<string, []>("op_47964_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_47964_end_mask_0 = const()[name = tensor<string, []>("op_47964_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47964_cast_fp16 = slice_by_index(begin = var_47964_begin_0, end = var_47964_end_0, end_mask = var_47964_end_mask_0, x = q_213_cast_fp16)[name = tensor<string, []>("op_47964_cast_fp16")];
            tensor<int32, [4]> var_47968_begin_0 = const()[name = tensor<string, []>("op_47968_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_47968_end_0 = const()[name = tensor<string, []>("op_47968_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_47968_end_mask_0 = const()[name = tensor<string, []>("op_47968_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47968_cast_fp16 = slice_by_index(begin = var_47968_begin_0, end = var_47968_end_0, end_mask = var_47968_end_mask_0, x = q_213_cast_fp16)[name = tensor<string, []>("op_47968_cast_fp16")];
            tensor<int32, [4]> var_47972_begin_0 = const()[name = tensor<string, []>("op_47972_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_47972_end_0 = const()[name = tensor<string, []>("op_47972_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_47972_end_mask_0 = const()[name = tensor<string, []>("op_47972_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47972_cast_fp16 = slice_by_index(begin = var_47972_begin_0, end = var_47972_end_0, end_mask = var_47972_end_mask_0, x = q_213_cast_fp16)[name = tensor<string, []>("op_47972_cast_fp16")];
            tensor<int32, [4]> var_47976_begin_0 = const()[name = tensor<string, []>("op_47976_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_47976_end_0 = const()[name = tensor<string, []>("op_47976_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_47976_end_mask_0 = const()[name = tensor<string, []>("op_47976_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47976_cast_fp16 = slice_by_index(begin = var_47976_begin_0, end = var_47976_end_0, end_mask = var_47976_end_mask_0, x = q_213_cast_fp16)[name = tensor<string, []>("op_47976_cast_fp16")];
            tensor<int32, [4]> var_47980_begin_0 = const()[name = tensor<string, []>("op_47980_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_47980_end_0 = const()[name = tensor<string, []>("op_47980_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_47980_end_mask_0 = const()[name = tensor<string, []>("op_47980_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47980_cast_fp16 = slice_by_index(begin = var_47980_begin_0, end = var_47980_end_0, end_mask = var_47980_end_mask_0, x = q_213_cast_fp16)[name = tensor<string, []>("op_47980_cast_fp16")];
            tensor<int32, [4]> var_47984_begin_0 = const()[name = tensor<string, []>("op_47984_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_47984_end_0 = const()[name = tensor<string, []>("op_47984_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_47984_end_mask_0 = const()[name = tensor<string, []>("op_47984_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47984_cast_fp16 = slice_by_index(begin = var_47984_begin_0, end = var_47984_end_0, end_mask = var_47984_end_mask_0, x = q_213_cast_fp16)[name = tensor<string, []>("op_47984_cast_fp16")];
            tensor<int32, [4]> var_47988_begin_0 = const()[name = tensor<string, []>("op_47988_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_47988_end_0 = const()[name = tensor<string, []>("op_47988_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_47988_end_mask_0 = const()[name = tensor<string, []>("op_47988_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47988_cast_fp16 = slice_by_index(begin = var_47988_begin_0, end = var_47988_end_0, end_mask = var_47988_end_mask_0, x = q_213_cast_fp16)[name = tensor<string, []>("op_47988_cast_fp16")];
            tensor<int32, [4]> var_47992_begin_0 = const()[name = tensor<string, []>("op_47992_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_47992_end_0 = const()[name = tensor<string, []>("op_47992_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_47992_end_mask_0 = const()[name = tensor<string, []>("op_47992_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47992_cast_fp16 = slice_by_index(begin = var_47992_begin_0, end = var_47992_end_0, end_mask = var_47992_end_mask_0, x = q_213_cast_fp16)[name = tensor<string, []>("op_47992_cast_fp16")];
            tensor<int32, [4]> k_427_perm_0 = const()[name = tensor<string, []>("k_427_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_47999_begin_0 = const()[name = tensor<string, []>("op_47999_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_47999_end_0 = const()[name = tensor<string, []>("op_47999_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_47999_end_mask_0 = const()[name = tensor<string, []>("op_47999_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> k_427_cast_fp16 = transpose(perm = k_427_perm_0, x = k_425_cast_fp16)[name = tensor<string, []>("transpose_33")];
            tensor<fp16, [2, 1024, 1, 64]> var_47999_cast_fp16 = slice_by_index(begin = var_47999_begin_0, end = var_47999_end_0, end_mask = var_47999_end_mask_0, x = k_427_cast_fp16)[name = tensor<string, []>("op_47999_cast_fp16")];
            tensor<int32, [4]> var_48003_begin_0 = const()[name = tensor<string, []>("op_48003_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_48003_end_0 = const()[name = tensor<string, []>("op_48003_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_48003_end_mask_0 = const()[name = tensor<string, []>("op_48003_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_48003_cast_fp16 = slice_by_index(begin = var_48003_begin_0, end = var_48003_end_0, end_mask = var_48003_end_mask_0, x = k_427_cast_fp16)[name = tensor<string, []>("op_48003_cast_fp16")];
            tensor<int32, [4]> var_48007_begin_0 = const()[name = tensor<string, []>("op_48007_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_48007_end_0 = const()[name = tensor<string, []>("op_48007_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_48007_end_mask_0 = const()[name = tensor<string, []>("op_48007_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_48007_cast_fp16 = slice_by_index(begin = var_48007_begin_0, end = var_48007_end_0, end_mask = var_48007_end_mask_0, x = k_427_cast_fp16)[name = tensor<string, []>("op_48007_cast_fp16")];
            tensor<int32, [4]> var_48011_begin_0 = const()[name = tensor<string, []>("op_48011_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_48011_end_0 = const()[name = tensor<string, []>("op_48011_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_48011_end_mask_0 = const()[name = tensor<string, []>("op_48011_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_48011_cast_fp16 = slice_by_index(begin = var_48011_begin_0, end = var_48011_end_0, end_mask = var_48011_end_mask_0, x = k_427_cast_fp16)[name = tensor<string, []>("op_48011_cast_fp16")];
            tensor<int32, [4]> var_48015_begin_0 = const()[name = tensor<string, []>("op_48015_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_48015_end_0 = const()[name = tensor<string, []>("op_48015_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_48015_end_mask_0 = const()[name = tensor<string, []>("op_48015_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_48015_cast_fp16 = slice_by_index(begin = var_48015_begin_0, end = var_48015_end_0, end_mask = var_48015_end_mask_0, x = k_427_cast_fp16)[name = tensor<string, []>("op_48015_cast_fp16")];
            tensor<int32, [4]> var_48019_begin_0 = const()[name = tensor<string, []>("op_48019_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_48019_end_0 = const()[name = tensor<string, []>("op_48019_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_48019_end_mask_0 = const()[name = tensor<string, []>("op_48019_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_48019_cast_fp16 = slice_by_index(begin = var_48019_begin_0, end = var_48019_end_0, end_mask = var_48019_end_mask_0, x = k_427_cast_fp16)[name = tensor<string, []>("op_48019_cast_fp16")];
            tensor<int32, [4]> var_48023_begin_0 = const()[name = tensor<string, []>("op_48023_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_48023_end_0 = const()[name = tensor<string, []>("op_48023_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_48023_end_mask_0 = const()[name = tensor<string, []>("op_48023_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_48023_cast_fp16 = slice_by_index(begin = var_48023_begin_0, end = var_48023_end_0, end_mask = var_48023_end_mask_0, x = k_427_cast_fp16)[name = tensor<string, []>("op_48023_cast_fp16")];
            tensor<int32, [4]> var_48027_begin_0 = const()[name = tensor<string, []>("op_48027_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_48027_end_0 = const()[name = tensor<string, []>("op_48027_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_48027_end_mask_0 = const()[name = tensor<string, []>("op_48027_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_48027_cast_fp16 = slice_by_index(begin = var_48027_begin_0, end = var_48027_end_0, end_mask = var_48027_end_mask_0, x = k_427_cast_fp16)[name = tensor<string, []>("op_48027_cast_fp16")];
            tensor<int32, [4]> var_48031_begin_0 = const()[name = tensor<string, []>("op_48031_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_48031_end_0 = const()[name = tensor<string, []>("op_48031_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_48031_end_mask_0 = const()[name = tensor<string, []>("op_48031_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_48031_cast_fp16 = slice_by_index(begin = var_48031_begin_0, end = var_48031_end_0, end_mask = var_48031_end_mask_0, x = k_427_cast_fp16)[name = tensor<string, []>("op_48031_cast_fp16")];
            tensor<int32, [4]> var_48035_begin_0 = const()[name = tensor<string, []>("op_48035_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_48035_end_0 = const()[name = tensor<string, []>("op_48035_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_48035_end_mask_0 = const()[name = tensor<string, []>("op_48035_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_48035_cast_fp16 = slice_by_index(begin = var_48035_begin_0, end = var_48035_end_0, end_mask = var_48035_end_mask_0, x = k_427_cast_fp16)[name = tensor<string, []>("op_48035_cast_fp16")];
            tensor<int32, [4]> var_48039_begin_0 = const()[name = tensor<string, []>("op_48039_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_48039_end_0 = const()[name = tensor<string, []>("op_48039_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_48039_end_mask_0 = const()[name = tensor<string, []>("op_48039_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_48039_cast_fp16 = slice_by_index(begin = var_48039_begin_0, end = var_48039_end_0, end_mask = var_48039_end_mask_0, x = k_427_cast_fp16)[name = tensor<string, []>("op_48039_cast_fp16")];
            tensor<int32, [4]> var_48043_begin_0 = const()[name = tensor<string, []>("op_48043_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_48043_end_0 = const()[name = tensor<string, []>("op_48043_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_48043_end_mask_0 = const()[name = tensor<string, []>("op_48043_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_48043_cast_fp16 = slice_by_index(begin = var_48043_begin_0, end = var_48043_end_0, end_mask = var_48043_end_mask_0, x = k_427_cast_fp16)[name = tensor<string, []>("op_48043_cast_fp16")];
            tensor<int32, [4]> var_48047_begin_0 = const()[name = tensor<string, []>("op_48047_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_48047_end_0 = const()[name = tensor<string, []>("op_48047_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_48047_end_mask_0 = const()[name = tensor<string, []>("op_48047_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_48047_cast_fp16 = slice_by_index(begin = var_48047_begin_0, end = var_48047_end_0, end_mask = var_48047_end_mask_0, x = k_427_cast_fp16)[name = tensor<string, []>("op_48047_cast_fp16")];
            tensor<int32, [4]> var_48051_begin_0 = const()[name = tensor<string, []>("op_48051_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_48051_end_0 = const()[name = tensor<string, []>("op_48051_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_48051_end_mask_0 = const()[name = tensor<string, []>("op_48051_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_48051_cast_fp16 = slice_by_index(begin = var_48051_begin_0, end = var_48051_end_0, end_mask = var_48051_end_mask_0, x = k_427_cast_fp16)[name = tensor<string, []>("op_48051_cast_fp16")];
            tensor<int32, [4]> var_48055_begin_0 = const()[name = tensor<string, []>("op_48055_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_48055_end_0 = const()[name = tensor<string, []>("op_48055_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_48055_end_mask_0 = const()[name = tensor<string, []>("op_48055_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_48055_cast_fp16 = slice_by_index(begin = var_48055_begin_0, end = var_48055_end_0, end_mask = var_48055_end_mask_0, x = k_427_cast_fp16)[name = tensor<string, []>("op_48055_cast_fp16")];
            tensor<int32, [4]> var_48059_begin_0 = const()[name = tensor<string, []>("op_48059_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_48059_end_0 = const()[name = tensor<string, []>("op_48059_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_48059_end_mask_0 = const()[name = tensor<string, []>("op_48059_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_48059_cast_fp16 = slice_by_index(begin = var_48059_begin_0, end = var_48059_end_0, end_mask = var_48059_end_mask_0, x = k_427_cast_fp16)[name = tensor<string, []>("op_48059_cast_fp16")];
            tensor<int32, [4]> var_48063_begin_0 = const()[name = tensor<string, []>("op_48063_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_48063_end_0 = const()[name = tensor<string, []>("op_48063_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_48063_end_mask_0 = const()[name = tensor<string, []>("op_48063_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_48063_cast_fp16 = slice_by_index(begin = var_48063_begin_0, end = var_48063_end_0, end_mask = var_48063_end_mask_0, x = k_427_cast_fp16)[name = tensor<string, []>("op_48063_cast_fp16")];
            tensor<int32, [4]> var_48067_begin_0 = const()[name = tensor<string, []>("op_48067_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_48067_end_0 = const()[name = tensor<string, []>("op_48067_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_48067_end_mask_0 = const()[name = tensor<string, []>("op_48067_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_48067_cast_fp16 = slice_by_index(begin = var_48067_begin_0, end = var_48067_end_0, end_mask = var_48067_end_mask_0, x = k_427_cast_fp16)[name = tensor<string, []>("op_48067_cast_fp16")];
            tensor<int32, [4]> var_48071_begin_0 = const()[name = tensor<string, []>("op_48071_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_48071_end_0 = const()[name = tensor<string, []>("op_48071_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_48071_end_mask_0 = const()[name = tensor<string, []>("op_48071_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_48071_cast_fp16 = slice_by_index(begin = var_48071_begin_0, end = var_48071_end_0, end_mask = var_48071_end_mask_0, x = k_427_cast_fp16)[name = tensor<string, []>("op_48071_cast_fp16")];
            tensor<int32, [4]> var_48075_begin_0 = const()[name = tensor<string, []>("op_48075_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_48075_end_0 = const()[name = tensor<string, []>("op_48075_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_48075_end_mask_0 = const()[name = tensor<string, []>("op_48075_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_48075_cast_fp16 = slice_by_index(begin = var_48075_begin_0, end = var_48075_end_0, end_mask = var_48075_end_mask_0, x = k_427_cast_fp16)[name = tensor<string, []>("op_48075_cast_fp16")];
            tensor<int32, [4]> var_48077_begin_0 = const()[name = tensor<string, []>("op_48077_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_48077_end_0 = const()[name = tensor<string, []>("op_48077_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_48077_end_mask_0 = const()[name = tensor<string, []>("op_48077_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48077_cast_fp16 = slice_by_index(begin = var_48077_begin_0, end = var_48077_end_0, end_mask = var_48077_end_mask_0, x = v_213_cast_fp16)[name = tensor<string, []>("op_48077_cast_fp16")];
            tensor<int32, [4]> var_48081_begin_0 = const()[name = tensor<string, []>("op_48081_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_48081_end_0 = const()[name = tensor<string, []>("op_48081_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_48081_end_mask_0 = const()[name = tensor<string, []>("op_48081_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48081_cast_fp16 = slice_by_index(begin = var_48081_begin_0, end = var_48081_end_0, end_mask = var_48081_end_mask_0, x = v_213_cast_fp16)[name = tensor<string, []>("op_48081_cast_fp16")];
            tensor<int32, [4]> var_48085_begin_0 = const()[name = tensor<string, []>("op_48085_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_48085_end_0 = const()[name = tensor<string, []>("op_48085_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_48085_end_mask_0 = const()[name = tensor<string, []>("op_48085_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48085_cast_fp16 = slice_by_index(begin = var_48085_begin_0, end = var_48085_end_0, end_mask = var_48085_end_mask_0, x = v_213_cast_fp16)[name = tensor<string, []>("op_48085_cast_fp16")];
            tensor<int32, [4]> var_48089_begin_0 = const()[name = tensor<string, []>("op_48089_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_48089_end_0 = const()[name = tensor<string, []>("op_48089_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_48089_end_mask_0 = const()[name = tensor<string, []>("op_48089_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48089_cast_fp16 = slice_by_index(begin = var_48089_begin_0, end = var_48089_end_0, end_mask = var_48089_end_mask_0, x = v_213_cast_fp16)[name = tensor<string, []>("op_48089_cast_fp16")];
            tensor<int32, [4]> var_48093_begin_0 = const()[name = tensor<string, []>("op_48093_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_48093_end_0 = const()[name = tensor<string, []>("op_48093_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_48093_end_mask_0 = const()[name = tensor<string, []>("op_48093_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48093_cast_fp16 = slice_by_index(begin = var_48093_begin_0, end = var_48093_end_0, end_mask = var_48093_end_mask_0, x = v_213_cast_fp16)[name = tensor<string, []>("op_48093_cast_fp16")];
            tensor<int32, [4]> var_48097_begin_0 = const()[name = tensor<string, []>("op_48097_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_48097_end_0 = const()[name = tensor<string, []>("op_48097_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_48097_end_mask_0 = const()[name = tensor<string, []>("op_48097_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48097_cast_fp16 = slice_by_index(begin = var_48097_begin_0, end = var_48097_end_0, end_mask = var_48097_end_mask_0, x = v_213_cast_fp16)[name = tensor<string, []>("op_48097_cast_fp16")];
            tensor<int32, [4]> var_48101_begin_0 = const()[name = tensor<string, []>("op_48101_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_48101_end_0 = const()[name = tensor<string, []>("op_48101_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_48101_end_mask_0 = const()[name = tensor<string, []>("op_48101_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48101_cast_fp16 = slice_by_index(begin = var_48101_begin_0, end = var_48101_end_0, end_mask = var_48101_end_mask_0, x = v_213_cast_fp16)[name = tensor<string, []>("op_48101_cast_fp16")];
            tensor<int32, [4]> var_48105_begin_0 = const()[name = tensor<string, []>("op_48105_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_48105_end_0 = const()[name = tensor<string, []>("op_48105_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_48105_end_mask_0 = const()[name = tensor<string, []>("op_48105_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48105_cast_fp16 = slice_by_index(begin = var_48105_begin_0, end = var_48105_end_0, end_mask = var_48105_end_mask_0, x = v_213_cast_fp16)[name = tensor<string, []>("op_48105_cast_fp16")];
            tensor<int32, [4]> var_48109_begin_0 = const()[name = tensor<string, []>("op_48109_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_48109_end_0 = const()[name = tensor<string, []>("op_48109_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_48109_end_mask_0 = const()[name = tensor<string, []>("op_48109_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48109_cast_fp16 = slice_by_index(begin = var_48109_begin_0, end = var_48109_end_0, end_mask = var_48109_end_mask_0, x = v_213_cast_fp16)[name = tensor<string, []>("op_48109_cast_fp16")];
            tensor<int32, [4]> var_48113_begin_0 = const()[name = tensor<string, []>("op_48113_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_48113_end_0 = const()[name = tensor<string, []>("op_48113_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_48113_end_mask_0 = const()[name = tensor<string, []>("op_48113_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48113_cast_fp16 = slice_by_index(begin = var_48113_begin_0, end = var_48113_end_0, end_mask = var_48113_end_mask_0, x = v_213_cast_fp16)[name = tensor<string, []>("op_48113_cast_fp16")];
            tensor<int32, [4]> var_48117_begin_0 = const()[name = tensor<string, []>("op_48117_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_48117_end_0 = const()[name = tensor<string, []>("op_48117_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_48117_end_mask_0 = const()[name = tensor<string, []>("op_48117_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48117_cast_fp16 = slice_by_index(begin = var_48117_begin_0, end = var_48117_end_0, end_mask = var_48117_end_mask_0, x = v_213_cast_fp16)[name = tensor<string, []>("op_48117_cast_fp16")];
            tensor<int32, [4]> var_48121_begin_0 = const()[name = tensor<string, []>("op_48121_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_48121_end_0 = const()[name = tensor<string, []>("op_48121_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_48121_end_mask_0 = const()[name = tensor<string, []>("op_48121_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48121_cast_fp16 = slice_by_index(begin = var_48121_begin_0, end = var_48121_end_0, end_mask = var_48121_end_mask_0, x = v_213_cast_fp16)[name = tensor<string, []>("op_48121_cast_fp16")];
            tensor<int32, [4]> var_48125_begin_0 = const()[name = tensor<string, []>("op_48125_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_48125_end_0 = const()[name = tensor<string, []>("op_48125_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_48125_end_mask_0 = const()[name = tensor<string, []>("op_48125_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48125_cast_fp16 = slice_by_index(begin = var_48125_begin_0, end = var_48125_end_0, end_mask = var_48125_end_mask_0, x = v_213_cast_fp16)[name = tensor<string, []>("op_48125_cast_fp16")];
            tensor<int32, [4]> var_48129_begin_0 = const()[name = tensor<string, []>("op_48129_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_48129_end_0 = const()[name = tensor<string, []>("op_48129_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_48129_end_mask_0 = const()[name = tensor<string, []>("op_48129_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48129_cast_fp16 = slice_by_index(begin = var_48129_begin_0, end = var_48129_end_0, end_mask = var_48129_end_mask_0, x = v_213_cast_fp16)[name = tensor<string, []>("op_48129_cast_fp16")];
            tensor<int32, [4]> var_48133_begin_0 = const()[name = tensor<string, []>("op_48133_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_48133_end_0 = const()[name = tensor<string, []>("op_48133_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_48133_end_mask_0 = const()[name = tensor<string, []>("op_48133_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48133_cast_fp16 = slice_by_index(begin = var_48133_begin_0, end = var_48133_end_0, end_mask = var_48133_end_mask_0, x = v_213_cast_fp16)[name = tensor<string, []>("op_48133_cast_fp16")];
            tensor<int32, [4]> var_48137_begin_0 = const()[name = tensor<string, []>("op_48137_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_48137_end_0 = const()[name = tensor<string, []>("op_48137_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_48137_end_mask_0 = const()[name = tensor<string, []>("op_48137_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48137_cast_fp16 = slice_by_index(begin = var_48137_begin_0, end = var_48137_end_0, end_mask = var_48137_end_mask_0, x = v_213_cast_fp16)[name = tensor<string, []>("op_48137_cast_fp16")];
            tensor<int32, [4]> var_48141_begin_0 = const()[name = tensor<string, []>("op_48141_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_48141_end_0 = const()[name = tensor<string, []>("op_48141_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_48141_end_mask_0 = const()[name = tensor<string, []>("op_48141_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48141_cast_fp16 = slice_by_index(begin = var_48141_begin_0, end = var_48141_end_0, end_mask = var_48141_end_mask_0, x = v_213_cast_fp16)[name = tensor<string, []>("op_48141_cast_fp16")];
            tensor<int32, [4]> var_48145_begin_0 = const()[name = tensor<string, []>("op_48145_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_48145_end_0 = const()[name = tensor<string, []>("op_48145_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_48145_end_mask_0 = const()[name = tensor<string, []>("op_48145_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48145_cast_fp16 = slice_by_index(begin = var_48145_begin_0, end = var_48145_end_0, end_mask = var_48145_end_mask_0, x = v_213_cast_fp16)[name = tensor<string, []>("op_48145_cast_fp16")];
            tensor<int32, [4]> var_48149_begin_0 = const()[name = tensor<string, []>("op_48149_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_48149_end_0 = const()[name = tensor<string, []>("op_48149_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_48149_end_mask_0 = const()[name = tensor<string, []>("op_48149_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48149_cast_fp16 = slice_by_index(begin = var_48149_begin_0, end = var_48149_end_0, end_mask = var_48149_end_mask_0, x = v_213_cast_fp16)[name = tensor<string, []>("op_48149_cast_fp16")];
            tensor<int32, [4]> var_48153_begin_0 = const()[name = tensor<string, []>("op_48153_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_48153_end_0 = const()[name = tensor<string, []>("op_48153_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_48153_end_mask_0 = const()[name = tensor<string, []>("op_48153_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48153_cast_fp16 = slice_by_index(begin = var_48153_begin_0, end = var_48153_end_0, end_mask = var_48153_end_mask_0, x = v_213_cast_fp16)[name = tensor<string, []>("op_48153_cast_fp16")];
            tensor<string, []> var_48157_equation_0 = const()[name = tensor<string, []>("op_48157_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48157_cast_fp16 = einsum(equation = var_48157_equation_0, values = (var_47999_cast_fp16, var_47916_cast_fp16))[name = tensor<string, []>("op_48157_cast_fp16")];
            tensor<fp16, []> var_48158_to_fp16 = const()[name = tensor<string, []>("op_48158_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4081_cast_fp16 = mul(x = var_48157_cast_fp16, y = var_48158_to_fp16)[name = tensor<string, []>("aw_4081_cast_fp16")];
            tensor<string, []> var_48161_equation_0 = const()[name = tensor<string, []>("op_48161_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48161_cast_fp16 = einsum(equation = var_48161_equation_0, values = (var_48003_cast_fp16, var_47920_cast_fp16))[name = tensor<string, []>("op_48161_cast_fp16")];
            tensor<fp16, []> var_48162_to_fp16 = const()[name = tensor<string, []>("op_48162_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4083_cast_fp16 = mul(x = var_48161_cast_fp16, y = var_48162_to_fp16)[name = tensor<string, []>("aw_4083_cast_fp16")];
            tensor<string, []> var_48165_equation_0 = const()[name = tensor<string, []>("op_48165_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48165_cast_fp16 = einsum(equation = var_48165_equation_0, values = (var_48007_cast_fp16, var_47924_cast_fp16))[name = tensor<string, []>("op_48165_cast_fp16")];
            tensor<fp16, []> var_48166_to_fp16 = const()[name = tensor<string, []>("op_48166_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4085_cast_fp16 = mul(x = var_48165_cast_fp16, y = var_48166_to_fp16)[name = tensor<string, []>("aw_4085_cast_fp16")];
            tensor<string, []> var_48169_equation_0 = const()[name = tensor<string, []>("op_48169_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48169_cast_fp16 = einsum(equation = var_48169_equation_0, values = (var_48011_cast_fp16, var_47928_cast_fp16))[name = tensor<string, []>("op_48169_cast_fp16")];
            tensor<fp16, []> var_48170_to_fp16 = const()[name = tensor<string, []>("op_48170_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4087_cast_fp16 = mul(x = var_48169_cast_fp16, y = var_48170_to_fp16)[name = tensor<string, []>("aw_4087_cast_fp16")];
            tensor<string, []> var_48173_equation_0 = const()[name = tensor<string, []>("op_48173_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48173_cast_fp16 = einsum(equation = var_48173_equation_0, values = (var_48015_cast_fp16, var_47932_cast_fp16))[name = tensor<string, []>("op_48173_cast_fp16")];
            tensor<fp16, []> var_48174_to_fp16 = const()[name = tensor<string, []>("op_48174_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4089_cast_fp16 = mul(x = var_48173_cast_fp16, y = var_48174_to_fp16)[name = tensor<string, []>("aw_4089_cast_fp16")];
            tensor<string, []> var_48177_equation_0 = const()[name = tensor<string, []>("op_48177_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48177_cast_fp16 = einsum(equation = var_48177_equation_0, values = (var_48019_cast_fp16, var_47936_cast_fp16))[name = tensor<string, []>("op_48177_cast_fp16")];
            tensor<fp16, []> var_48178_to_fp16 = const()[name = tensor<string, []>("op_48178_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4091_cast_fp16 = mul(x = var_48177_cast_fp16, y = var_48178_to_fp16)[name = tensor<string, []>("aw_4091_cast_fp16")];
            tensor<string, []> var_48181_equation_0 = const()[name = tensor<string, []>("op_48181_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48181_cast_fp16 = einsum(equation = var_48181_equation_0, values = (var_48023_cast_fp16, var_47940_cast_fp16))[name = tensor<string, []>("op_48181_cast_fp16")];
            tensor<fp16, []> var_48182_to_fp16 = const()[name = tensor<string, []>("op_48182_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4093_cast_fp16 = mul(x = var_48181_cast_fp16, y = var_48182_to_fp16)[name = tensor<string, []>("aw_4093_cast_fp16")];
            tensor<string, []> var_48185_equation_0 = const()[name = tensor<string, []>("op_48185_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48185_cast_fp16 = einsum(equation = var_48185_equation_0, values = (var_48027_cast_fp16, var_47944_cast_fp16))[name = tensor<string, []>("op_48185_cast_fp16")];
            tensor<fp16, []> var_48186_to_fp16 = const()[name = tensor<string, []>("op_48186_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4095_cast_fp16 = mul(x = var_48185_cast_fp16, y = var_48186_to_fp16)[name = tensor<string, []>("aw_4095_cast_fp16")];
            tensor<string, []> var_48189_equation_0 = const()[name = tensor<string, []>("op_48189_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48189_cast_fp16 = einsum(equation = var_48189_equation_0, values = (var_48031_cast_fp16, var_47948_cast_fp16))[name = tensor<string, []>("op_48189_cast_fp16")];
            tensor<fp16, []> var_48190_to_fp16 = const()[name = tensor<string, []>("op_48190_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4097_cast_fp16 = mul(x = var_48189_cast_fp16, y = var_48190_to_fp16)[name = tensor<string, []>("aw_4097_cast_fp16")];
            tensor<string, []> var_48193_equation_0 = const()[name = tensor<string, []>("op_48193_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48193_cast_fp16 = einsum(equation = var_48193_equation_0, values = (var_48035_cast_fp16, var_47952_cast_fp16))[name = tensor<string, []>("op_48193_cast_fp16")];
            tensor<fp16, []> var_48194_to_fp16 = const()[name = tensor<string, []>("op_48194_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4099_cast_fp16 = mul(x = var_48193_cast_fp16, y = var_48194_to_fp16)[name = tensor<string, []>("aw_4099_cast_fp16")];
            tensor<string, []> var_48197_equation_0 = const()[name = tensor<string, []>("op_48197_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48197_cast_fp16 = einsum(equation = var_48197_equation_0, values = (var_48039_cast_fp16, var_47956_cast_fp16))[name = tensor<string, []>("op_48197_cast_fp16")];
            tensor<fp16, []> var_48198_to_fp16 = const()[name = tensor<string, []>("op_48198_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4101_cast_fp16 = mul(x = var_48197_cast_fp16, y = var_48198_to_fp16)[name = tensor<string, []>("aw_4101_cast_fp16")];
            tensor<string, []> var_48201_equation_0 = const()[name = tensor<string, []>("op_48201_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48201_cast_fp16 = einsum(equation = var_48201_equation_0, values = (var_48043_cast_fp16, var_47960_cast_fp16))[name = tensor<string, []>("op_48201_cast_fp16")];
            tensor<fp16, []> var_48202_to_fp16 = const()[name = tensor<string, []>("op_48202_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4103_cast_fp16 = mul(x = var_48201_cast_fp16, y = var_48202_to_fp16)[name = tensor<string, []>("aw_4103_cast_fp16")];
            tensor<string, []> var_48205_equation_0 = const()[name = tensor<string, []>("op_48205_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48205_cast_fp16 = einsum(equation = var_48205_equation_0, values = (var_48047_cast_fp16, var_47964_cast_fp16))[name = tensor<string, []>("op_48205_cast_fp16")];
            tensor<fp16, []> var_48206_to_fp16 = const()[name = tensor<string, []>("op_48206_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4105_cast_fp16 = mul(x = var_48205_cast_fp16, y = var_48206_to_fp16)[name = tensor<string, []>("aw_4105_cast_fp16")];
            tensor<string, []> var_48209_equation_0 = const()[name = tensor<string, []>("op_48209_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48209_cast_fp16 = einsum(equation = var_48209_equation_0, values = (var_48051_cast_fp16, var_47968_cast_fp16))[name = tensor<string, []>("op_48209_cast_fp16")];
            tensor<fp16, []> var_48210_to_fp16 = const()[name = tensor<string, []>("op_48210_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4107_cast_fp16 = mul(x = var_48209_cast_fp16, y = var_48210_to_fp16)[name = tensor<string, []>("aw_4107_cast_fp16")];
            tensor<string, []> var_48213_equation_0 = const()[name = tensor<string, []>("op_48213_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48213_cast_fp16 = einsum(equation = var_48213_equation_0, values = (var_48055_cast_fp16, var_47972_cast_fp16))[name = tensor<string, []>("op_48213_cast_fp16")];
            tensor<fp16, []> var_48214_to_fp16 = const()[name = tensor<string, []>("op_48214_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4109_cast_fp16 = mul(x = var_48213_cast_fp16, y = var_48214_to_fp16)[name = tensor<string, []>("aw_4109_cast_fp16")];
            tensor<string, []> var_48217_equation_0 = const()[name = tensor<string, []>("op_48217_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48217_cast_fp16 = einsum(equation = var_48217_equation_0, values = (var_48059_cast_fp16, var_47976_cast_fp16))[name = tensor<string, []>("op_48217_cast_fp16")];
            tensor<fp16, []> var_48218_to_fp16 = const()[name = tensor<string, []>("op_48218_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4111_cast_fp16 = mul(x = var_48217_cast_fp16, y = var_48218_to_fp16)[name = tensor<string, []>("aw_4111_cast_fp16")];
            tensor<string, []> var_48221_equation_0 = const()[name = tensor<string, []>("op_48221_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48221_cast_fp16 = einsum(equation = var_48221_equation_0, values = (var_48063_cast_fp16, var_47980_cast_fp16))[name = tensor<string, []>("op_48221_cast_fp16")];
            tensor<fp16, []> var_48222_to_fp16 = const()[name = tensor<string, []>("op_48222_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4113_cast_fp16 = mul(x = var_48221_cast_fp16, y = var_48222_to_fp16)[name = tensor<string, []>("aw_4113_cast_fp16")];
            tensor<string, []> var_48225_equation_0 = const()[name = tensor<string, []>("op_48225_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48225_cast_fp16 = einsum(equation = var_48225_equation_0, values = (var_48067_cast_fp16, var_47984_cast_fp16))[name = tensor<string, []>("op_48225_cast_fp16")];
            tensor<fp16, []> var_48226_to_fp16 = const()[name = tensor<string, []>("op_48226_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4115_cast_fp16 = mul(x = var_48225_cast_fp16, y = var_48226_to_fp16)[name = tensor<string, []>("aw_4115_cast_fp16")];
            tensor<string, []> var_48229_equation_0 = const()[name = tensor<string, []>("op_48229_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48229_cast_fp16 = einsum(equation = var_48229_equation_0, values = (var_48071_cast_fp16, var_47988_cast_fp16))[name = tensor<string, []>("op_48229_cast_fp16")];
            tensor<fp16, []> var_48230_to_fp16 = const()[name = tensor<string, []>("op_48230_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4117_cast_fp16 = mul(x = var_48229_cast_fp16, y = var_48230_to_fp16)[name = tensor<string, []>("aw_4117_cast_fp16")];
            tensor<string, []> var_48233_equation_0 = const()[name = tensor<string, []>("op_48233_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48233_cast_fp16 = einsum(equation = var_48233_equation_0, values = (var_48075_cast_fp16, var_47992_cast_fp16))[name = tensor<string, []>("op_48233_cast_fp16")];
            tensor<fp16, []> var_48234_to_fp16 = const()[name = tensor<string, []>("op_48234_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4119_cast_fp16 = mul(x = var_48233_cast_fp16, y = var_48234_to_fp16)[name = tensor<string, []>("aw_4119_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48236_cast_fp16 = softmax(axis = var_30355, x = aw_4081_cast_fp16)[name = tensor<string, []>("op_48236_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48237_cast_fp16 = softmax(axis = var_30355, x = aw_4083_cast_fp16)[name = tensor<string, []>("op_48237_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48238_cast_fp16 = softmax(axis = var_30355, x = aw_4085_cast_fp16)[name = tensor<string, []>("op_48238_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48239_cast_fp16 = softmax(axis = var_30355, x = aw_4087_cast_fp16)[name = tensor<string, []>("op_48239_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48240_cast_fp16 = softmax(axis = var_30355, x = aw_4089_cast_fp16)[name = tensor<string, []>("op_48240_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48241_cast_fp16 = softmax(axis = var_30355, x = aw_4091_cast_fp16)[name = tensor<string, []>("op_48241_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48242_cast_fp16 = softmax(axis = var_30355, x = aw_4093_cast_fp16)[name = tensor<string, []>("op_48242_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48243_cast_fp16 = softmax(axis = var_30355, x = aw_4095_cast_fp16)[name = tensor<string, []>("op_48243_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48244_cast_fp16 = softmax(axis = var_30355, x = aw_4097_cast_fp16)[name = tensor<string, []>("op_48244_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48245_cast_fp16 = softmax(axis = var_30355, x = aw_4099_cast_fp16)[name = tensor<string, []>("op_48245_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48246_cast_fp16 = softmax(axis = var_30355, x = aw_4101_cast_fp16)[name = tensor<string, []>("op_48246_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48247_cast_fp16 = softmax(axis = var_30355, x = aw_4103_cast_fp16)[name = tensor<string, []>("op_48247_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48248_cast_fp16 = softmax(axis = var_30355, x = aw_4105_cast_fp16)[name = tensor<string, []>("op_48248_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48249_cast_fp16 = softmax(axis = var_30355, x = aw_4107_cast_fp16)[name = tensor<string, []>("op_48249_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48250_cast_fp16 = softmax(axis = var_30355, x = aw_4109_cast_fp16)[name = tensor<string, []>("op_48250_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48251_cast_fp16 = softmax(axis = var_30355, x = aw_4111_cast_fp16)[name = tensor<string, []>("op_48251_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48252_cast_fp16 = softmax(axis = var_30355, x = aw_4113_cast_fp16)[name = tensor<string, []>("op_48252_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48253_cast_fp16 = softmax(axis = var_30355, x = aw_4115_cast_fp16)[name = tensor<string, []>("op_48253_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48254_cast_fp16 = softmax(axis = var_30355, x = aw_4117_cast_fp16)[name = tensor<string, []>("op_48254_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48255_cast_fp16 = softmax(axis = var_30355, x = aw_4119_cast_fp16)[name = tensor<string, []>("op_48255_cast_fp16")];
            tensor<string, []> var_48257_equation_0 = const()[name = tensor<string, []>("op_48257_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48257_cast_fp16 = einsum(equation = var_48257_equation_0, values = (var_48077_cast_fp16, var_48236_cast_fp16))[name = tensor<string, []>("op_48257_cast_fp16")];
            tensor<string, []> var_48259_equation_0 = const()[name = tensor<string, []>("op_48259_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48259_cast_fp16 = einsum(equation = var_48259_equation_0, values = (var_48081_cast_fp16, var_48237_cast_fp16))[name = tensor<string, []>("op_48259_cast_fp16")];
            tensor<string, []> var_48261_equation_0 = const()[name = tensor<string, []>("op_48261_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48261_cast_fp16 = einsum(equation = var_48261_equation_0, values = (var_48085_cast_fp16, var_48238_cast_fp16))[name = tensor<string, []>("op_48261_cast_fp16")];
            tensor<string, []> var_48263_equation_0 = const()[name = tensor<string, []>("op_48263_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48263_cast_fp16 = einsum(equation = var_48263_equation_0, values = (var_48089_cast_fp16, var_48239_cast_fp16))[name = tensor<string, []>("op_48263_cast_fp16")];
            tensor<string, []> var_48265_equation_0 = const()[name = tensor<string, []>("op_48265_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48265_cast_fp16 = einsum(equation = var_48265_equation_0, values = (var_48093_cast_fp16, var_48240_cast_fp16))[name = tensor<string, []>("op_48265_cast_fp16")];
            tensor<string, []> var_48267_equation_0 = const()[name = tensor<string, []>("op_48267_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48267_cast_fp16 = einsum(equation = var_48267_equation_0, values = (var_48097_cast_fp16, var_48241_cast_fp16))[name = tensor<string, []>("op_48267_cast_fp16")];
            tensor<string, []> var_48269_equation_0 = const()[name = tensor<string, []>("op_48269_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48269_cast_fp16 = einsum(equation = var_48269_equation_0, values = (var_48101_cast_fp16, var_48242_cast_fp16))[name = tensor<string, []>("op_48269_cast_fp16")];
            tensor<string, []> var_48271_equation_0 = const()[name = tensor<string, []>("op_48271_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48271_cast_fp16 = einsum(equation = var_48271_equation_0, values = (var_48105_cast_fp16, var_48243_cast_fp16))[name = tensor<string, []>("op_48271_cast_fp16")];
            tensor<string, []> var_48273_equation_0 = const()[name = tensor<string, []>("op_48273_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48273_cast_fp16 = einsum(equation = var_48273_equation_0, values = (var_48109_cast_fp16, var_48244_cast_fp16))[name = tensor<string, []>("op_48273_cast_fp16")];
            tensor<string, []> var_48275_equation_0 = const()[name = tensor<string, []>("op_48275_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48275_cast_fp16 = einsum(equation = var_48275_equation_0, values = (var_48113_cast_fp16, var_48245_cast_fp16))[name = tensor<string, []>("op_48275_cast_fp16")];
            tensor<string, []> var_48277_equation_0 = const()[name = tensor<string, []>("op_48277_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48277_cast_fp16 = einsum(equation = var_48277_equation_0, values = (var_48117_cast_fp16, var_48246_cast_fp16))[name = tensor<string, []>("op_48277_cast_fp16")];
            tensor<string, []> var_48279_equation_0 = const()[name = tensor<string, []>("op_48279_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48279_cast_fp16 = einsum(equation = var_48279_equation_0, values = (var_48121_cast_fp16, var_48247_cast_fp16))[name = tensor<string, []>("op_48279_cast_fp16")];
            tensor<string, []> var_48281_equation_0 = const()[name = tensor<string, []>("op_48281_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48281_cast_fp16 = einsum(equation = var_48281_equation_0, values = (var_48125_cast_fp16, var_48248_cast_fp16))[name = tensor<string, []>("op_48281_cast_fp16")];
            tensor<string, []> var_48283_equation_0 = const()[name = tensor<string, []>("op_48283_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48283_cast_fp16 = einsum(equation = var_48283_equation_0, values = (var_48129_cast_fp16, var_48249_cast_fp16))[name = tensor<string, []>("op_48283_cast_fp16")];
            tensor<string, []> var_48285_equation_0 = const()[name = tensor<string, []>("op_48285_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48285_cast_fp16 = einsum(equation = var_48285_equation_0, values = (var_48133_cast_fp16, var_48250_cast_fp16))[name = tensor<string, []>("op_48285_cast_fp16")];
            tensor<string, []> var_48287_equation_0 = const()[name = tensor<string, []>("op_48287_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48287_cast_fp16 = einsum(equation = var_48287_equation_0, values = (var_48137_cast_fp16, var_48251_cast_fp16))[name = tensor<string, []>("op_48287_cast_fp16")];
            tensor<string, []> var_48289_equation_0 = const()[name = tensor<string, []>("op_48289_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48289_cast_fp16 = einsum(equation = var_48289_equation_0, values = (var_48141_cast_fp16, var_48252_cast_fp16))[name = tensor<string, []>("op_48289_cast_fp16")];
            tensor<string, []> var_48291_equation_0 = const()[name = tensor<string, []>("op_48291_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48291_cast_fp16 = einsum(equation = var_48291_equation_0, values = (var_48145_cast_fp16, var_48253_cast_fp16))[name = tensor<string, []>("op_48291_cast_fp16")];
            tensor<string, []> var_48293_equation_0 = const()[name = tensor<string, []>("op_48293_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48293_cast_fp16 = einsum(equation = var_48293_equation_0, values = (var_48149_cast_fp16, var_48254_cast_fp16))[name = tensor<string, []>("op_48293_cast_fp16")];
            tensor<string, []> var_48295_equation_0 = const()[name = tensor<string, []>("op_48295_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48295_cast_fp16 = einsum(equation = var_48295_equation_0, values = (var_48153_cast_fp16, var_48255_cast_fp16))[name = tensor<string, []>("op_48295_cast_fp16")];
            tensor<bool, []> input_609_interleave_0 = const()[name = tensor<string, []>("input_609_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_609_cast_fp16 = concat(axis = var_30355, interleave = input_609_interleave_0, values = (var_48257_cast_fp16, var_48259_cast_fp16, var_48261_cast_fp16, var_48263_cast_fp16, var_48265_cast_fp16, var_48267_cast_fp16, var_48269_cast_fp16, var_48271_cast_fp16, var_48273_cast_fp16, var_48275_cast_fp16, var_48277_cast_fp16, var_48279_cast_fp16, var_48281_cast_fp16, var_48283_cast_fp16, var_48285_cast_fp16, var_48287_cast_fp16, var_48289_cast_fp16, var_48291_cast_fp16, var_48293_cast_fp16, var_48295_cast_fp16))[name = tensor<string, []>("input_609_cast_fp16")];
            tensor<string, []> var_48305_pad_type_0 = const()[name = tensor<string, []>("op_48305_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_48305_strides_0 = const()[name = tensor<string, []>("op_48305_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_48305_pad_0 = const()[name = tensor<string, []>("op_48305_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_48305_dilations_0 = const()[name = tensor<string, []>("op_48305_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_48305_groups_0 = const()[name = tensor<string, []>("op_48305_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_9_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(542617664))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(543846528))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_9_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_9_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_9_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(543846720)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_48305_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_9_attn1_to_out_0_bias_to_fp16, dilations = var_48305_dilations_0, groups = var_48305_groups_0, pad = var_48305_pad_0, pad_type = var_48305_pad_type_0, strides = var_48305_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_9_attn1_to_out_0_weight_to_fp16_palettized, x = input_609_cast_fp16)[name = tensor<string, []>("op_48305_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_321_cast_fp16 = add(x = var_48305_cast_fp16, y = inputs_319_cast_fp16)[name = tensor<string, []>("inputs_321_cast_fp16")];
            tensor<int32, [1]> hidden_states_417_axes_0 = const()[name = tensor<string, []>("hidden_states_417_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_417_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_417_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(543849344)))];
            tensor<fp16, [1280]> hidden_states_417_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_417_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(543851968)))];
            tensor<fp16, []> var_48315_to_fp16 = const()[name = tensor<string, []>("op_48315_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_417_cast_fp16 = layer_norm(axes = hidden_states_417_axes_0, beta = hidden_states_417_beta_0_to_fp16, epsilon = var_48315_to_fp16, gamma = hidden_states_417_gamma_0_to_fp16, x = inputs_321_cast_fp16)[name = tensor<string, []>("hidden_states_417_cast_fp16")];
            tensor<string, []> q_215_pad_type_0 = const()[name = tensor<string, []>("q_215_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_215_strides_0 = const()[name = tensor<string, []>("q_215_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_215_pad_0 = const()[name = tensor<string, []>("q_215_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_215_dilations_0 = const()[name = tensor<string, []>("q_215_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_215_groups_0 = const()[name = tensor<string, []>("q_215_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_9_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(543854592))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(545083456))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_9_attn2_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_215_cast_fp16 = conv(dilations = q_215_dilations_0, groups = q_215_groups_0, pad = q_215_pad_0, pad_type = q_215_pad_type_0, strides = q_215_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_9_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_417_cast_fp16)[name = tensor<string, []>("q_215_cast_fp16")];
            tensor<string, []> k_429_pad_type_0 = const()[name = tensor<string, []>("k_429_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_429_strides_0 = const()[name = tensor<string, []>("k_429_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_429_pad_0 = const()[name = tensor<string, []>("k_429_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_429_dilations_0 = const()[name = tensor<string, []>("k_429_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_429_groups_0 = const()[name = tensor<string, []>("k_429_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_9_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(545083648))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(547049792))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_9_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> k_429_cast_fp16 = conv(dilations = k_429_dilations_0, groups = k_429_groups_0, pad = k_429_pad_0, pad_type = k_429_pad_type_0, strides = k_429_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_9_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("k_429_cast_fp16")];
            tensor<string, []> v_215_pad_type_0 = const()[name = tensor<string, []>("v_215_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_215_strides_0 = const()[name = tensor<string, []>("v_215_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_215_pad_0 = const()[name = tensor<string, []>("v_215_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_215_dilations_0 = const()[name = tensor<string, []>("v_215_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_215_groups_0 = const()[name = tensor<string, []>("v_215_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_9_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(547049984))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(549016128))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_9_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> v_215_cast_fp16 = conv(dilations = v_215_dilations_0, groups = v_215_groups_0, pad = v_215_pad_0, pad_type = v_215_pad_type_0, strides = v_215_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_9_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("v_215_cast_fp16")];
            tensor<int32, [4]> var_48348_begin_0 = const()[name = tensor<string, []>("op_48348_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_48348_end_0 = const()[name = tensor<string, []>("op_48348_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_48348_end_mask_0 = const()[name = tensor<string, []>("op_48348_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48348_cast_fp16 = slice_by_index(begin = var_48348_begin_0, end = var_48348_end_0, end_mask = var_48348_end_mask_0, x = q_215_cast_fp16)[name = tensor<string, []>("op_48348_cast_fp16")];
            tensor<int32, [4]> var_48352_begin_0 = const()[name = tensor<string, []>("op_48352_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_48352_end_0 = const()[name = tensor<string, []>("op_48352_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_48352_end_mask_0 = const()[name = tensor<string, []>("op_48352_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48352_cast_fp16 = slice_by_index(begin = var_48352_begin_0, end = var_48352_end_0, end_mask = var_48352_end_mask_0, x = q_215_cast_fp16)[name = tensor<string, []>("op_48352_cast_fp16")];
            tensor<int32, [4]> var_48356_begin_0 = const()[name = tensor<string, []>("op_48356_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_48356_end_0 = const()[name = tensor<string, []>("op_48356_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_48356_end_mask_0 = const()[name = tensor<string, []>("op_48356_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48356_cast_fp16 = slice_by_index(begin = var_48356_begin_0, end = var_48356_end_0, end_mask = var_48356_end_mask_0, x = q_215_cast_fp16)[name = tensor<string, []>("op_48356_cast_fp16")];
            tensor<int32, [4]> var_48360_begin_0 = const()[name = tensor<string, []>("op_48360_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_48360_end_0 = const()[name = tensor<string, []>("op_48360_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_48360_end_mask_0 = const()[name = tensor<string, []>("op_48360_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48360_cast_fp16 = slice_by_index(begin = var_48360_begin_0, end = var_48360_end_0, end_mask = var_48360_end_mask_0, x = q_215_cast_fp16)[name = tensor<string, []>("op_48360_cast_fp16")];
            tensor<int32, [4]> var_48364_begin_0 = const()[name = tensor<string, []>("op_48364_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_48364_end_0 = const()[name = tensor<string, []>("op_48364_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_48364_end_mask_0 = const()[name = tensor<string, []>("op_48364_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48364_cast_fp16 = slice_by_index(begin = var_48364_begin_0, end = var_48364_end_0, end_mask = var_48364_end_mask_0, x = q_215_cast_fp16)[name = tensor<string, []>("op_48364_cast_fp16")];
            tensor<int32, [4]> var_48368_begin_0 = const()[name = tensor<string, []>("op_48368_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_48368_end_0 = const()[name = tensor<string, []>("op_48368_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_48368_end_mask_0 = const()[name = tensor<string, []>("op_48368_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48368_cast_fp16 = slice_by_index(begin = var_48368_begin_0, end = var_48368_end_0, end_mask = var_48368_end_mask_0, x = q_215_cast_fp16)[name = tensor<string, []>("op_48368_cast_fp16")];
            tensor<int32, [4]> var_48372_begin_0 = const()[name = tensor<string, []>("op_48372_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_48372_end_0 = const()[name = tensor<string, []>("op_48372_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_48372_end_mask_0 = const()[name = tensor<string, []>("op_48372_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48372_cast_fp16 = slice_by_index(begin = var_48372_begin_0, end = var_48372_end_0, end_mask = var_48372_end_mask_0, x = q_215_cast_fp16)[name = tensor<string, []>("op_48372_cast_fp16")];
            tensor<int32, [4]> var_48376_begin_0 = const()[name = tensor<string, []>("op_48376_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_48376_end_0 = const()[name = tensor<string, []>("op_48376_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_48376_end_mask_0 = const()[name = tensor<string, []>("op_48376_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48376_cast_fp16 = slice_by_index(begin = var_48376_begin_0, end = var_48376_end_0, end_mask = var_48376_end_mask_0, x = q_215_cast_fp16)[name = tensor<string, []>("op_48376_cast_fp16")];
            tensor<int32, [4]> var_48380_begin_0 = const()[name = tensor<string, []>("op_48380_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_48380_end_0 = const()[name = tensor<string, []>("op_48380_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_48380_end_mask_0 = const()[name = tensor<string, []>("op_48380_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48380_cast_fp16 = slice_by_index(begin = var_48380_begin_0, end = var_48380_end_0, end_mask = var_48380_end_mask_0, x = q_215_cast_fp16)[name = tensor<string, []>("op_48380_cast_fp16")];
            tensor<int32, [4]> var_48384_begin_0 = const()[name = tensor<string, []>("op_48384_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_48384_end_0 = const()[name = tensor<string, []>("op_48384_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_48384_end_mask_0 = const()[name = tensor<string, []>("op_48384_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48384_cast_fp16 = slice_by_index(begin = var_48384_begin_0, end = var_48384_end_0, end_mask = var_48384_end_mask_0, x = q_215_cast_fp16)[name = tensor<string, []>("op_48384_cast_fp16")];
            tensor<int32, [4]> var_48388_begin_0 = const()[name = tensor<string, []>("op_48388_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_48388_end_0 = const()[name = tensor<string, []>("op_48388_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_48388_end_mask_0 = const()[name = tensor<string, []>("op_48388_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48388_cast_fp16 = slice_by_index(begin = var_48388_begin_0, end = var_48388_end_0, end_mask = var_48388_end_mask_0, x = q_215_cast_fp16)[name = tensor<string, []>("op_48388_cast_fp16")];
            tensor<int32, [4]> var_48392_begin_0 = const()[name = tensor<string, []>("op_48392_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_48392_end_0 = const()[name = tensor<string, []>("op_48392_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_48392_end_mask_0 = const()[name = tensor<string, []>("op_48392_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48392_cast_fp16 = slice_by_index(begin = var_48392_begin_0, end = var_48392_end_0, end_mask = var_48392_end_mask_0, x = q_215_cast_fp16)[name = tensor<string, []>("op_48392_cast_fp16")];
            tensor<int32, [4]> var_48396_begin_0 = const()[name = tensor<string, []>("op_48396_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_48396_end_0 = const()[name = tensor<string, []>("op_48396_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_48396_end_mask_0 = const()[name = tensor<string, []>("op_48396_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48396_cast_fp16 = slice_by_index(begin = var_48396_begin_0, end = var_48396_end_0, end_mask = var_48396_end_mask_0, x = q_215_cast_fp16)[name = tensor<string, []>("op_48396_cast_fp16")];
            tensor<int32, [4]> var_48400_begin_0 = const()[name = tensor<string, []>("op_48400_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_48400_end_0 = const()[name = tensor<string, []>("op_48400_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_48400_end_mask_0 = const()[name = tensor<string, []>("op_48400_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48400_cast_fp16 = slice_by_index(begin = var_48400_begin_0, end = var_48400_end_0, end_mask = var_48400_end_mask_0, x = q_215_cast_fp16)[name = tensor<string, []>("op_48400_cast_fp16")];
            tensor<int32, [4]> var_48404_begin_0 = const()[name = tensor<string, []>("op_48404_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_48404_end_0 = const()[name = tensor<string, []>("op_48404_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_48404_end_mask_0 = const()[name = tensor<string, []>("op_48404_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48404_cast_fp16 = slice_by_index(begin = var_48404_begin_0, end = var_48404_end_0, end_mask = var_48404_end_mask_0, x = q_215_cast_fp16)[name = tensor<string, []>("op_48404_cast_fp16")];
            tensor<int32, [4]> var_48408_begin_0 = const()[name = tensor<string, []>("op_48408_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_48408_end_0 = const()[name = tensor<string, []>("op_48408_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_48408_end_mask_0 = const()[name = tensor<string, []>("op_48408_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48408_cast_fp16 = slice_by_index(begin = var_48408_begin_0, end = var_48408_end_0, end_mask = var_48408_end_mask_0, x = q_215_cast_fp16)[name = tensor<string, []>("op_48408_cast_fp16")];
            tensor<int32, [4]> var_48412_begin_0 = const()[name = tensor<string, []>("op_48412_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_48412_end_0 = const()[name = tensor<string, []>("op_48412_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_48412_end_mask_0 = const()[name = tensor<string, []>("op_48412_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48412_cast_fp16 = slice_by_index(begin = var_48412_begin_0, end = var_48412_end_0, end_mask = var_48412_end_mask_0, x = q_215_cast_fp16)[name = tensor<string, []>("op_48412_cast_fp16")];
            tensor<int32, [4]> var_48416_begin_0 = const()[name = tensor<string, []>("op_48416_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_48416_end_0 = const()[name = tensor<string, []>("op_48416_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_48416_end_mask_0 = const()[name = tensor<string, []>("op_48416_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48416_cast_fp16 = slice_by_index(begin = var_48416_begin_0, end = var_48416_end_0, end_mask = var_48416_end_mask_0, x = q_215_cast_fp16)[name = tensor<string, []>("op_48416_cast_fp16")];
            tensor<int32, [4]> var_48420_begin_0 = const()[name = tensor<string, []>("op_48420_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_48420_end_0 = const()[name = tensor<string, []>("op_48420_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_48420_end_mask_0 = const()[name = tensor<string, []>("op_48420_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48420_cast_fp16 = slice_by_index(begin = var_48420_begin_0, end = var_48420_end_0, end_mask = var_48420_end_mask_0, x = q_215_cast_fp16)[name = tensor<string, []>("op_48420_cast_fp16")];
            tensor<int32, [4]> var_48424_begin_0 = const()[name = tensor<string, []>("op_48424_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_48424_end_0 = const()[name = tensor<string, []>("op_48424_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_48424_end_mask_0 = const()[name = tensor<string, []>("op_48424_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48424_cast_fp16 = slice_by_index(begin = var_48424_begin_0, end = var_48424_end_0, end_mask = var_48424_end_mask_0, x = q_215_cast_fp16)[name = tensor<string, []>("op_48424_cast_fp16")];
            tensor<int32, [4]> k_431_perm_0 = const()[name = tensor<string, []>("k_431_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_48431_begin_0 = const()[name = tensor<string, []>("op_48431_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_48431_end_0 = const()[name = tensor<string, []>("op_48431_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_48431_end_mask_0 = const()[name = tensor<string, []>("op_48431_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> k_431_cast_fp16 = transpose(perm = k_431_perm_0, x = k_429_cast_fp16)[name = tensor<string, []>("transpose_32")];
            tensor<fp16, [2, 77, 1, 64]> var_48431_cast_fp16 = slice_by_index(begin = var_48431_begin_0, end = var_48431_end_0, end_mask = var_48431_end_mask_0, x = k_431_cast_fp16)[name = tensor<string, []>("op_48431_cast_fp16")];
            tensor<int32, [4]> var_48435_begin_0 = const()[name = tensor<string, []>("op_48435_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_48435_end_0 = const()[name = tensor<string, []>("op_48435_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_48435_end_mask_0 = const()[name = tensor<string, []>("op_48435_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_48435_cast_fp16 = slice_by_index(begin = var_48435_begin_0, end = var_48435_end_0, end_mask = var_48435_end_mask_0, x = k_431_cast_fp16)[name = tensor<string, []>("op_48435_cast_fp16")];
            tensor<int32, [4]> var_48439_begin_0 = const()[name = tensor<string, []>("op_48439_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_48439_end_0 = const()[name = tensor<string, []>("op_48439_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_48439_end_mask_0 = const()[name = tensor<string, []>("op_48439_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_48439_cast_fp16 = slice_by_index(begin = var_48439_begin_0, end = var_48439_end_0, end_mask = var_48439_end_mask_0, x = k_431_cast_fp16)[name = tensor<string, []>("op_48439_cast_fp16")];
            tensor<int32, [4]> var_48443_begin_0 = const()[name = tensor<string, []>("op_48443_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_48443_end_0 = const()[name = tensor<string, []>("op_48443_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_48443_end_mask_0 = const()[name = tensor<string, []>("op_48443_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_48443_cast_fp16 = slice_by_index(begin = var_48443_begin_0, end = var_48443_end_0, end_mask = var_48443_end_mask_0, x = k_431_cast_fp16)[name = tensor<string, []>("op_48443_cast_fp16")];
            tensor<int32, [4]> var_48447_begin_0 = const()[name = tensor<string, []>("op_48447_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_48447_end_0 = const()[name = tensor<string, []>("op_48447_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_48447_end_mask_0 = const()[name = tensor<string, []>("op_48447_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_48447_cast_fp16 = slice_by_index(begin = var_48447_begin_0, end = var_48447_end_0, end_mask = var_48447_end_mask_0, x = k_431_cast_fp16)[name = tensor<string, []>("op_48447_cast_fp16")];
            tensor<int32, [4]> var_48451_begin_0 = const()[name = tensor<string, []>("op_48451_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_48451_end_0 = const()[name = tensor<string, []>("op_48451_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_48451_end_mask_0 = const()[name = tensor<string, []>("op_48451_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_48451_cast_fp16 = slice_by_index(begin = var_48451_begin_0, end = var_48451_end_0, end_mask = var_48451_end_mask_0, x = k_431_cast_fp16)[name = tensor<string, []>("op_48451_cast_fp16")];
            tensor<int32, [4]> var_48455_begin_0 = const()[name = tensor<string, []>("op_48455_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_48455_end_0 = const()[name = tensor<string, []>("op_48455_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_48455_end_mask_0 = const()[name = tensor<string, []>("op_48455_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_48455_cast_fp16 = slice_by_index(begin = var_48455_begin_0, end = var_48455_end_0, end_mask = var_48455_end_mask_0, x = k_431_cast_fp16)[name = tensor<string, []>("op_48455_cast_fp16")];
            tensor<int32, [4]> var_48459_begin_0 = const()[name = tensor<string, []>("op_48459_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_48459_end_0 = const()[name = tensor<string, []>("op_48459_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_48459_end_mask_0 = const()[name = tensor<string, []>("op_48459_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_48459_cast_fp16 = slice_by_index(begin = var_48459_begin_0, end = var_48459_end_0, end_mask = var_48459_end_mask_0, x = k_431_cast_fp16)[name = tensor<string, []>("op_48459_cast_fp16")];
            tensor<int32, [4]> var_48463_begin_0 = const()[name = tensor<string, []>("op_48463_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_48463_end_0 = const()[name = tensor<string, []>("op_48463_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_48463_end_mask_0 = const()[name = tensor<string, []>("op_48463_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_48463_cast_fp16 = slice_by_index(begin = var_48463_begin_0, end = var_48463_end_0, end_mask = var_48463_end_mask_0, x = k_431_cast_fp16)[name = tensor<string, []>("op_48463_cast_fp16")];
            tensor<int32, [4]> var_48467_begin_0 = const()[name = tensor<string, []>("op_48467_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_48467_end_0 = const()[name = tensor<string, []>("op_48467_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_48467_end_mask_0 = const()[name = tensor<string, []>("op_48467_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_48467_cast_fp16 = slice_by_index(begin = var_48467_begin_0, end = var_48467_end_0, end_mask = var_48467_end_mask_0, x = k_431_cast_fp16)[name = tensor<string, []>("op_48467_cast_fp16")];
            tensor<int32, [4]> var_48471_begin_0 = const()[name = tensor<string, []>("op_48471_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_48471_end_0 = const()[name = tensor<string, []>("op_48471_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_48471_end_mask_0 = const()[name = tensor<string, []>("op_48471_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_48471_cast_fp16 = slice_by_index(begin = var_48471_begin_0, end = var_48471_end_0, end_mask = var_48471_end_mask_0, x = k_431_cast_fp16)[name = tensor<string, []>("op_48471_cast_fp16")];
            tensor<int32, [4]> var_48475_begin_0 = const()[name = tensor<string, []>("op_48475_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_48475_end_0 = const()[name = tensor<string, []>("op_48475_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_48475_end_mask_0 = const()[name = tensor<string, []>("op_48475_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_48475_cast_fp16 = slice_by_index(begin = var_48475_begin_0, end = var_48475_end_0, end_mask = var_48475_end_mask_0, x = k_431_cast_fp16)[name = tensor<string, []>("op_48475_cast_fp16")];
            tensor<int32, [4]> var_48479_begin_0 = const()[name = tensor<string, []>("op_48479_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_48479_end_0 = const()[name = tensor<string, []>("op_48479_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_48479_end_mask_0 = const()[name = tensor<string, []>("op_48479_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_48479_cast_fp16 = slice_by_index(begin = var_48479_begin_0, end = var_48479_end_0, end_mask = var_48479_end_mask_0, x = k_431_cast_fp16)[name = tensor<string, []>("op_48479_cast_fp16")];
            tensor<int32, [4]> var_48483_begin_0 = const()[name = tensor<string, []>("op_48483_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_48483_end_0 = const()[name = tensor<string, []>("op_48483_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_48483_end_mask_0 = const()[name = tensor<string, []>("op_48483_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_48483_cast_fp16 = slice_by_index(begin = var_48483_begin_0, end = var_48483_end_0, end_mask = var_48483_end_mask_0, x = k_431_cast_fp16)[name = tensor<string, []>("op_48483_cast_fp16")];
            tensor<int32, [4]> var_48487_begin_0 = const()[name = tensor<string, []>("op_48487_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_48487_end_0 = const()[name = tensor<string, []>("op_48487_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_48487_end_mask_0 = const()[name = tensor<string, []>("op_48487_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_48487_cast_fp16 = slice_by_index(begin = var_48487_begin_0, end = var_48487_end_0, end_mask = var_48487_end_mask_0, x = k_431_cast_fp16)[name = tensor<string, []>("op_48487_cast_fp16")];
            tensor<int32, [4]> var_48491_begin_0 = const()[name = tensor<string, []>("op_48491_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_48491_end_0 = const()[name = tensor<string, []>("op_48491_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_48491_end_mask_0 = const()[name = tensor<string, []>("op_48491_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_48491_cast_fp16 = slice_by_index(begin = var_48491_begin_0, end = var_48491_end_0, end_mask = var_48491_end_mask_0, x = k_431_cast_fp16)[name = tensor<string, []>("op_48491_cast_fp16")];
            tensor<int32, [4]> var_48495_begin_0 = const()[name = tensor<string, []>("op_48495_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_48495_end_0 = const()[name = tensor<string, []>("op_48495_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_48495_end_mask_0 = const()[name = tensor<string, []>("op_48495_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_48495_cast_fp16 = slice_by_index(begin = var_48495_begin_0, end = var_48495_end_0, end_mask = var_48495_end_mask_0, x = k_431_cast_fp16)[name = tensor<string, []>("op_48495_cast_fp16")];
            tensor<int32, [4]> var_48499_begin_0 = const()[name = tensor<string, []>("op_48499_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_48499_end_0 = const()[name = tensor<string, []>("op_48499_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_48499_end_mask_0 = const()[name = tensor<string, []>("op_48499_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_48499_cast_fp16 = slice_by_index(begin = var_48499_begin_0, end = var_48499_end_0, end_mask = var_48499_end_mask_0, x = k_431_cast_fp16)[name = tensor<string, []>("op_48499_cast_fp16")];
            tensor<int32, [4]> var_48503_begin_0 = const()[name = tensor<string, []>("op_48503_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_48503_end_0 = const()[name = tensor<string, []>("op_48503_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_48503_end_mask_0 = const()[name = tensor<string, []>("op_48503_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_48503_cast_fp16 = slice_by_index(begin = var_48503_begin_0, end = var_48503_end_0, end_mask = var_48503_end_mask_0, x = k_431_cast_fp16)[name = tensor<string, []>("op_48503_cast_fp16")];
            tensor<int32, [4]> var_48507_begin_0 = const()[name = tensor<string, []>("op_48507_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_48507_end_0 = const()[name = tensor<string, []>("op_48507_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_48507_end_mask_0 = const()[name = tensor<string, []>("op_48507_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_48507_cast_fp16 = slice_by_index(begin = var_48507_begin_0, end = var_48507_end_0, end_mask = var_48507_end_mask_0, x = k_431_cast_fp16)[name = tensor<string, []>("op_48507_cast_fp16")];
            tensor<int32, [4]> var_48509_begin_0 = const()[name = tensor<string, []>("op_48509_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_48509_end_0 = const()[name = tensor<string, []>("op_48509_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_48509_end_mask_0 = const()[name = tensor<string, []>("op_48509_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_48509_cast_fp16 = slice_by_index(begin = var_48509_begin_0, end = var_48509_end_0, end_mask = var_48509_end_mask_0, x = v_215_cast_fp16)[name = tensor<string, []>("op_48509_cast_fp16")];
            tensor<int32, [4]> var_48513_begin_0 = const()[name = tensor<string, []>("op_48513_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_48513_end_0 = const()[name = tensor<string, []>("op_48513_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_48513_end_mask_0 = const()[name = tensor<string, []>("op_48513_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_48513_cast_fp16 = slice_by_index(begin = var_48513_begin_0, end = var_48513_end_0, end_mask = var_48513_end_mask_0, x = v_215_cast_fp16)[name = tensor<string, []>("op_48513_cast_fp16")];
            tensor<int32, [4]> var_48517_begin_0 = const()[name = tensor<string, []>("op_48517_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_48517_end_0 = const()[name = tensor<string, []>("op_48517_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_48517_end_mask_0 = const()[name = tensor<string, []>("op_48517_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_48517_cast_fp16 = slice_by_index(begin = var_48517_begin_0, end = var_48517_end_0, end_mask = var_48517_end_mask_0, x = v_215_cast_fp16)[name = tensor<string, []>("op_48517_cast_fp16")];
            tensor<int32, [4]> var_48521_begin_0 = const()[name = tensor<string, []>("op_48521_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_48521_end_0 = const()[name = tensor<string, []>("op_48521_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_48521_end_mask_0 = const()[name = tensor<string, []>("op_48521_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_48521_cast_fp16 = slice_by_index(begin = var_48521_begin_0, end = var_48521_end_0, end_mask = var_48521_end_mask_0, x = v_215_cast_fp16)[name = tensor<string, []>("op_48521_cast_fp16")];
            tensor<int32, [4]> var_48525_begin_0 = const()[name = tensor<string, []>("op_48525_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_48525_end_0 = const()[name = tensor<string, []>("op_48525_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_48525_end_mask_0 = const()[name = tensor<string, []>("op_48525_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_48525_cast_fp16 = slice_by_index(begin = var_48525_begin_0, end = var_48525_end_0, end_mask = var_48525_end_mask_0, x = v_215_cast_fp16)[name = tensor<string, []>("op_48525_cast_fp16")];
            tensor<int32, [4]> var_48529_begin_0 = const()[name = tensor<string, []>("op_48529_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_48529_end_0 = const()[name = tensor<string, []>("op_48529_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_48529_end_mask_0 = const()[name = tensor<string, []>("op_48529_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_48529_cast_fp16 = slice_by_index(begin = var_48529_begin_0, end = var_48529_end_0, end_mask = var_48529_end_mask_0, x = v_215_cast_fp16)[name = tensor<string, []>("op_48529_cast_fp16")];
            tensor<int32, [4]> var_48533_begin_0 = const()[name = tensor<string, []>("op_48533_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_48533_end_0 = const()[name = tensor<string, []>("op_48533_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_48533_end_mask_0 = const()[name = tensor<string, []>("op_48533_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_48533_cast_fp16 = slice_by_index(begin = var_48533_begin_0, end = var_48533_end_0, end_mask = var_48533_end_mask_0, x = v_215_cast_fp16)[name = tensor<string, []>("op_48533_cast_fp16")];
            tensor<int32, [4]> var_48537_begin_0 = const()[name = tensor<string, []>("op_48537_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_48537_end_0 = const()[name = tensor<string, []>("op_48537_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_48537_end_mask_0 = const()[name = tensor<string, []>("op_48537_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_48537_cast_fp16 = slice_by_index(begin = var_48537_begin_0, end = var_48537_end_0, end_mask = var_48537_end_mask_0, x = v_215_cast_fp16)[name = tensor<string, []>("op_48537_cast_fp16")];
            tensor<int32, [4]> var_48541_begin_0 = const()[name = tensor<string, []>("op_48541_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_48541_end_0 = const()[name = tensor<string, []>("op_48541_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_48541_end_mask_0 = const()[name = tensor<string, []>("op_48541_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_48541_cast_fp16 = slice_by_index(begin = var_48541_begin_0, end = var_48541_end_0, end_mask = var_48541_end_mask_0, x = v_215_cast_fp16)[name = tensor<string, []>("op_48541_cast_fp16")];
            tensor<int32, [4]> var_48545_begin_0 = const()[name = tensor<string, []>("op_48545_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_48545_end_0 = const()[name = tensor<string, []>("op_48545_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_48545_end_mask_0 = const()[name = tensor<string, []>("op_48545_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_48545_cast_fp16 = slice_by_index(begin = var_48545_begin_0, end = var_48545_end_0, end_mask = var_48545_end_mask_0, x = v_215_cast_fp16)[name = tensor<string, []>("op_48545_cast_fp16")];
            tensor<int32, [4]> var_48549_begin_0 = const()[name = tensor<string, []>("op_48549_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_48549_end_0 = const()[name = tensor<string, []>("op_48549_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_48549_end_mask_0 = const()[name = tensor<string, []>("op_48549_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_48549_cast_fp16 = slice_by_index(begin = var_48549_begin_0, end = var_48549_end_0, end_mask = var_48549_end_mask_0, x = v_215_cast_fp16)[name = tensor<string, []>("op_48549_cast_fp16")];
            tensor<int32, [4]> var_48553_begin_0 = const()[name = tensor<string, []>("op_48553_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_48553_end_0 = const()[name = tensor<string, []>("op_48553_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_48553_end_mask_0 = const()[name = tensor<string, []>("op_48553_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_48553_cast_fp16 = slice_by_index(begin = var_48553_begin_0, end = var_48553_end_0, end_mask = var_48553_end_mask_0, x = v_215_cast_fp16)[name = tensor<string, []>("op_48553_cast_fp16")];
            tensor<int32, [4]> var_48557_begin_0 = const()[name = tensor<string, []>("op_48557_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_48557_end_0 = const()[name = tensor<string, []>("op_48557_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_48557_end_mask_0 = const()[name = tensor<string, []>("op_48557_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_48557_cast_fp16 = slice_by_index(begin = var_48557_begin_0, end = var_48557_end_0, end_mask = var_48557_end_mask_0, x = v_215_cast_fp16)[name = tensor<string, []>("op_48557_cast_fp16")];
            tensor<int32, [4]> var_48561_begin_0 = const()[name = tensor<string, []>("op_48561_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_48561_end_0 = const()[name = tensor<string, []>("op_48561_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_48561_end_mask_0 = const()[name = tensor<string, []>("op_48561_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_48561_cast_fp16 = slice_by_index(begin = var_48561_begin_0, end = var_48561_end_0, end_mask = var_48561_end_mask_0, x = v_215_cast_fp16)[name = tensor<string, []>("op_48561_cast_fp16")];
            tensor<int32, [4]> var_48565_begin_0 = const()[name = tensor<string, []>("op_48565_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_48565_end_0 = const()[name = tensor<string, []>("op_48565_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_48565_end_mask_0 = const()[name = tensor<string, []>("op_48565_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_48565_cast_fp16 = slice_by_index(begin = var_48565_begin_0, end = var_48565_end_0, end_mask = var_48565_end_mask_0, x = v_215_cast_fp16)[name = tensor<string, []>("op_48565_cast_fp16")];
            tensor<int32, [4]> var_48569_begin_0 = const()[name = tensor<string, []>("op_48569_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_48569_end_0 = const()[name = tensor<string, []>("op_48569_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_48569_end_mask_0 = const()[name = tensor<string, []>("op_48569_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_48569_cast_fp16 = slice_by_index(begin = var_48569_begin_0, end = var_48569_end_0, end_mask = var_48569_end_mask_0, x = v_215_cast_fp16)[name = tensor<string, []>("op_48569_cast_fp16")];
            tensor<int32, [4]> var_48573_begin_0 = const()[name = tensor<string, []>("op_48573_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_48573_end_0 = const()[name = tensor<string, []>("op_48573_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_48573_end_mask_0 = const()[name = tensor<string, []>("op_48573_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_48573_cast_fp16 = slice_by_index(begin = var_48573_begin_0, end = var_48573_end_0, end_mask = var_48573_end_mask_0, x = v_215_cast_fp16)[name = tensor<string, []>("op_48573_cast_fp16")];
            tensor<int32, [4]> var_48577_begin_0 = const()[name = tensor<string, []>("op_48577_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_48577_end_0 = const()[name = tensor<string, []>("op_48577_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_48577_end_mask_0 = const()[name = tensor<string, []>("op_48577_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_48577_cast_fp16 = slice_by_index(begin = var_48577_begin_0, end = var_48577_end_0, end_mask = var_48577_end_mask_0, x = v_215_cast_fp16)[name = tensor<string, []>("op_48577_cast_fp16")];
            tensor<int32, [4]> var_48581_begin_0 = const()[name = tensor<string, []>("op_48581_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_48581_end_0 = const()[name = tensor<string, []>("op_48581_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_48581_end_mask_0 = const()[name = tensor<string, []>("op_48581_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_48581_cast_fp16 = slice_by_index(begin = var_48581_begin_0, end = var_48581_end_0, end_mask = var_48581_end_mask_0, x = v_215_cast_fp16)[name = tensor<string, []>("op_48581_cast_fp16")];
            tensor<int32, [4]> var_48585_begin_0 = const()[name = tensor<string, []>("op_48585_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_48585_end_0 = const()[name = tensor<string, []>("op_48585_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_48585_end_mask_0 = const()[name = tensor<string, []>("op_48585_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_48585_cast_fp16 = slice_by_index(begin = var_48585_begin_0, end = var_48585_end_0, end_mask = var_48585_end_mask_0, x = v_215_cast_fp16)[name = tensor<string, []>("op_48585_cast_fp16")];
            tensor<string, []> var_48589_equation_0 = const()[name = tensor<string, []>("op_48589_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_48589_cast_fp16 = einsum(equation = var_48589_equation_0, values = (var_48431_cast_fp16, var_48348_cast_fp16))[name = tensor<string, []>("op_48589_cast_fp16")];
            tensor<fp16, []> var_48590_to_fp16 = const()[name = tensor<string, []>("op_48590_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4121_cast_fp16 = mul(x = var_48589_cast_fp16, y = var_48590_to_fp16)[name = tensor<string, []>("aw_4121_cast_fp16")];
            tensor<string, []> var_48593_equation_0 = const()[name = tensor<string, []>("op_48593_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_48593_cast_fp16 = einsum(equation = var_48593_equation_0, values = (var_48435_cast_fp16, var_48352_cast_fp16))[name = tensor<string, []>("op_48593_cast_fp16")];
            tensor<fp16, []> var_48594_to_fp16 = const()[name = tensor<string, []>("op_48594_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4123_cast_fp16 = mul(x = var_48593_cast_fp16, y = var_48594_to_fp16)[name = tensor<string, []>("aw_4123_cast_fp16")];
            tensor<string, []> var_48597_equation_0 = const()[name = tensor<string, []>("op_48597_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_48597_cast_fp16 = einsum(equation = var_48597_equation_0, values = (var_48439_cast_fp16, var_48356_cast_fp16))[name = tensor<string, []>("op_48597_cast_fp16")];
            tensor<fp16, []> var_48598_to_fp16 = const()[name = tensor<string, []>("op_48598_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4125_cast_fp16 = mul(x = var_48597_cast_fp16, y = var_48598_to_fp16)[name = tensor<string, []>("aw_4125_cast_fp16")];
            tensor<string, []> var_48601_equation_0 = const()[name = tensor<string, []>("op_48601_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_48601_cast_fp16 = einsum(equation = var_48601_equation_0, values = (var_48443_cast_fp16, var_48360_cast_fp16))[name = tensor<string, []>("op_48601_cast_fp16")];
            tensor<fp16, []> var_48602_to_fp16 = const()[name = tensor<string, []>("op_48602_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4127_cast_fp16 = mul(x = var_48601_cast_fp16, y = var_48602_to_fp16)[name = tensor<string, []>("aw_4127_cast_fp16")];
            tensor<string, []> var_48605_equation_0 = const()[name = tensor<string, []>("op_48605_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_48605_cast_fp16 = einsum(equation = var_48605_equation_0, values = (var_48447_cast_fp16, var_48364_cast_fp16))[name = tensor<string, []>("op_48605_cast_fp16")];
            tensor<fp16, []> var_48606_to_fp16 = const()[name = tensor<string, []>("op_48606_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4129_cast_fp16 = mul(x = var_48605_cast_fp16, y = var_48606_to_fp16)[name = tensor<string, []>("aw_4129_cast_fp16")];
            tensor<string, []> var_48609_equation_0 = const()[name = tensor<string, []>("op_48609_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_48609_cast_fp16 = einsum(equation = var_48609_equation_0, values = (var_48451_cast_fp16, var_48368_cast_fp16))[name = tensor<string, []>("op_48609_cast_fp16")];
            tensor<fp16, []> var_48610_to_fp16 = const()[name = tensor<string, []>("op_48610_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4131_cast_fp16 = mul(x = var_48609_cast_fp16, y = var_48610_to_fp16)[name = tensor<string, []>("aw_4131_cast_fp16")];
            tensor<string, []> var_48613_equation_0 = const()[name = tensor<string, []>("op_48613_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_48613_cast_fp16 = einsum(equation = var_48613_equation_0, values = (var_48455_cast_fp16, var_48372_cast_fp16))[name = tensor<string, []>("op_48613_cast_fp16")];
            tensor<fp16, []> var_48614_to_fp16 = const()[name = tensor<string, []>("op_48614_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4133_cast_fp16 = mul(x = var_48613_cast_fp16, y = var_48614_to_fp16)[name = tensor<string, []>("aw_4133_cast_fp16")];
            tensor<string, []> var_48617_equation_0 = const()[name = tensor<string, []>("op_48617_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_48617_cast_fp16 = einsum(equation = var_48617_equation_0, values = (var_48459_cast_fp16, var_48376_cast_fp16))[name = tensor<string, []>("op_48617_cast_fp16")];
            tensor<fp16, []> var_48618_to_fp16 = const()[name = tensor<string, []>("op_48618_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4135_cast_fp16 = mul(x = var_48617_cast_fp16, y = var_48618_to_fp16)[name = tensor<string, []>("aw_4135_cast_fp16")];
            tensor<string, []> var_48621_equation_0 = const()[name = tensor<string, []>("op_48621_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_48621_cast_fp16 = einsum(equation = var_48621_equation_0, values = (var_48463_cast_fp16, var_48380_cast_fp16))[name = tensor<string, []>("op_48621_cast_fp16")];
            tensor<fp16, []> var_48622_to_fp16 = const()[name = tensor<string, []>("op_48622_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4137_cast_fp16 = mul(x = var_48621_cast_fp16, y = var_48622_to_fp16)[name = tensor<string, []>("aw_4137_cast_fp16")];
            tensor<string, []> var_48625_equation_0 = const()[name = tensor<string, []>("op_48625_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_48625_cast_fp16 = einsum(equation = var_48625_equation_0, values = (var_48467_cast_fp16, var_48384_cast_fp16))[name = tensor<string, []>("op_48625_cast_fp16")];
            tensor<fp16, []> var_48626_to_fp16 = const()[name = tensor<string, []>("op_48626_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4139_cast_fp16 = mul(x = var_48625_cast_fp16, y = var_48626_to_fp16)[name = tensor<string, []>("aw_4139_cast_fp16")];
            tensor<string, []> var_48629_equation_0 = const()[name = tensor<string, []>("op_48629_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_48629_cast_fp16 = einsum(equation = var_48629_equation_0, values = (var_48471_cast_fp16, var_48388_cast_fp16))[name = tensor<string, []>("op_48629_cast_fp16")];
            tensor<fp16, []> var_48630_to_fp16 = const()[name = tensor<string, []>("op_48630_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4141_cast_fp16 = mul(x = var_48629_cast_fp16, y = var_48630_to_fp16)[name = tensor<string, []>("aw_4141_cast_fp16")];
            tensor<string, []> var_48633_equation_0 = const()[name = tensor<string, []>("op_48633_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_48633_cast_fp16 = einsum(equation = var_48633_equation_0, values = (var_48475_cast_fp16, var_48392_cast_fp16))[name = tensor<string, []>("op_48633_cast_fp16")];
            tensor<fp16, []> var_48634_to_fp16 = const()[name = tensor<string, []>("op_48634_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4143_cast_fp16 = mul(x = var_48633_cast_fp16, y = var_48634_to_fp16)[name = tensor<string, []>("aw_4143_cast_fp16")];
            tensor<string, []> var_48637_equation_0 = const()[name = tensor<string, []>("op_48637_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_48637_cast_fp16 = einsum(equation = var_48637_equation_0, values = (var_48479_cast_fp16, var_48396_cast_fp16))[name = tensor<string, []>("op_48637_cast_fp16")];
            tensor<fp16, []> var_48638_to_fp16 = const()[name = tensor<string, []>("op_48638_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4145_cast_fp16 = mul(x = var_48637_cast_fp16, y = var_48638_to_fp16)[name = tensor<string, []>("aw_4145_cast_fp16")];
            tensor<string, []> var_48641_equation_0 = const()[name = tensor<string, []>("op_48641_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_48641_cast_fp16 = einsum(equation = var_48641_equation_0, values = (var_48483_cast_fp16, var_48400_cast_fp16))[name = tensor<string, []>("op_48641_cast_fp16")];
            tensor<fp16, []> var_48642_to_fp16 = const()[name = tensor<string, []>("op_48642_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4147_cast_fp16 = mul(x = var_48641_cast_fp16, y = var_48642_to_fp16)[name = tensor<string, []>("aw_4147_cast_fp16")];
            tensor<string, []> var_48645_equation_0 = const()[name = tensor<string, []>("op_48645_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_48645_cast_fp16 = einsum(equation = var_48645_equation_0, values = (var_48487_cast_fp16, var_48404_cast_fp16))[name = tensor<string, []>("op_48645_cast_fp16")];
            tensor<fp16, []> var_48646_to_fp16 = const()[name = tensor<string, []>("op_48646_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4149_cast_fp16 = mul(x = var_48645_cast_fp16, y = var_48646_to_fp16)[name = tensor<string, []>("aw_4149_cast_fp16")];
            tensor<string, []> var_48649_equation_0 = const()[name = tensor<string, []>("op_48649_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_48649_cast_fp16 = einsum(equation = var_48649_equation_0, values = (var_48491_cast_fp16, var_48408_cast_fp16))[name = tensor<string, []>("op_48649_cast_fp16")];
            tensor<fp16, []> var_48650_to_fp16 = const()[name = tensor<string, []>("op_48650_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4151_cast_fp16 = mul(x = var_48649_cast_fp16, y = var_48650_to_fp16)[name = tensor<string, []>("aw_4151_cast_fp16")];
            tensor<string, []> var_48653_equation_0 = const()[name = tensor<string, []>("op_48653_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_48653_cast_fp16 = einsum(equation = var_48653_equation_0, values = (var_48495_cast_fp16, var_48412_cast_fp16))[name = tensor<string, []>("op_48653_cast_fp16")];
            tensor<fp16, []> var_48654_to_fp16 = const()[name = tensor<string, []>("op_48654_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4153_cast_fp16 = mul(x = var_48653_cast_fp16, y = var_48654_to_fp16)[name = tensor<string, []>("aw_4153_cast_fp16")];
            tensor<string, []> var_48657_equation_0 = const()[name = tensor<string, []>("op_48657_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_48657_cast_fp16 = einsum(equation = var_48657_equation_0, values = (var_48499_cast_fp16, var_48416_cast_fp16))[name = tensor<string, []>("op_48657_cast_fp16")];
            tensor<fp16, []> var_48658_to_fp16 = const()[name = tensor<string, []>("op_48658_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4155_cast_fp16 = mul(x = var_48657_cast_fp16, y = var_48658_to_fp16)[name = tensor<string, []>("aw_4155_cast_fp16")];
            tensor<string, []> var_48661_equation_0 = const()[name = tensor<string, []>("op_48661_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_48661_cast_fp16 = einsum(equation = var_48661_equation_0, values = (var_48503_cast_fp16, var_48420_cast_fp16))[name = tensor<string, []>("op_48661_cast_fp16")];
            tensor<fp16, []> var_48662_to_fp16 = const()[name = tensor<string, []>("op_48662_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4157_cast_fp16 = mul(x = var_48661_cast_fp16, y = var_48662_to_fp16)[name = tensor<string, []>("aw_4157_cast_fp16")];
            tensor<string, []> var_48665_equation_0 = const()[name = tensor<string, []>("op_48665_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_48665_cast_fp16 = einsum(equation = var_48665_equation_0, values = (var_48507_cast_fp16, var_48424_cast_fp16))[name = tensor<string, []>("op_48665_cast_fp16")];
            tensor<fp16, []> var_48666_to_fp16 = const()[name = tensor<string, []>("op_48666_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4159_cast_fp16 = mul(x = var_48665_cast_fp16, y = var_48666_to_fp16)[name = tensor<string, []>("aw_4159_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_48668_cast_fp16 = softmax(axis = var_30355, x = aw_4121_cast_fp16)[name = tensor<string, []>("op_48668_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_48669_cast_fp16 = softmax(axis = var_30355, x = aw_4123_cast_fp16)[name = tensor<string, []>("op_48669_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_48670_cast_fp16 = softmax(axis = var_30355, x = aw_4125_cast_fp16)[name = tensor<string, []>("op_48670_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_48671_cast_fp16 = softmax(axis = var_30355, x = aw_4127_cast_fp16)[name = tensor<string, []>("op_48671_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_48672_cast_fp16 = softmax(axis = var_30355, x = aw_4129_cast_fp16)[name = tensor<string, []>("op_48672_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_48673_cast_fp16 = softmax(axis = var_30355, x = aw_4131_cast_fp16)[name = tensor<string, []>("op_48673_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_48674_cast_fp16 = softmax(axis = var_30355, x = aw_4133_cast_fp16)[name = tensor<string, []>("op_48674_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_48675_cast_fp16 = softmax(axis = var_30355, x = aw_4135_cast_fp16)[name = tensor<string, []>("op_48675_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_48676_cast_fp16 = softmax(axis = var_30355, x = aw_4137_cast_fp16)[name = tensor<string, []>("op_48676_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_48677_cast_fp16 = softmax(axis = var_30355, x = aw_4139_cast_fp16)[name = tensor<string, []>("op_48677_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_48678_cast_fp16 = softmax(axis = var_30355, x = aw_4141_cast_fp16)[name = tensor<string, []>("op_48678_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_48679_cast_fp16 = softmax(axis = var_30355, x = aw_4143_cast_fp16)[name = tensor<string, []>("op_48679_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_48680_cast_fp16 = softmax(axis = var_30355, x = aw_4145_cast_fp16)[name = tensor<string, []>("op_48680_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_48681_cast_fp16 = softmax(axis = var_30355, x = aw_4147_cast_fp16)[name = tensor<string, []>("op_48681_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_48682_cast_fp16 = softmax(axis = var_30355, x = aw_4149_cast_fp16)[name = tensor<string, []>("op_48682_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_48683_cast_fp16 = softmax(axis = var_30355, x = aw_4151_cast_fp16)[name = tensor<string, []>("op_48683_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_48684_cast_fp16 = softmax(axis = var_30355, x = aw_4153_cast_fp16)[name = tensor<string, []>("op_48684_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_48685_cast_fp16 = softmax(axis = var_30355, x = aw_4155_cast_fp16)[name = tensor<string, []>("op_48685_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_48686_cast_fp16 = softmax(axis = var_30355, x = aw_4157_cast_fp16)[name = tensor<string, []>("op_48686_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_48687_cast_fp16 = softmax(axis = var_30355, x = aw_4159_cast_fp16)[name = tensor<string, []>("op_48687_cast_fp16")];
            tensor<string, []> var_48689_equation_0 = const()[name = tensor<string, []>("op_48689_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48689_cast_fp16 = einsum(equation = var_48689_equation_0, values = (var_48509_cast_fp16, var_48668_cast_fp16))[name = tensor<string, []>("op_48689_cast_fp16")];
            tensor<string, []> var_48691_equation_0 = const()[name = tensor<string, []>("op_48691_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48691_cast_fp16 = einsum(equation = var_48691_equation_0, values = (var_48513_cast_fp16, var_48669_cast_fp16))[name = tensor<string, []>("op_48691_cast_fp16")];
            tensor<string, []> var_48693_equation_0 = const()[name = tensor<string, []>("op_48693_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48693_cast_fp16 = einsum(equation = var_48693_equation_0, values = (var_48517_cast_fp16, var_48670_cast_fp16))[name = tensor<string, []>("op_48693_cast_fp16")];
            tensor<string, []> var_48695_equation_0 = const()[name = tensor<string, []>("op_48695_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48695_cast_fp16 = einsum(equation = var_48695_equation_0, values = (var_48521_cast_fp16, var_48671_cast_fp16))[name = tensor<string, []>("op_48695_cast_fp16")];
            tensor<string, []> var_48697_equation_0 = const()[name = tensor<string, []>("op_48697_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48697_cast_fp16 = einsum(equation = var_48697_equation_0, values = (var_48525_cast_fp16, var_48672_cast_fp16))[name = tensor<string, []>("op_48697_cast_fp16")];
            tensor<string, []> var_48699_equation_0 = const()[name = tensor<string, []>("op_48699_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48699_cast_fp16 = einsum(equation = var_48699_equation_0, values = (var_48529_cast_fp16, var_48673_cast_fp16))[name = tensor<string, []>("op_48699_cast_fp16")];
            tensor<string, []> var_48701_equation_0 = const()[name = tensor<string, []>("op_48701_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48701_cast_fp16 = einsum(equation = var_48701_equation_0, values = (var_48533_cast_fp16, var_48674_cast_fp16))[name = tensor<string, []>("op_48701_cast_fp16")];
            tensor<string, []> var_48703_equation_0 = const()[name = tensor<string, []>("op_48703_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48703_cast_fp16 = einsum(equation = var_48703_equation_0, values = (var_48537_cast_fp16, var_48675_cast_fp16))[name = tensor<string, []>("op_48703_cast_fp16")];
            tensor<string, []> var_48705_equation_0 = const()[name = tensor<string, []>("op_48705_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48705_cast_fp16 = einsum(equation = var_48705_equation_0, values = (var_48541_cast_fp16, var_48676_cast_fp16))[name = tensor<string, []>("op_48705_cast_fp16")];
            tensor<string, []> var_48707_equation_0 = const()[name = tensor<string, []>("op_48707_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48707_cast_fp16 = einsum(equation = var_48707_equation_0, values = (var_48545_cast_fp16, var_48677_cast_fp16))[name = tensor<string, []>("op_48707_cast_fp16")];
            tensor<string, []> var_48709_equation_0 = const()[name = tensor<string, []>("op_48709_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48709_cast_fp16 = einsum(equation = var_48709_equation_0, values = (var_48549_cast_fp16, var_48678_cast_fp16))[name = tensor<string, []>("op_48709_cast_fp16")];
            tensor<string, []> var_48711_equation_0 = const()[name = tensor<string, []>("op_48711_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48711_cast_fp16 = einsum(equation = var_48711_equation_0, values = (var_48553_cast_fp16, var_48679_cast_fp16))[name = tensor<string, []>("op_48711_cast_fp16")];
            tensor<string, []> var_48713_equation_0 = const()[name = tensor<string, []>("op_48713_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48713_cast_fp16 = einsum(equation = var_48713_equation_0, values = (var_48557_cast_fp16, var_48680_cast_fp16))[name = tensor<string, []>("op_48713_cast_fp16")];
            tensor<string, []> var_48715_equation_0 = const()[name = tensor<string, []>("op_48715_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48715_cast_fp16 = einsum(equation = var_48715_equation_0, values = (var_48561_cast_fp16, var_48681_cast_fp16))[name = tensor<string, []>("op_48715_cast_fp16")];
            tensor<string, []> var_48717_equation_0 = const()[name = tensor<string, []>("op_48717_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48717_cast_fp16 = einsum(equation = var_48717_equation_0, values = (var_48565_cast_fp16, var_48682_cast_fp16))[name = tensor<string, []>("op_48717_cast_fp16")];
            tensor<string, []> var_48719_equation_0 = const()[name = tensor<string, []>("op_48719_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48719_cast_fp16 = einsum(equation = var_48719_equation_0, values = (var_48569_cast_fp16, var_48683_cast_fp16))[name = tensor<string, []>("op_48719_cast_fp16")];
            tensor<string, []> var_48721_equation_0 = const()[name = tensor<string, []>("op_48721_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48721_cast_fp16 = einsum(equation = var_48721_equation_0, values = (var_48573_cast_fp16, var_48684_cast_fp16))[name = tensor<string, []>("op_48721_cast_fp16")];
            tensor<string, []> var_48723_equation_0 = const()[name = tensor<string, []>("op_48723_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48723_cast_fp16 = einsum(equation = var_48723_equation_0, values = (var_48577_cast_fp16, var_48685_cast_fp16))[name = tensor<string, []>("op_48723_cast_fp16")];
            tensor<string, []> var_48725_equation_0 = const()[name = tensor<string, []>("op_48725_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48725_cast_fp16 = einsum(equation = var_48725_equation_0, values = (var_48581_cast_fp16, var_48686_cast_fp16))[name = tensor<string, []>("op_48725_cast_fp16")];
            tensor<string, []> var_48727_equation_0 = const()[name = tensor<string, []>("op_48727_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48727_cast_fp16 = einsum(equation = var_48727_equation_0, values = (var_48585_cast_fp16, var_48687_cast_fp16))[name = tensor<string, []>("op_48727_cast_fp16")];
            tensor<bool, []> input_611_interleave_0 = const()[name = tensor<string, []>("input_611_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_611_cast_fp16 = concat(axis = var_30355, interleave = input_611_interleave_0, values = (var_48689_cast_fp16, var_48691_cast_fp16, var_48693_cast_fp16, var_48695_cast_fp16, var_48697_cast_fp16, var_48699_cast_fp16, var_48701_cast_fp16, var_48703_cast_fp16, var_48705_cast_fp16, var_48707_cast_fp16, var_48709_cast_fp16, var_48711_cast_fp16, var_48713_cast_fp16, var_48715_cast_fp16, var_48717_cast_fp16, var_48719_cast_fp16, var_48721_cast_fp16, var_48723_cast_fp16, var_48725_cast_fp16, var_48727_cast_fp16))[name = tensor<string, []>("input_611_cast_fp16")];
            tensor<string, []> var_48737_pad_type_0 = const()[name = tensor<string, []>("op_48737_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_48737_strides_0 = const()[name = tensor<string, []>("op_48737_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_48737_pad_0 = const()[name = tensor<string, []>("op_48737_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_48737_dilations_0 = const()[name = tensor<string, []>("op_48737_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_48737_groups_0 = const()[name = tensor<string, []>("op_48737_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_9_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(549016320))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(550245184))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_9_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_9_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_9_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(550245376)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_48737_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_9_attn2_to_out_0_bias_to_fp16, dilations = var_48737_dilations_0, groups = var_48737_groups_0, pad = var_48737_pad_0, pad_type = var_48737_pad_type_0, strides = var_48737_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_9_attn2_to_out_0_weight_to_fp16_palettized, x = input_611_cast_fp16)[name = tensor<string, []>("op_48737_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_323_cast_fp16 = add(x = var_48737_cast_fp16, y = inputs_321_cast_fp16)[name = tensor<string, []>("inputs_323_cast_fp16")];
            tensor<int32, [1]> input_613_axes_0 = const()[name = tensor<string, []>("input_613_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> input_613_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_613_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(550248000)))];
            tensor<fp16, [1280]> input_613_beta_0_to_fp16 = const()[name = tensor<string, []>("input_613_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(550250624)))];
            tensor<fp16, []> var_48747_to_fp16 = const()[name = tensor<string, []>("op_48747_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> input_613_cast_fp16 = layer_norm(axes = input_613_axes_0, beta = input_613_beta_0_to_fp16, epsilon = var_48747_to_fp16, gamma = input_613_gamma_0_to_fp16, x = inputs_323_cast_fp16)[name = tensor<string, []>("input_613_cast_fp16")];
            tensor<string, []> var_48767_pad_type_0 = const()[name = tensor<string, []>("op_48767_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_48767_strides_0 = const()[name = tensor<string, []>("op_48767_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_48767_pad_0 = const()[name = tensor<string, []>("op_48767_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_48767_dilations_0 = const()[name = tensor<string, []>("op_48767_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_48767_groups_0 = const()[name = tensor<string, []>("op_48767_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_9_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [9830400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(550253248))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(560083712))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_9_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([10240, 1280, 1, 1])];
            tensor<fp16, [10240]> up_blocks_0_attentions_1_transformer_blocks_9_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_9_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(560083904)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_48767_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_9_ff_net_0_proj_bias_to_fp16, dilations = var_48767_dilations_0, groups = var_48767_groups_0, pad = var_48767_pad_0, pad_type = var_48767_pad_type_0, strides = var_48767_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_9_ff_net_0_proj_weight_to_fp16_palettized, x = input_613_cast_fp16)[name = tensor<string, []>("op_48767_cast_fp16")];
            tensor<int32, [2]> var_48768_split_sizes_0 = const()[name = tensor<string, []>("op_48768_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_48768_axis_0 = const()[name = tensor<string, []>("op_48768_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_48768_cast_fp16_0, tensor<fp16, [2, 5120, 1, 1024]> var_48768_cast_fp16_1 = split(axis = var_48768_axis_0, split_sizes = var_48768_split_sizes_0, x = var_48767_cast_fp16)[name = tensor<string, []>("op_48768_cast_fp16")];
            tensor<string, []> var_48770_mode_0 = const()[name = tensor<string, []>("op_48770_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_48770_cast_fp16 = gelu(mode = var_48770_mode_0, x = var_48768_cast_fp16_1)[name = tensor<string, []>("op_48770_cast_fp16")];
            tensor<fp16, [2, 5120, 1, 1024]> input_615_cast_fp16 = mul(x = var_48768_cast_fp16_0, y = var_48770_cast_fp16)[name = tensor<string, []>("input_615_cast_fp16")];
            tensor<string, []> var_48778_pad_type_0 = const()[name = tensor<string, []>("op_48778_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_48778_strides_0 = const()[name = tensor<string, []>("op_48778_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_48778_pad_0 = const()[name = tensor<string, []>("op_48778_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_48778_dilations_0 = const()[name = tensor<string, []>("op_48778_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_48778_groups_0 = const()[name = tensor<string, []>("op_48778_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_9_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [4915200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(560104448))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(565019712))), name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_9_ff_net_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 5120, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_9_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_9_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(565019904)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_48778_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_9_ff_net_2_bias_to_fp16, dilations = var_48778_dilations_0, groups = var_48778_groups_0, pad = var_48778_pad_0, pad_type = var_48778_pad_type_0, strides = var_48778_strides_0, weight = up_blocks_0_attentions_1_transformer_blocks_9_ff_net_2_weight_to_fp16_palettized, x = input_615_cast_fp16)[name = tensor<string, []>("op_48778_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_421_cast_fp16 = add(x = var_48778_cast_fp16, y = inputs_323_cast_fp16)[name = tensor<string, []>("hidden_states_421_cast_fp16")];
            tensor<int32, [4]> var_48780 = const()[name = tensor<string, []>("op_48780"), val = tensor<int32, [4]>([2, 1280, 32, 32])];
            tensor<fp16, [2, 1280, 32, 32]> input_617_cast_fp16 = reshape(shape = var_48780, x = hidden_states_421_cast_fp16)[name = tensor<string, []>("input_617_cast_fp16")];
            tensor<string, []> hidden_states_423_pad_type_0 = const()[name = tensor<string, []>("hidden_states_423_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> hidden_states_423_strides_0 = const()[name = tensor<string, []>("hidden_states_423_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_423_pad_0 = const()[name = tensor<string, []>("hidden_states_423_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_423_dilations_0 = const()[name = tensor<string, []>("hidden_states_423_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_423_groups_0 = const()[name = tensor<string, []>("hidden_states_423_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_proj_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(565022528))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(566251392))), name = tensor<string, []>("up_blocks_0_attentions_1_proj_out_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_proj_out_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_proj_out_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(566251584)))];
            tensor<fp16, [2, 1280, 32, 32]> hidden_states_423_cast_fp16 = conv(bias = up_blocks_0_attentions_1_proj_out_bias_to_fp16, dilations = hidden_states_423_dilations_0, groups = hidden_states_423_groups_0, pad = hidden_states_423_pad_0, pad_type = hidden_states_423_pad_type_0, strides = hidden_states_423_strides_0, weight = up_blocks_0_attentions_1_proj_out_weight_to_fp16_palettized, x = input_617_cast_fp16)[name = tensor<string, []>("hidden_states_423_cast_fp16")];
            tensor<fp16, [2, 1280, 32, 32]> hidden_states_425_cast_fp16 = add(x = hidden_states_423_cast_fp16, y = hidden_states_357_cast_fp16)[name = tensor<string, []>("hidden_states_425_cast_fp16")];
            tensor<bool, []> input_619_interleave_0 = const()[name = tensor<string, []>("input_619_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 640, 32, 32]> cast_10 = cast(dtype = cast_10_dtype_0, x = input_115_cast_fp16)[name = tensor<string, []>("cast_4")];
            tensor<fp16, [2, 1920, 32, 32]> input_619_cast_fp16 = concat(axis = var_30355, interleave = input_619_interleave_0, values = (hidden_states_425_cast_fp16, cast_10))[name = tensor<string, []>("input_619_cast_fp16")];
            tensor<int32, [5]> reshape_108_shape_0 = const()[name = tensor<string, []>("reshape_108_shape_0"), val = tensor<int32, [5]>([2, 32, 60, 32, 32])];
            tensor<fp16, [2, 32, 60, 32, 32]> reshape_108_cast_fp16 = reshape(shape = reshape_108_shape_0, x = input_619_cast_fp16)[name = tensor<string, []>("reshape_108_cast_fp16")];
            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_fp16 = reduce_mean(axes = reduce_mean_81_axes_0, keep_dims = reduce_mean_81_keep_dims_0, x = reshape_108_cast_fp16)[name = tensor<string, []>("reduce_mean_81_cast_fp16")];
            tensor<fp16, [2, 32, 60, 32, 32]> sub_54_cast_fp16 = sub(x = reshape_108_cast_fp16, y = reduce_mean_81_cast_fp16)[name = tensor<string, []>("sub_54_cast_fp16")];
            tensor<fp16, [2, 32, 60, 32, 32]> square_27_cast_fp16 = square(x = sub_54_cast_fp16)[name = tensor<string, []>("square_27_cast_fp16")];
            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_fp16 = reduce_mean(axes = reduce_mean_83_axes_0, keep_dims = reduce_mean_83_keep_dims_0, x = square_27_cast_fp16)[name = tensor<string, []>("reduce_mean_83_cast_fp16")];
            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_fp16 = add(x = reduce_mean_83_cast_fp16, y = add_54_y_0_to_fp16)[name = tensor<string, []>("add_54_cast_fp16")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_27_cast_fp16 = sqrt(x = add_54_cast_fp16)[name = tensor<string, []>("sqrt_27_cast_fp16")];
            tensor<fp16, [2, 32, 60, 32, 32]> real_div_27_cast_fp16 = real_div(x = sub_54_cast_fp16, y = sqrt_27_cast_fp16)[name = tensor<string, []>("real_div_27_cast_fp16")];
            tensor<int32, [4]> reshape_109_shape_0 = const()[name = tensor<string, []>("reshape_109_shape_0"), val = tensor<int32, [4]>([2, 1920, 32, 32])];
            tensor<fp16, [2, 1920, 32, 32]> reshape_109_cast_fp16 = reshape(shape = reshape_109_shape_0, x = real_div_27_cast_fp16)[name = tensor<string, []>("reshape_109_cast_fp16")];
            tensor<fp16, [1920]> add_55_mean_0_to_fp16 = const()[name = tensor<string, []>("add_55_mean_0_to_fp16"), val = tensor<fp16, [1920]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(566254208)))];
            tensor<fp16, [1920]> add_55_variance_0_to_fp16 = const()[name = tensor<string, []>("add_55_variance_0_to_fp16"), val = tensor<fp16, [1920]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(566258112)))];
            tensor<fp16, [1920]> add_55_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_55_gamma_0_to_fp16"), val = tensor<fp16, [1920]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(566262016)))];
            tensor<fp16, [1920]> add_55_beta_0_to_fp16 = const()[name = tensor<string, []>("add_55_beta_0_to_fp16"), val = tensor<fp16, [1920]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(566265920)))];
            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, 1920, 32, 32]> add_55_cast_fp16 = batch_norm(beta = add_55_beta_0_to_fp16, epsilon = add_55_epsilon_0_to_fp16, gamma = add_55_gamma_0_to_fp16, mean = add_55_mean_0_to_fp16, variance = add_55_variance_0_to_fp16, x = reshape_109_cast_fp16)[name = tensor<string, []>("add_55_cast_fp16")];
            tensor<fp16, [2, 1920, 32, 32]> input_623_cast_fp16 = silu(x = add_55_cast_fp16)[name = tensor<string, []>("input_623_cast_fp16")];
            tensor<string, []> hidden_states_427_pad_type_0 = const()[name = tensor<string, []>("hidden_states_427_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_427_pad_0 = const()[name = tensor<string, []>("hidden_states_427_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> hidden_states_427_strides_0 = const()[name = tensor<string, []>("hidden_states_427_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> hidden_states_427_dilations_0 = const()[name = tensor<string, []>("hidden_states_427_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_427_groups_0 = const()[name = tensor<string, []>("hidden_states_427_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1920, 3, 3]> up_blocks_0_resnets_2_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [16588800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(566269824))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(582858688))), name = tensor<string, []>("up_blocks_0_resnets_2_conv1_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1920, 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, []>(582858880)))];
            tensor<fp16, [2, 1280, 32, 32]> hidden_states_427_cast_fp16 = conv(bias = up_blocks_0_resnets_2_conv1_bias_to_fp16, dilations = hidden_states_427_dilations_0, groups = hidden_states_427_groups_0, pad = hidden_states_427_pad_0, pad_type = hidden_states_427_pad_type_0, strides = hidden_states_427_strides_0, weight = up_blocks_0_resnets_2_conv1_weight_to_fp16_palettized, x = input_623_cast_fp16)[name = tensor<string, []>("hidden_states_427_cast_fp16")];
            tensor<string, []> temb_21_pad_type_0 = const()[name = tensor<string, []>("temb_21_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> temb_21_strides_0 = const()[name = tensor<string, []>("temb_21_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> temb_21_pad_0 = const()[name = tensor<string, []>("temb_21_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> temb_21_dilations_0 = const()[name = tensor<string, []>("temb_21_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> temb_21_groups_0 = const()[name = tensor<string, []>("temb_21_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_resnets_2_time_emb_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(582861504))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(584090368))), 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, []>(584090560)))];
            tensor<fp16, [2, 1280, 1, 1]> temb_21_cast_fp16 = conv(bias = up_blocks_0_resnets_2_time_emb_proj_bias_to_fp16, dilations = temb_21_dilations_0, groups = temb_21_groups_0, pad = temb_21_pad_0, pad_type = temb_21_pad_type_0, strides = temb_21_strides_0, weight = up_blocks_0_resnets_2_time_emb_proj_weight_to_fp16_palettized, x = cast_2)[name = tensor<string, []>("temb_21_cast_fp16")];
            tensor<fp16, [2, 1280, 32, 32]> input_627_cast_fp16 = add(x = hidden_states_427_cast_fp16, y = temb_21_cast_fp16)[name = tensor<string, []>("input_627_cast_fp16")];
            tensor<int32, [5]> reshape_112_shape_0 = const()[name = tensor<string, []>("reshape_112_shape_0"), val = tensor<int32, [5]>([2, 32, 40, 32, 32])];
            tensor<fp16, [2, 32, 40, 32, 32]> reshape_112_cast_fp16 = reshape(shape = reshape_112_shape_0, x = input_627_cast_fp16)[name = tensor<string, []>("reshape_112_cast_fp16")];
            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_fp16 = reduce_mean(axes = reduce_mean_84_axes_0, keep_dims = reduce_mean_84_keep_dims_0, x = reshape_112_cast_fp16)[name = tensor<string, []>("reduce_mean_84_cast_fp16")];
            tensor<fp16, [2, 32, 40, 32, 32]> sub_56_cast_fp16 = sub(x = reshape_112_cast_fp16, y = reduce_mean_84_cast_fp16)[name = tensor<string, []>("sub_56_cast_fp16")];
            tensor<fp16, [2, 32, 40, 32, 32]> square_28_cast_fp16 = square(x = sub_56_cast_fp16)[name = tensor<string, []>("square_28_cast_fp16")];
            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_fp16 = reduce_mean(axes = reduce_mean_86_axes_0, keep_dims = reduce_mean_86_keep_dims_0, x = square_28_cast_fp16)[name = tensor<string, []>("reduce_mean_86_cast_fp16")];
            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_fp16 = add(x = reduce_mean_86_cast_fp16, y = add_56_y_0_to_fp16)[name = tensor<string, []>("add_56_cast_fp16")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_28_cast_fp16 = sqrt(x = add_56_cast_fp16)[name = tensor<string, []>("sqrt_28_cast_fp16")];
            tensor<fp16, [2, 32, 40, 32, 32]> real_div_28_cast_fp16 = real_div(x = sub_56_cast_fp16, y = sqrt_28_cast_fp16)[name = tensor<string, []>("real_div_28_cast_fp16")];
            tensor<int32, [4]> reshape_113_shape_0 = const()[name = tensor<string, []>("reshape_113_shape_0"), val = tensor<int32, [4]>([2, 1280, 32, 32])];
            tensor<fp16, [2, 1280, 32, 32]> reshape_113_cast_fp16 = reshape(shape = reshape_113_shape_0, x = real_div_28_cast_fp16)[name = tensor<string, []>("reshape_113_cast_fp16")];
            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, []>(584093184)))];
            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, []>(584095808)))];
            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, 32, 32]> add_57_cast_fp16 = 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_23_mean_0_to_fp16, variance = add_23_variance_0_to_fp16, x = reshape_113_cast_fp16)[name = tensor<string, []>("add_57_cast_fp16")];
            tensor<fp16, [2, 1280, 32, 32]> input_631_cast_fp16 = silu(x = add_57_cast_fp16)[name = tensor<string, []>("input_631_cast_fp16")];
            tensor<string, []> hidden_states_429_pad_type_0 = const()[name = tensor<string, []>("hidden_states_429_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_429_pad_0 = const()[name = tensor<string, []>("hidden_states_429_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> hidden_states_429_strides_0 = const()[name = tensor<string, []>("hidden_states_429_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> hidden_states_429_dilations_0 = const()[name = tensor<string, []>("hidden_states_429_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_429_groups_0 = const()[name = tensor<string, []>("hidden_states_429_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 3, 3]> up_blocks_0_resnets_2_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [11059200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(584098432))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(595157696))), 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, []>(595157888)))];
            tensor<fp16, [2, 1280, 32, 32]> hidden_states_429_cast_fp16 = conv(bias = up_blocks_0_resnets_2_conv2_bias_to_fp16, dilations = hidden_states_429_dilations_0, groups = hidden_states_429_groups_0, pad = hidden_states_429_pad_0, pad_type = hidden_states_429_pad_type_0, strides = hidden_states_429_strides_0, weight = up_blocks_0_resnets_2_conv2_weight_to_fp16_palettized, x = input_631_cast_fp16)[name = tensor<string, []>("hidden_states_429_cast_fp16")];
            tensor<string, []> x_9_pad_type_0 = const()[name = tensor<string, []>("x_9_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> x_9_strides_0 = const()[name = tensor<string, []>("x_9_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> x_9_pad_0 = const()[name = tensor<string, []>("x_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> x_9_dilations_0 = const()[name = tensor<string, []>("x_9_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> x_9_groups_0 = const()[name = tensor<string, []>("x_9_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1920, 1, 1]> up_blocks_0_resnets_2_conv_shortcut_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1843200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(595160512))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(597003776))), name = tensor<string, []>("up_blocks_0_resnets_2_conv_shortcut_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1920, 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, []>(597003968)))];
            tensor<fp16, [2, 1280, 32, 32]> x_9_cast_fp16 = conv(bias = up_blocks_0_resnets_2_conv_shortcut_bias_to_fp16, dilations = x_9_dilations_0, groups = x_9_groups_0, pad = x_9_pad_0, pad_type = x_9_pad_type_0, strides = x_9_strides_0, weight = up_blocks_0_resnets_2_conv_shortcut_weight_to_fp16_palettized, x = input_619_cast_fp16)[name = tensor<string, []>("x_9_cast_fp16")];
            tensor<fp16, [2, 1280, 32, 32]> hidden_states_431_cast_fp16 = add(x = x_9_cast_fp16, y = hidden_states_429_cast_fp16)[name = tensor<string, []>("hidden_states_431_cast_fp16")];
            tensor<int32, [5]> reshape_116_shape_0 = const()[name = tensor<string, []>("reshape_116_shape_0"), val = tensor<int32, [5]>([2, 32, 40, 32, 32])];
            tensor<fp16, [2, 32, 40, 32, 32]> reshape_116_cast_fp16 = reshape(shape = reshape_116_shape_0, x = hidden_states_431_cast_fp16)[name = tensor<string, []>("reshape_116_cast_fp16")];
            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_fp16 = reduce_mean(axes = reduce_mean_87_axes_0, keep_dims = reduce_mean_87_keep_dims_0, x = reshape_116_cast_fp16)[name = tensor<string, []>("reduce_mean_87_cast_fp16")];
            tensor<fp16, [2, 32, 40, 32, 32]> sub_58_cast_fp16 = sub(x = reshape_116_cast_fp16, y = reduce_mean_87_cast_fp16)[name = tensor<string, []>("sub_58_cast_fp16")];
            tensor<fp16, [2, 32, 40, 32, 32]> square_29_cast_fp16 = square(x = sub_58_cast_fp16)[name = tensor<string, []>("square_29_cast_fp16")];
            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_fp16 = reduce_mean(axes = reduce_mean_89_axes_0, keep_dims = reduce_mean_89_keep_dims_0, x = square_29_cast_fp16)[name = tensor<string, []>("reduce_mean_89_cast_fp16")];
            tensor<fp16, []> add_58_y_0_to_fp16 = const()[name = tensor<string, []>("add_58_y_0_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_58_cast_fp16 = add(x = reduce_mean_89_cast_fp16, y = add_58_y_0_to_fp16)[name = tensor<string, []>("add_58_cast_fp16")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_29_cast_fp16 = sqrt(x = add_58_cast_fp16)[name = tensor<string, []>("sqrt_29_cast_fp16")];
            tensor<fp16, [2, 32, 40, 32, 32]> real_div_29_cast_fp16 = real_div(x = sub_58_cast_fp16, y = sqrt_29_cast_fp16)[name = tensor<string, []>("real_div_29_cast_fp16")];
            tensor<int32, [4]> reshape_117_shape_0 = const()[name = tensor<string, []>("reshape_117_shape_0"), val = tensor<int32, [4]>([2, 1280, 32, 32])];
            tensor<fp16, [2, 1280, 32, 32]> reshape_117_cast_fp16 = reshape(shape = reshape_117_shape_0, x = real_div_29_cast_fp16)[name = tensor<string, []>("reshape_117_cast_fp16")];
            tensor<fp16, [1280]> add_59_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_59_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(597006592)))];
            tensor<fp16, [1280]> add_59_beta_0_to_fp16 = const()[name = tensor<string, []>("add_59_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(597009216)))];
            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, 1280, 32, 32]> add_59_cast_fp16 = batch_norm(beta = add_59_beta_0_to_fp16, epsilon = add_59_epsilon_0_to_fp16, gamma = add_59_gamma_0_to_fp16, mean = add_23_mean_0_to_fp16, variance = add_23_variance_0_to_fp16, x = reshape_117_cast_fp16)[name = tensor<string, []>("add_59_cast_fp16")];
            tensor<string, []> hidden_states_433_pad_type_0 = const()[name = tensor<string, []>("hidden_states_433_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> hidden_states_433_strides_0 = const()[name = tensor<string, []>("hidden_states_433_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_433_pad_0 = const()[name = tensor<string, []>("hidden_states_433_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_433_dilations_0 = const()[name = tensor<string, []>("hidden_states_433_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_433_groups_0 = const()[name = tensor<string, []>("hidden_states_433_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_proj_in_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(597011840))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(598240704))), name = tensor<string, []>("up_blocks_0_attentions_2_proj_in_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_proj_in_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_proj_in_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(598240896)))];
            tensor<fp16, [2, 1280, 32, 32]> hidden_states_433_cast_fp16 = conv(bias = up_blocks_0_attentions_2_proj_in_bias_to_fp16, dilations = hidden_states_433_dilations_0, groups = hidden_states_433_groups_0, pad = hidden_states_433_pad_0, pad_type = hidden_states_433_pad_type_0, strides = hidden_states_433_strides_0, weight = up_blocks_0_attentions_2_proj_in_weight_to_fp16_palettized, x = add_59_cast_fp16)[name = tensor<string, []>("hidden_states_433_cast_fp16")];
            tensor<int32, [4]> var_48878 = const()[name = tensor<string, []>("op_48878"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_325_cast_fp16 = reshape(shape = var_48878, x = hidden_states_433_cast_fp16)[name = tensor<string, []>("inputs_325_cast_fp16")];
            tensor<int32, [1]> hidden_states_435_axes_0 = const()[name = tensor<string, []>("hidden_states_435_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_435_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_435_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(598243520)))];
            tensor<fp16, [1280]> hidden_states_435_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_435_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(598246144)))];
            tensor<fp16, []> var_48894_to_fp16 = const()[name = tensor<string, []>("op_48894_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_435_cast_fp16 = layer_norm(axes = hidden_states_435_axes_0, beta = hidden_states_435_beta_0_to_fp16, epsilon = var_48894_to_fp16, gamma = hidden_states_435_gamma_0_to_fp16, x = inputs_325_cast_fp16)[name = tensor<string, []>("hidden_states_435_cast_fp16")];
            tensor<string, []> q_217_pad_type_0 = const()[name = tensor<string, []>("q_217_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_217_strides_0 = const()[name = tensor<string, []>("q_217_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_217_pad_0 = const()[name = tensor<string, []>("q_217_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_217_dilations_0 = const()[name = tensor<string, []>("q_217_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_217_groups_0 = const()[name = tensor<string, []>("q_217_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(598248768))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(599477632))), name = tensor<string, []>("up_blocks_0_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, 1024]> q_217_cast_fp16 = conv(dilations = q_217_dilations_0, groups = q_217_groups_0, pad = q_217_pad_0, pad_type = q_217_pad_type_0, strides = q_217_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_435_cast_fp16)[name = tensor<string, []>("q_217_cast_fp16")];
            tensor<string, []> k_433_pad_type_0 = const()[name = tensor<string, []>("k_433_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_433_strides_0 = const()[name = tensor<string, []>("k_433_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_433_pad_0 = const()[name = tensor<string, []>("k_433_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_433_dilations_0 = const()[name = tensor<string, []>("k_433_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_433_groups_0 = const()[name = tensor<string, []>("k_433_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(599477824))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(600706688))), name = tensor<string, []>("up_blocks_0_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, 1024]> k_433_cast_fp16 = conv(dilations = k_433_dilations_0, groups = k_433_groups_0, pad = k_433_pad_0, pad_type = k_433_pad_type_0, strides = k_433_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_435_cast_fp16)[name = tensor<string, []>("k_433_cast_fp16")];
            tensor<string, []> v_217_pad_type_0 = const()[name = tensor<string, []>("v_217_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_217_strides_0 = const()[name = tensor<string, []>("v_217_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_217_pad_0 = const()[name = tensor<string, []>("v_217_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_217_dilations_0 = const()[name = tensor<string, []>("v_217_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_217_groups_0 = const()[name = tensor<string, []>("v_217_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(600706880))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(601935744))), name = tensor<string, []>("up_blocks_0_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, 1024]> v_217_cast_fp16 = conv(dilations = v_217_dilations_0, groups = v_217_groups_0, pad = v_217_pad_0, pad_type = v_217_pad_type_0, strides = v_217_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_435_cast_fp16)[name = tensor<string, []>("v_217_cast_fp16")];
            tensor<int32, [4]> var_48927_begin_0 = const()[name = tensor<string, []>("op_48927_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_48927_end_0 = const()[name = tensor<string, []>("op_48927_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_48927_end_mask_0 = const()[name = tensor<string, []>("op_48927_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48927_cast_fp16 = slice_by_index(begin = var_48927_begin_0, end = var_48927_end_0, end_mask = var_48927_end_mask_0, x = q_217_cast_fp16)[name = tensor<string, []>("op_48927_cast_fp16")];
            tensor<int32, [4]> var_48931_begin_0 = const()[name = tensor<string, []>("op_48931_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_48931_end_0 = const()[name = tensor<string, []>("op_48931_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_48931_end_mask_0 = const()[name = tensor<string, []>("op_48931_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48931_cast_fp16 = slice_by_index(begin = var_48931_begin_0, end = var_48931_end_0, end_mask = var_48931_end_mask_0, x = q_217_cast_fp16)[name = tensor<string, []>("op_48931_cast_fp16")];
            tensor<int32, [4]> var_48935_begin_0 = const()[name = tensor<string, []>("op_48935_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_48935_end_0 = const()[name = tensor<string, []>("op_48935_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_48935_end_mask_0 = const()[name = tensor<string, []>("op_48935_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48935_cast_fp16 = slice_by_index(begin = var_48935_begin_0, end = var_48935_end_0, end_mask = var_48935_end_mask_0, x = q_217_cast_fp16)[name = tensor<string, []>("op_48935_cast_fp16")];
            tensor<int32, [4]> var_48939_begin_0 = const()[name = tensor<string, []>("op_48939_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_48939_end_0 = const()[name = tensor<string, []>("op_48939_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_48939_end_mask_0 = const()[name = tensor<string, []>("op_48939_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48939_cast_fp16 = slice_by_index(begin = var_48939_begin_0, end = var_48939_end_0, end_mask = var_48939_end_mask_0, x = q_217_cast_fp16)[name = tensor<string, []>("op_48939_cast_fp16")];
            tensor<int32, [4]> var_48943_begin_0 = const()[name = tensor<string, []>("op_48943_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_48943_end_0 = const()[name = tensor<string, []>("op_48943_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_48943_end_mask_0 = const()[name = tensor<string, []>("op_48943_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48943_cast_fp16 = slice_by_index(begin = var_48943_begin_0, end = var_48943_end_0, end_mask = var_48943_end_mask_0, x = q_217_cast_fp16)[name = tensor<string, []>("op_48943_cast_fp16")];
            tensor<int32, [4]> var_48947_begin_0 = const()[name = tensor<string, []>("op_48947_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_48947_end_0 = const()[name = tensor<string, []>("op_48947_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_48947_end_mask_0 = const()[name = tensor<string, []>("op_48947_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48947_cast_fp16 = slice_by_index(begin = var_48947_begin_0, end = var_48947_end_0, end_mask = var_48947_end_mask_0, x = q_217_cast_fp16)[name = tensor<string, []>("op_48947_cast_fp16")];
            tensor<int32, [4]> var_48951_begin_0 = const()[name = tensor<string, []>("op_48951_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_48951_end_0 = const()[name = tensor<string, []>("op_48951_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_48951_end_mask_0 = const()[name = tensor<string, []>("op_48951_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48951_cast_fp16 = slice_by_index(begin = var_48951_begin_0, end = var_48951_end_0, end_mask = var_48951_end_mask_0, x = q_217_cast_fp16)[name = tensor<string, []>("op_48951_cast_fp16")];
            tensor<int32, [4]> var_48955_begin_0 = const()[name = tensor<string, []>("op_48955_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_48955_end_0 = const()[name = tensor<string, []>("op_48955_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_48955_end_mask_0 = const()[name = tensor<string, []>("op_48955_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48955_cast_fp16 = slice_by_index(begin = var_48955_begin_0, end = var_48955_end_0, end_mask = var_48955_end_mask_0, x = q_217_cast_fp16)[name = tensor<string, []>("op_48955_cast_fp16")];
            tensor<int32, [4]> var_48959_begin_0 = const()[name = tensor<string, []>("op_48959_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_48959_end_0 = const()[name = tensor<string, []>("op_48959_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_48959_end_mask_0 = const()[name = tensor<string, []>("op_48959_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48959_cast_fp16 = slice_by_index(begin = var_48959_begin_0, end = var_48959_end_0, end_mask = var_48959_end_mask_0, x = q_217_cast_fp16)[name = tensor<string, []>("op_48959_cast_fp16")];
            tensor<int32, [4]> var_48963_begin_0 = const()[name = tensor<string, []>("op_48963_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_48963_end_0 = const()[name = tensor<string, []>("op_48963_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_48963_end_mask_0 = const()[name = tensor<string, []>("op_48963_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48963_cast_fp16 = slice_by_index(begin = var_48963_begin_0, end = var_48963_end_0, end_mask = var_48963_end_mask_0, x = q_217_cast_fp16)[name = tensor<string, []>("op_48963_cast_fp16")];
            tensor<int32, [4]> var_48967_begin_0 = const()[name = tensor<string, []>("op_48967_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_48967_end_0 = const()[name = tensor<string, []>("op_48967_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_48967_end_mask_0 = const()[name = tensor<string, []>("op_48967_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48967_cast_fp16 = slice_by_index(begin = var_48967_begin_0, end = var_48967_end_0, end_mask = var_48967_end_mask_0, x = q_217_cast_fp16)[name = tensor<string, []>("op_48967_cast_fp16")];
            tensor<int32, [4]> var_48971_begin_0 = const()[name = tensor<string, []>("op_48971_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_48971_end_0 = const()[name = tensor<string, []>("op_48971_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_48971_end_mask_0 = const()[name = tensor<string, []>("op_48971_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48971_cast_fp16 = slice_by_index(begin = var_48971_begin_0, end = var_48971_end_0, end_mask = var_48971_end_mask_0, x = q_217_cast_fp16)[name = tensor<string, []>("op_48971_cast_fp16")];
            tensor<int32, [4]> var_48975_begin_0 = const()[name = tensor<string, []>("op_48975_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_48975_end_0 = const()[name = tensor<string, []>("op_48975_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_48975_end_mask_0 = const()[name = tensor<string, []>("op_48975_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48975_cast_fp16 = slice_by_index(begin = var_48975_begin_0, end = var_48975_end_0, end_mask = var_48975_end_mask_0, x = q_217_cast_fp16)[name = tensor<string, []>("op_48975_cast_fp16")];
            tensor<int32, [4]> var_48979_begin_0 = const()[name = tensor<string, []>("op_48979_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_48979_end_0 = const()[name = tensor<string, []>("op_48979_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_48979_end_mask_0 = const()[name = tensor<string, []>("op_48979_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48979_cast_fp16 = slice_by_index(begin = var_48979_begin_0, end = var_48979_end_0, end_mask = var_48979_end_mask_0, x = q_217_cast_fp16)[name = tensor<string, []>("op_48979_cast_fp16")];
            tensor<int32, [4]> var_48983_begin_0 = const()[name = tensor<string, []>("op_48983_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_48983_end_0 = const()[name = tensor<string, []>("op_48983_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_48983_end_mask_0 = const()[name = tensor<string, []>("op_48983_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48983_cast_fp16 = slice_by_index(begin = var_48983_begin_0, end = var_48983_end_0, end_mask = var_48983_end_mask_0, x = q_217_cast_fp16)[name = tensor<string, []>("op_48983_cast_fp16")];
            tensor<int32, [4]> var_48987_begin_0 = const()[name = tensor<string, []>("op_48987_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_48987_end_0 = const()[name = tensor<string, []>("op_48987_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_48987_end_mask_0 = const()[name = tensor<string, []>("op_48987_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48987_cast_fp16 = slice_by_index(begin = var_48987_begin_0, end = var_48987_end_0, end_mask = var_48987_end_mask_0, x = q_217_cast_fp16)[name = tensor<string, []>("op_48987_cast_fp16")];
            tensor<int32, [4]> var_48991_begin_0 = const()[name = tensor<string, []>("op_48991_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_48991_end_0 = const()[name = tensor<string, []>("op_48991_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_48991_end_mask_0 = const()[name = tensor<string, []>("op_48991_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48991_cast_fp16 = slice_by_index(begin = var_48991_begin_0, end = var_48991_end_0, end_mask = var_48991_end_mask_0, x = q_217_cast_fp16)[name = tensor<string, []>("op_48991_cast_fp16")];
            tensor<int32, [4]> var_48995_begin_0 = const()[name = tensor<string, []>("op_48995_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_48995_end_0 = const()[name = tensor<string, []>("op_48995_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_48995_end_mask_0 = const()[name = tensor<string, []>("op_48995_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48995_cast_fp16 = slice_by_index(begin = var_48995_begin_0, end = var_48995_end_0, end_mask = var_48995_end_mask_0, x = q_217_cast_fp16)[name = tensor<string, []>("op_48995_cast_fp16")];
            tensor<int32, [4]> var_48999_begin_0 = const()[name = tensor<string, []>("op_48999_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_48999_end_0 = const()[name = tensor<string, []>("op_48999_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_48999_end_mask_0 = const()[name = tensor<string, []>("op_48999_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48999_cast_fp16 = slice_by_index(begin = var_48999_begin_0, end = var_48999_end_0, end_mask = var_48999_end_mask_0, x = q_217_cast_fp16)[name = tensor<string, []>("op_48999_cast_fp16")];
            tensor<int32, [4]> var_49003_begin_0 = const()[name = tensor<string, []>("op_49003_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_49003_end_0 = const()[name = tensor<string, []>("op_49003_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_49003_end_mask_0 = const()[name = tensor<string, []>("op_49003_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49003_cast_fp16 = slice_by_index(begin = var_49003_begin_0, end = var_49003_end_0, end_mask = var_49003_end_mask_0, x = q_217_cast_fp16)[name = tensor<string, []>("op_49003_cast_fp16")];
            tensor<int32, [4]> k_435_perm_0 = const()[name = tensor<string, []>("k_435_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_49010_begin_0 = const()[name = tensor<string, []>("op_49010_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_49010_end_0 = const()[name = tensor<string, []>("op_49010_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_49010_end_mask_0 = const()[name = tensor<string, []>("op_49010_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> k_435_cast_fp16 = transpose(perm = k_435_perm_0, x = k_433_cast_fp16)[name = tensor<string, []>("transpose_31")];
            tensor<fp16, [2, 1024, 1, 64]> var_49010_cast_fp16 = slice_by_index(begin = var_49010_begin_0, end = var_49010_end_0, end_mask = var_49010_end_mask_0, x = k_435_cast_fp16)[name = tensor<string, []>("op_49010_cast_fp16")];
            tensor<int32, [4]> var_49014_begin_0 = const()[name = tensor<string, []>("op_49014_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_49014_end_0 = const()[name = tensor<string, []>("op_49014_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_49014_end_mask_0 = const()[name = tensor<string, []>("op_49014_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49014_cast_fp16 = slice_by_index(begin = var_49014_begin_0, end = var_49014_end_0, end_mask = var_49014_end_mask_0, x = k_435_cast_fp16)[name = tensor<string, []>("op_49014_cast_fp16")];
            tensor<int32, [4]> var_49018_begin_0 = const()[name = tensor<string, []>("op_49018_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_49018_end_0 = const()[name = tensor<string, []>("op_49018_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_49018_end_mask_0 = const()[name = tensor<string, []>("op_49018_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49018_cast_fp16 = slice_by_index(begin = var_49018_begin_0, end = var_49018_end_0, end_mask = var_49018_end_mask_0, x = k_435_cast_fp16)[name = tensor<string, []>("op_49018_cast_fp16")];
            tensor<int32, [4]> var_49022_begin_0 = const()[name = tensor<string, []>("op_49022_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_49022_end_0 = const()[name = tensor<string, []>("op_49022_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_49022_end_mask_0 = const()[name = tensor<string, []>("op_49022_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49022_cast_fp16 = slice_by_index(begin = var_49022_begin_0, end = var_49022_end_0, end_mask = var_49022_end_mask_0, x = k_435_cast_fp16)[name = tensor<string, []>("op_49022_cast_fp16")];
            tensor<int32, [4]> var_49026_begin_0 = const()[name = tensor<string, []>("op_49026_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_49026_end_0 = const()[name = tensor<string, []>("op_49026_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_49026_end_mask_0 = const()[name = tensor<string, []>("op_49026_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49026_cast_fp16 = slice_by_index(begin = var_49026_begin_0, end = var_49026_end_0, end_mask = var_49026_end_mask_0, x = k_435_cast_fp16)[name = tensor<string, []>("op_49026_cast_fp16")];
            tensor<int32, [4]> var_49030_begin_0 = const()[name = tensor<string, []>("op_49030_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_49030_end_0 = const()[name = tensor<string, []>("op_49030_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_49030_end_mask_0 = const()[name = tensor<string, []>("op_49030_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49030_cast_fp16 = slice_by_index(begin = var_49030_begin_0, end = var_49030_end_0, end_mask = var_49030_end_mask_0, x = k_435_cast_fp16)[name = tensor<string, []>("op_49030_cast_fp16")];
            tensor<int32, [4]> var_49034_begin_0 = const()[name = tensor<string, []>("op_49034_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_49034_end_0 = const()[name = tensor<string, []>("op_49034_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_49034_end_mask_0 = const()[name = tensor<string, []>("op_49034_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49034_cast_fp16 = slice_by_index(begin = var_49034_begin_0, end = var_49034_end_0, end_mask = var_49034_end_mask_0, x = k_435_cast_fp16)[name = tensor<string, []>("op_49034_cast_fp16")];
            tensor<int32, [4]> var_49038_begin_0 = const()[name = tensor<string, []>("op_49038_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_49038_end_0 = const()[name = tensor<string, []>("op_49038_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_49038_end_mask_0 = const()[name = tensor<string, []>("op_49038_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49038_cast_fp16 = slice_by_index(begin = var_49038_begin_0, end = var_49038_end_0, end_mask = var_49038_end_mask_0, x = k_435_cast_fp16)[name = tensor<string, []>("op_49038_cast_fp16")];
            tensor<int32, [4]> var_49042_begin_0 = const()[name = tensor<string, []>("op_49042_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_49042_end_0 = const()[name = tensor<string, []>("op_49042_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_49042_end_mask_0 = const()[name = tensor<string, []>("op_49042_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49042_cast_fp16 = slice_by_index(begin = var_49042_begin_0, end = var_49042_end_0, end_mask = var_49042_end_mask_0, x = k_435_cast_fp16)[name = tensor<string, []>("op_49042_cast_fp16")];
            tensor<int32, [4]> var_49046_begin_0 = const()[name = tensor<string, []>("op_49046_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_49046_end_0 = const()[name = tensor<string, []>("op_49046_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_49046_end_mask_0 = const()[name = tensor<string, []>("op_49046_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49046_cast_fp16 = slice_by_index(begin = var_49046_begin_0, end = var_49046_end_0, end_mask = var_49046_end_mask_0, x = k_435_cast_fp16)[name = tensor<string, []>("op_49046_cast_fp16")];
            tensor<int32, [4]> var_49050_begin_0 = const()[name = tensor<string, []>("op_49050_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_49050_end_0 = const()[name = tensor<string, []>("op_49050_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_49050_end_mask_0 = const()[name = tensor<string, []>("op_49050_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49050_cast_fp16 = slice_by_index(begin = var_49050_begin_0, end = var_49050_end_0, end_mask = var_49050_end_mask_0, x = k_435_cast_fp16)[name = tensor<string, []>("op_49050_cast_fp16")];
            tensor<int32, [4]> var_49054_begin_0 = const()[name = tensor<string, []>("op_49054_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_49054_end_0 = const()[name = tensor<string, []>("op_49054_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_49054_end_mask_0 = const()[name = tensor<string, []>("op_49054_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49054_cast_fp16 = slice_by_index(begin = var_49054_begin_0, end = var_49054_end_0, end_mask = var_49054_end_mask_0, x = k_435_cast_fp16)[name = tensor<string, []>("op_49054_cast_fp16")];
            tensor<int32, [4]> var_49058_begin_0 = const()[name = tensor<string, []>("op_49058_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_49058_end_0 = const()[name = tensor<string, []>("op_49058_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_49058_end_mask_0 = const()[name = tensor<string, []>("op_49058_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49058_cast_fp16 = slice_by_index(begin = var_49058_begin_0, end = var_49058_end_0, end_mask = var_49058_end_mask_0, x = k_435_cast_fp16)[name = tensor<string, []>("op_49058_cast_fp16")];
            tensor<int32, [4]> var_49062_begin_0 = const()[name = tensor<string, []>("op_49062_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_49062_end_0 = const()[name = tensor<string, []>("op_49062_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_49062_end_mask_0 = const()[name = tensor<string, []>("op_49062_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49062_cast_fp16 = slice_by_index(begin = var_49062_begin_0, end = var_49062_end_0, end_mask = var_49062_end_mask_0, x = k_435_cast_fp16)[name = tensor<string, []>("op_49062_cast_fp16")];
            tensor<int32, [4]> var_49066_begin_0 = const()[name = tensor<string, []>("op_49066_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_49066_end_0 = const()[name = tensor<string, []>("op_49066_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_49066_end_mask_0 = const()[name = tensor<string, []>("op_49066_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49066_cast_fp16 = slice_by_index(begin = var_49066_begin_0, end = var_49066_end_0, end_mask = var_49066_end_mask_0, x = k_435_cast_fp16)[name = tensor<string, []>("op_49066_cast_fp16")];
            tensor<int32, [4]> var_49070_begin_0 = const()[name = tensor<string, []>("op_49070_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_49070_end_0 = const()[name = tensor<string, []>("op_49070_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_49070_end_mask_0 = const()[name = tensor<string, []>("op_49070_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49070_cast_fp16 = slice_by_index(begin = var_49070_begin_0, end = var_49070_end_0, end_mask = var_49070_end_mask_0, x = k_435_cast_fp16)[name = tensor<string, []>("op_49070_cast_fp16")];
            tensor<int32, [4]> var_49074_begin_0 = const()[name = tensor<string, []>("op_49074_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_49074_end_0 = const()[name = tensor<string, []>("op_49074_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_49074_end_mask_0 = const()[name = tensor<string, []>("op_49074_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49074_cast_fp16 = slice_by_index(begin = var_49074_begin_0, end = var_49074_end_0, end_mask = var_49074_end_mask_0, x = k_435_cast_fp16)[name = tensor<string, []>("op_49074_cast_fp16")];
            tensor<int32, [4]> var_49078_begin_0 = const()[name = tensor<string, []>("op_49078_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_49078_end_0 = const()[name = tensor<string, []>("op_49078_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_49078_end_mask_0 = const()[name = tensor<string, []>("op_49078_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49078_cast_fp16 = slice_by_index(begin = var_49078_begin_0, end = var_49078_end_0, end_mask = var_49078_end_mask_0, x = k_435_cast_fp16)[name = tensor<string, []>("op_49078_cast_fp16")];
            tensor<int32, [4]> var_49082_begin_0 = const()[name = tensor<string, []>("op_49082_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_49082_end_0 = const()[name = tensor<string, []>("op_49082_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_49082_end_mask_0 = const()[name = tensor<string, []>("op_49082_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49082_cast_fp16 = slice_by_index(begin = var_49082_begin_0, end = var_49082_end_0, end_mask = var_49082_end_mask_0, x = k_435_cast_fp16)[name = tensor<string, []>("op_49082_cast_fp16")];
            tensor<int32, [4]> var_49086_begin_0 = const()[name = tensor<string, []>("op_49086_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_49086_end_0 = const()[name = tensor<string, []>("op_49086_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_49086_end_mask_0 = const()[name = tensor<string, []>("op_49086_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49086_cast_fp16 = slice_by_index(begin = var_49086_begin_0, end = var_49086_end_0, end_mask = var_49086_end_mask_0, x = k_435_cast_fp16)[name = tensor<string, []>("op_49086_cast_fp16")];
            tensor<int32, [4]> var_49088_begin_0 = const()[name = tensor<string, []>("op_49088_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_49088_end_0 = const()[name = tensor<string, []>("op_49088_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_49088_end_mask_0 = const()[name = tensor<string, []>("op_49088_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49088_cast_fp16 = slice_by_index(begin = var_49088_begin_0, end = var_49088_end_0, end_mask = var_49088_end_mask_0, x = v_217_cast_fp16)[name = tensor<string, []>("op_49088_cast_fp16")];
            tensor<int32, [4]> var_49092_begin_0 = const()[name = tensor<string, []>("op_49092_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_49092_end_0 = const()[name = tensor<string, []>("op_49092_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_49092_end_mask_0 = const()[name = tensor<string, []>("op_49092_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49092_cast_fp16 = slice_by_index(begin = var_49092_begin_0, end = var_49092_end_0, end_mask = var_49092_end_mask_0, x = v_217_cast_fp16)[name = tensor<string, []>("op_49092_cast_fp16")];
            tensor<int32, [4]> var_49096_begin_0 = const()[name = tensor<string, []>("op_49096_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_49096_end_0 = const()[name = tensor<string, []>("op_49096_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_49096_end_mask_0 = const()[name = tensor<string, []>("op_49096_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49096_cast_fp16 = slice_by_index(begin = var_49096_begin_0, end = var_49096_end_0, end_mask = var_49096_end_mask_0, x = v_217_cast_fp16)[name = tensor<string, []>("op_49096_cast_fp16")];
            tensor<int32, [4]> var_49100_begin_0 = const()[name = tensor<string, []>("op_49100_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_49100_end_0 = const()[name = tensor<string, []>("op_49100_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_49100_end_mask_0 = const()[name = tensor<string, []>("op_49100_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49100_cast_fp16 = slice_by_index(begin = var_49100_begin_0, end = var_49100_end_0, end_mask = var_49100_end_mask_0, x = v_217_cast_fp16)[name = tensor<string, []>("op_49100_cast_fp16")];
            tensor<int32, [4]> var_49104_begin_0 = const()[name = tensor<string, []>("op_49104_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_49104_end_0 = const()[name = tensor<string, []>("op_49104_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_49104_end_mask_0 = const()[name = tensor<string, []>("op_49104_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49104_cast_fp16 = slice_by_index(begin = var_49104_begin_0, end = var_49104_end_0, end_mask = var_49104_end_mask_0, x = v_217_cast_fp16)[name = tensor<string, []>("op_49104_cast_fp16")];
            tensor<int32, [4]> var_49108_begin_0 = const()[name = tensor<string, []>("op_49108_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_49108_end_0 = const()[name = tensor<string, []>("op_49108_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_49108_end_mask_0 = const()[name = tensor<string, []>("op_49108_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49108_cast_fp16 = slice_by_index(begin = var_49108_begin_0, end = var_49108_end_0, end_mask = var_49108_end_mask_0, x = v_217_cast_fp16)[name = tensor<string, []>("op_49108_cast_fp16")];
            tensor<int32, [4]> var_49112_begin_0 = const()[name = tensor<string, []>("op_49112_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_49112_end_0 = const()[name = tensor<string, []>("op_49112_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_49112_end_mask_0 = const()[name = tensor<string, []>("op_49112_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49112_cast_fp16 = slice_by_index(begin = var_49112_begin_0, end = var_49112_end_0, end_mask = var_49112_end_mask_0, x = v_217_cast_fp16)[name = tensor<string, []>("op_49112_cast_fp16")];
            tensor<int32, [4]> var_49116_begin_0 = const()[name = tensor<string, []>("op_49116_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_49116_end_0 = const()[name = tensor<string, []>("op_49116_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_49116_end_mask_0 = const()[name = tensor<string, []>("op_49116_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49116_cast_fp16 = slice_by_index(begin = var_49116_begin_0, end = var_49116_end_0, end_mask = var_49116_end_mask_0, x = v_217_cast_fp16)[name = tensor<string, []>("op_49116_cast_fp16")];
            tensor<int32, [4]> var_49120_begin_0 = const()[name = tensor<string, []>("op_49120_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_49120_end_0 = const()[name = tensor<string, []>("op_49120_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_49120_end_mask_0 = const()[name = tensor<string, []>("op_49120_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49120_cast_fp16 = slice_by_index(begin = var_49120_begin_0, end = var_49120_end_0, end_mask = var_49120_end_mask_0, x = v_217_cast_fp16)[name = tensor<string, []>("op_49120_cast_fp16")];
            tensor<int32, [4]> var_49124_begin_0 = const()[name = tensor<string, []>("op_49124_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_49124_end_0 = const()[name = tensor<string, []>("op_49124_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_49124_end_mask_0 = const()[name = tensor<string, []>("op_49124_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49124_cast_fp16 = slice_by_index(begin = var_49124_begin_0, end = var_49124_end_0, end_mask = var_49124_end_mask_0, x = v_217_cast_fp16)[name = tensor<string, []>("op_49124_cast_fp16")];
            tensor<int32, [4]> var_49128_begin_0 = const()[name = tensor<string, []>("op_49128_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_49128_end_0 = const()[name = tensor<string, []>("op_49128_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_49128_end_mask_0 = const()[name = tensor<string, []>("op_49128_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49128_cast_fp16 = slice_by_index(begin = var_49128_begin_0, end = var_49128_end_0, end_mask = var_49128_end_mask_0, x = v_217_cast_fp16)[name = tensor<string, []>("op_49128_cast_fp16")];
            tensor<int32, [4]> var_49132_begin_0 = const()[name = tensor<string, []>("op_49132_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_49132_end_0 = const()[name = tensor<string, []>("op_49132_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_49132_end_mask_0 = const()[name = tensor<string, []>("op_49132_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49132_cast_fp16 = slice_by_index(begin = var_49132_begin_0, end = var_49132_end_0, end_mask = var_49132_end_mask_0, x = v_217_cast_fp16)[name = tensor<string, []>("op_49132_cast_fp16")];
            tensor<int32, [4]> var_49136_begin_0 = const()[name = tensor<string, []>("op_49136_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_49136_end_0 = const()[name = tensor<string, []>("op_49136_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_49136_end_mask_0 = const()[name = tensor<string, []>("op_49136_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49136_cast_fp16 = slice_by_index(begin = var_49136_begin_0, end = var_49136_end_0, end_mask = var_49136_end_mask_0, x = v_217_cast_fp16)[name = tensor<string, []>("op_49136_cast_fp16")];
            tensor<int32, [4]> var_49140_begin_0 = const()[name = tensor<string, []>("op_49140_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_49140_end_0 = const()[name = tensor<string, []>("op_49140_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_49140_end_mask_0 = const()[name = tensor<string, []>("op_49140_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49140_cast_fp16 = slice_by_index(begin = var_49140_begin_0, end = var_49140_end_0, end_mask = var_49140_end_mask_0, x = v_217_cast_fp16)[name = tensor<string, []>("op_49140_cast_fp16")];
            tensor<int32, [4]> var_49144_begin_0 = const()[name = tensor<string, []>("op_49144_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_49144_end_0 = const()[name = tensor<string, []>("op_49144_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_49144_end_mask_0 = const()[name = tensor<string, []>("op_49144_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49144_cast_fp16 = slice_by_index(begin = var_49144_begin_0, end = var_49144_end_0, end_mask = var_49144_end_mask_0, x = v_217_cast_fp16)[name = tensor<string, []>("op_49144_cast_fp16")];
            tensor<int32, [4]> var_49148_begin_0 = const()[name = tensor<string, []>("op_49148_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_49148_end_0 = const()[name = tensor<string, []>("op_49148_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_49148_end_mask_0 = const()[name = tensor<string, []>("op_49148_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49148_cast_fp16 = slice_by_index(begin = var_49148_begin_0, end = var_49148_end_0, end_mask = var_49148_end_mask_0, x = v_217_cast_fp16)[name = tensor<string, []>("op_49148_cast_fp16")];
            tensor<int32, [4]> var_49152_begin_0 = const()[name = tensor<string, []>("op_49152_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_49152_end_0 = const()[name = tensor<string, []>("op_49152_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_49152_end_mask_0 = const()[name = tensor<string, []>("op_49152_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49152_cast_fp16 = slice_by_index(begin = var_49152_begin_0, end = var_49152_end_0, end_mask = var_49152_end_mask_0, x = v_217_cast_fp16)[name = tensor<string, []>("op_49152_cast_fp16")];
            tensor<int32, [4]> var_49156_begin_0 = const()[name = tensor<string, []>("op_49156_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_49156_end_0 = const()[name = tensor<string, []>("op_49156_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_49156_end_mask_0 = const()[name = tensor<string, []>("op_49156_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49156_cast_fp16 = slice_by_index(begin = var_49156_begin_0, end = var_49156_end_0, end_mask = var_49156_end_mask_0, x = v_217_cast_fp16)[name = tensor<string, []>("op_49156_cast_fp16")];
            tensor<int32, [4]> var_49160_begin_0 = const()[name = tensor<string, []>("op_49160_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_49160_end_0 = const()[name = tensor<string, []>("op_49160_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_49160_end_mask_0 = const()[name = tensor<string, []>("op_49160_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49160_cast_fp16 = slice_by_index(begin = var_49160_begin_0, end = var_49160_end_0, end_mask = var_49160_end_mask_0, x = v_217_cast_fp16)[name = tensor<string, []>("op_49160_cast_fp16")];
            tensor<int32, [4]> var_49164_begin_0 = const()[name = tensor<string, []>("op_49164_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_49164_end_0 = const()[name = tensor<string, []>("op_49164_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_49164_end_mask_0 = const()[name = tensor<string, []>("op_49164_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49164_cast_fp16 = slice_by_index(begin = var_49164_begin_0, end = var_49164_end_0, end_mask = var_49164_end_mask_0, x = v_217_cast_fp16)[name = tensor<string, []>("op_49164_cast_fp16")];
            tensor<string, []> var_49168_equation_0 = const()[name = tensor<string, []>("op_49168_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49168_cast_fp16 = einsum(equation = var_49168_equation_0, values = (var_49010_cast_fp16, var_48927_cast_fp16))[name = tensor<string, []>("op_49168_cast_fp16")];
            tensor<fp16, []> var_49169_to_fp16 = const()[name = tensor<string, []>("op_49169_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4161_cast_fp16 = mul(x = var_49168_cast_fp16, y = var_49169_to_fp16)[name = tensor<string, []>("aw_4161_cast_fp16")];
            tensor<string, []> var_49172_equation_0 = const()[name = tensor<string, []>("op_49172_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49172_cast_fp16 = einsum(equation = var_49172_equation_0, values = (var_49014_cast_fp16, var_48931_cast_fp16))[name = tensor<string, []>("op_49172_cast_fp16")];
            tensor<fp16, []> var_49173_to_fp16 = const()[name = tensor<string, []>("op_49173_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4163_cast_fp16 = mul(x = var_49172_cast_fp16, y = var_49173_to_fp16)[name = tensor<string, []>("aw_4163_cast_fp16")];
            tensor<string, []> var_49176_equation_0 = const()[name = tensor<string, []>("op_49176_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49176_cast_fp16 = einsum(equation = var_49176_equation_0, values = (var_49018_cast_fp16, var_48935_cast_fp16))[name = tensor<string, []>("op_49176_cast_fp16")];
            tensor<fp16, []> var_49177_to_fp16 = const()[name = tensor<string, []>("op_49177_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4165_cast_fp16 = mul(x = var_49176_cast_fp16, y = var_49177_to_fp16)[name = tensor<string, []>("aw_4165_cast_fp16")];
            tensor<string, []> var_49180_equation_0 = const()[name = tensor<string, []>("op_49180_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49180_cast_fp16 = einsum(equation = var_49180_equation_0, values = (var_49022_cast_fp16, var_48939_cast_fp16))[name = tensor<string, []>("op_49180_cast_fp16")];
            tensor<fp16, []> var_49181_to_fp16 = const()[name = tensor<string, []>("op_49181_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4167_cast_fp16 = mul(x = var_49180_cast_fp16, y = var_49181_to_fp16)[name = tensor<string, []>("aw_4167_cast_fp16")];
            tensor<string, []> var_49184_equation_0 = const()[name = tensor<string, []>("op_49184_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49184_cast_fp16 = einsum(equation = var_49184_equation_0, values = (var_49026_cast_fp16, var_48943_cast_fp16))[name = tensor<string, []>("op_49184_cast_fp16")];
            tensor<fp16, []> var_49185_to_fp16 = const()[name = tensor<string, []>("op_49185_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4169_cast_fp16 = mul(x = var_49184_cast_fp16, y = var_49185_to_fp16)[name = tensor<string, []>("aw_4169_cast_fp16")];
            tensor<string, []> var_49188_equation_0 = const()[name = tensor<string, []>("op_49188_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49188_cast_fp16 = einsum(equation = var_49188_equation_0, values = (var_49030_cast_fp16, var_48947_cast_fp16))[name = tensor<string, []>("op_49188_cast_fp16")];
            tensor<fp16, []> var_49189_to_fp16 = const()[name = tensor<string, []>("op_49189_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4171_cast_fp16 = mul(x = var_49188_cast_fp16, y = var_49189_to_fp16)[name = tensor<string, []>("aw_4171_cast_fp16")];
            tensor<string, []> var_49192_equation_0 = const()[name = tensor<string, []>("op_49192_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49192_cast_fp16 = einsum(equation = var_49192_equation_0, values = (var_49034_cast_fp16, var_48951_cast_fp16))[name = tensor<string, []>("op_49192_cast_fp16")];
            tensor<fp16, []> var_49193_to_fp16 = const()[name = tensor<string, []>("op_49193_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4173_cast_fp16 = mul(x = var_49192_cast_fp16, y = var_49193_to_fp16)[name = tensor<string, []>("aw_4173_cast_fp16")];
            tensor<string, []> var_49196_equation_0 = const()[name = tensor<string, []>("op_49196_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49196_cast_fp16 = einsum(equation = var_49196_equation_0, values = (var_49038_cast_fp16, var_48955_cast_fp16))[name = tensor<string, []>("op_49196_cast_fp16")];
            tensor<fp16, []> var_49197_to_fp16 = const()[name = tensor<string, []>("op_49197_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4175_cast_fp16 = mul(x = var_49196_cast_fp16, y = var_49197_to_fp16)[name = tensor<string, []>("aw_4175_cast_fp16")];
            tensor<string, []> var_49200_equation_0 = const()[name = tensor<string, []>("op_49200_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49200_cast_fp16 = einsum(equation = var_49200_equation_0, values = (var_49042_cast_fp16, var_48959_cast_fp16))[name = tensor<string, []>("op_49200_cast_fp16")];
            tensor<fp16, []> var_49201_to_fp16 = const()[name = tensor<string, []>("op_49201_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4177_cast_fp16 = mul(x = var_49200_cast_fp16, y = var_49201_to_fp16)[name = tensor<string, []>("aw_4177_cast_fp16")];
            tensor<string, []> var_49204_equation_0 = const()[name = tensor<string, []>("op_49204_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49204_cast_fp16 = einsum(equation = var_49204_equation_0, values = (var_49046_cast_fp16, var_48963_cast_fp16))[name = tensor<string, []>("op_49204_cast_fp16")];
            tensor<fp16, []> var_49205_to_fp16 = const()[name = tensor<string, []>("op_49205_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4179_cast_fp16 = mul(x = var_49204_cast_fp16, y = var_49205_to_fp16)[name = tensor<string, []>("aw_4179_cast_fp16")];
            tensor<string, []> var_49208_equation_0 = const()[name = tensor<string, []>("op_49208_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49208_cast_fp16 = einsum(equation = var_49208_equation_0, values = (var_49050_cast_fp16, var_48967_cast_fp16))[name = tensor<string, []>("op_49208_cast_fp16")];
            tensor<fp16, []> var_49209_to_fp16 = const()[name = tensor<string, []>("op_49209_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4181_cast_fp16 = mul(x = var_49208_cast_fp16, y = var_49209_to_fp16)[name = tensor<string, []>("aw_4181_cast_fp16")];
            tensor<string, []> var_49212_equation_0 = const()[name = tensor<string, []>("op_49212_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49212_cast_fp16 = einsum(equation = var_49212_equation_0, values = (var_49054_cast_fp16, var_48971_cast_fp16))[name = tensor<string, []>("op_49212_cast_fp16")];
            tensor<fp16, []> var_49213_to_fp16 = const()[name = tensor<string, []>("op_49213_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4183_cast_fp16 = mul(x = var_49212_cast_fp16, y = var_49213_to_fp16)[name = tensor<string, []>("aw_4183_cast_fp16")];
            tensor<string, []> var_49216_equation_0 = const()[name = tensor<string, []>("op_49216_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49216_cast_fp16 = einsum(equation = var_49216_equation_0, values = (var_49058_cast_fp16, var_48975_cast_fp16))[name = tensor<string, []>("op_49216_cast_fp16")];
            tensor<fp16, []> var_49217_to_fp16 = const()[name = tensor<string, []>("op_49217_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4185_cast_fp16 = mul(x = var_49216_cast_fp16, y = var_49217_to_fp16)[name = tensor<string, []>("aw_4185_cast_fp16")];
            tensor<string, []> var_49220_equation_0 = const()[name = tensor<string, []>("op_49220_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49220_cast_fp16 = einsum(equation = var_49220_equation_0, values = (var_49062_cast_fp16, var_48979_cast_fp16))[name = tensor<string, []>("op_49220_cast_fp16")];
            tensor<fp16, []> var_49221_to_fp16 = const()[name = tensor<string, []>("op_49221_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4187_cast_fp16 = mul(x = var_49220_cast_fp16, y = var_49221_to_fp16)[name = tensor<string, []>("aw_4187_cast_fp16")];
            tensor<string, []> var_49224_equation_0 = const()[name = tensor<string, []>("op_49224_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49224_cast_fp16 = einsum(equation = var_49224_equation_0, values = (var_49066_cast_fp16, var_48983_cast_fp16))[name = tensor<string, []>("op_49224_cast_fp16")];
            tensor<fp16, []> var_49225_to_fp16 = const()[name = tensor<string, []>("op_49225_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4189_cast_fp16 = mul(x = var_49224_cast_fp16, y = var_49225_to_fp16)[name = tensor<string, []>("aw_4189_cast_fp16")];
            tensor<string, []> var_49228_equation_0 = const()[name = tensor<string, []>("op_49228_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49228_cast_fp16 = einsum(equation = var_49228_equation_0, values = (var_49070_cast_fp16, var_48987_cast_fp16))[name = tensor<string, []>("op_49228_cast_fp16")];
            tensor<fp16, []> var_49229_to_fp16 = const()[name = tensor<string, []>("op_49229_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4191_cast_fp16 = mul(x = var_49228_cast_fp16, y = var_49229_to_fp16)[name = tensor<string, []>("aw_4191_cast_fp16")];
            tensor<string, []> var_49232_equation_0 = const()[name = tensor<string, []>("op_49232_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49232_cast_fp16 = einsum(equation = var_49232_equation_0, values = (var_49074_cast_fp16, var_48991_cast_fp16))[name = tensor<string, []>("op_49232_cast_fp16")];
            tensor<fp16, []> var_49233_to_fp16 = const()[name = tensor<string, []>("op_49233_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4193_cast_fp16 = mul(x = var_49232_cast_fp16, y = var_49233_to_fp16)[name = tensor<string, []>("aw_4193_cast_fp16")];
            tensor<string, []> var_49236_equation_0 = const()[name = tensor<string, []>("op_49236_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49236_cast_fp16 = einsum(equation = var_49236_equation_0, values = (var_49078_cast_fp16, var_48995_cast_fp16))[name = tensor<string, []>("op_49236_cast_fp16")];
            tensor<fp16, []> var_49237_to_fp16 = const()[name = tensor<string, []>("op_49237_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4195_cast_fp16 = mul(x = var_49236_cast_fp16, y = var_49237_to_fp16)[name = tensor<string, []>("aw_4195_cast_fp16")];
            tensor<string, []> var_49240_equation_0 = const()[name = tensor<string, []>("op_49240_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49240_cast_fp16 = einsum(equation = var_49240_equation_0, values = (var_49082_cast_fp16, var_48999_cast_fp16))[name = tensor<string, []>("op_49240_cast_fp16")];
            tensor<fp16, []> var_49241_to_fp16 = const()[name = tensor<string, []>("op_49241_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4197_cast_fp16 = mul(x = var_49240_cast_fp16, y = var_49241_to_fp16)[name = tensor<string, []>("aw_4197_cast_fp16")];
            tensor<string, []> var_49244_equation_0 = const()[name = tensor<string, []>("op_49244_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49244_cast_fp16 = einsum(equation = var_49244_equation_0, values = (var_49086_cast_fp16, var_49003_cast_fp16))[name = tensor<string, []>("op_49244_cast_fp16")];
            tensor<fp16, []> var_49245_to_fp16 = const()[name = tensor<string, []>("op_49245_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4199_cast_fp16 = mul(x = var_49244_cast_fp16, y = var_49245_to_fp16)[name = tensor<string, []>("aw_4199_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49247_cast_fp16 = softmax(axis = var_30355, x = aw_4161_cast_fp16)[name = tensor<string, []>("op_49247_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49248_cast_fp16 = softmax(axis = var_30355, x = aw_4163_cast_fp16)[name = tensor<string, []>("op_49248_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49249_cast_fp16 = softmax(axis = var_30355, x = aw_4165_cast_fp16)[name = tensor<string, []>("op_49249_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49250_cast_fp16 = softmax(axis = var_30355, x = aw_4167_cast_fp16)[name = tensor<string, []>("op_49250_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49251_cast_fp16 = softmax(axis = var_30355, x = aw_4169_cast_fp16)[name = tensor<string, []>("op_49251_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49252_cast_fp16 = softmax(axis = var_30355, x = aw_4171_cast_fp16)[name = tensor<string, []>("op_49252_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49253_cast_fp16 = softmax(axis = var_30355, x = aw_4173_cast_fp16)[name = tensor<string, []>("op_49253_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49254_cast_fp16 = softmax(axis = var_30355, x = aw_4175_cast_fp16)[name = tensor<string, []>("op_49254_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49255_cast_fp16 = softmax(axis = var_30355, x = aw_4177_cast_fp16)[name = tensor<string, []>("op_49255_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49256_cast_fp16 = softmax(axis = var_30355, x = aw_4179_cast_fp16)[name = tensor<string, []>("op_49256_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49257_cast_fp16 = softmax(axis = var_30355, x = aw_4181_cast_fp16)[name = tensor<string, []>("op_49257_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49258_cast_fp16 = softmax(axis = var_30355, x = aw_4183_cast_fp16)[name = tensor<string, []>("op_49258_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49259_cast_fp16 = softmax(axis = var_30355, x = aw_4185_cast_fp16)[name = tensor<string, []>("op_49259_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49260_cast_fp16 = softmax(axis = var_30355, x = aw_4187_cast_fp16)[name = tensor<string, []>("op_49260_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49261_cast_fp16 = softmax(axis = var_30355, x = aw_4189_cast_fp16)[name = tensor<string, []>("op_49261_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49262_cast_fp16 = softmax(axis = var_30355, x = aw_4191_cast_fp16)[name = tensor<string, []>("op_49262_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49263_cast_fp16 = softmax(axis = var_30355, x = aw_4193_cast_fp16)[name = tensor<string, []>("op_49263_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49264_cast_fp16 = softmax(axis = var_30355, x = aw_4195_cast_fp16)[name = tensor<string, []>("op_49264_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49265_cast_fp16 = softmax(axis = var_30355, x = aw_4197_cast_fp16)[name = tensor<string, []>("op_49265_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49266_cast_fp16 = softmax(axis = var_30355, x = aw_4199_cast_fp16)[name = tensor<string, []>("op_49266_cast_fp16")];
            tensor<string, []> var_49268_equation_0 = const()[name = tensor<string, []>("op_49268_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49268_cast_fp16 = einsum(equation = var_49268_equation_0, values = (var_49088_cast_fp16, var_49247_cast_fp16))[name = tensor<string, []>("op_49268_cast_fp16")];
            tensor<string, []> var_49270_equation_0 = const()[name = tensor<string, []>("op_49270_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49270_cast_fp16 = einsum(equation = var_49270_equation_0, values = (var_49092_cast_fp16, var_49248_cast_fp16))[name = tensor<string, []>("op_49270_cast_fp16")];
            tensor<string, []> var_49272_equation_0 = const()[name = tensor<string, []>("op_49272_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49272_cast_fp16 = einsum(equation = var_49272_equation_0, values = (var_49096_cast_fp16, var_49249_cast_fp16))[name = tensor<string, []>("op_49272_cast_fp16")];
            tensor<string, []> var_49274_equation_0 = const()[name = tensor<string, []>("op_49274_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49274_cast_fp16 = einsum(equation = var_49274_equation_0, values = (var_49100_cast_fp16, var_49250_cast_fp16))[name = tensor<string, []>("op_49274_cast_fp16")];
            tensor<string, []> var_49276_equation_0 = const()[name = tensor<string, []>("op_49276_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49276_cast_fp16 = einsum(equation = var_49276_equation_0, values = (var_49104_cast_fp16, var_49251_cast_fp16))[name = tensor<string, []>("op_49276_cast_fp16")];
            tensor<string, []> var_49278_equation_0 = const()[name = tensor<string, []>("op_49278_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49278_cast_fp16 = einsum(equation = var_49278_equation_0, values = (var_49108_cast_fp16, var_49252_cast_fp16))[name = tensor<string, []>("op_49278_cast_fp16")];
            tensor<string, []> var_49280_equation_0 = const()[name = tensor<string, []>("op_49280_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49280_cast_fp16 = einsum(equation = var_49280_equation_0, values = (var_49112_cast_fp16, var_49253_cast_fp16))[name = tensor<string, []>("op_49280_cast_fp16")];
            tensor<string, []> var_49282_equation_0 = const()[name = tensor<string, []>("op_49282_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49282_cast_fp16 = einsum(equation = var_49282_equation_0, values = (var_49116_cast_fp16, var_49254_cast_fp16))[name = tensor<string, []>("op_49282_cast_fp16")];
            tensor<string, []> var_49284_equation_0 = const()[name = tensor<string, []>("op_49284_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49284_cast_fp16 = einsum(equation = var_49284_equation_0, values = (var_49120_cast_fp16, var_49255_cast_fp16))[name = tensor<string, []>("op_49284_cast_fp16")];
            tensor<string, []> var_49286_equation_0 = const()[name = tensor<string, []>("op_49286_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49286_cast_fp16 = einsum(equation = var_49286_equation_0, values = (var_49124_cast_fp16, var_49256_cast_fp16))[name = tensor<string, []>("op_49286_cast_fp16")];
            tensor<string, []> var_49288_equation_0 = const()[name = tensor<string, []>("op_49288_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49288_cast_fp16 = einsum(equation = var_49288_equation_0, values = (var_49128_cast_fp16, var_49257_cast_fp16))[name = tensor<string, []>("op_49288_cast_fp16")];
            tensor<string, []> var_49290_equation_0 = const()[name = tensor<string, []>("op_49290_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49290_cast_fp16 = einsum(equation = var_49290_equation_0, values = (var_49132_cast_fp16, var_49258_cast_fp16))[name = tensor<string, []>("op_49290_cast_fp16")];
            tensor<string, []> var_49292_equation_0 = const()[name = tensor<string, []>("op_49292_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49292_cast_fp16 = einsum(equation = var_49292_equation_0, values = (var_49136_cast_fp16, var_49259_cast_fp16))[name = tensor<string, []>("op_49292_cast_fp16")];
            tensor<string, []> var_49294_equation_0 = const()[name = tensor<string, []>("op_49294_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49294_cast_fp16 = einsum(equation = var_49294_equation_0, values = (var_49140_cast_fp16, var_49260_cast_fp16))[name = tensor<string, []>("op_49294_cast_fp16")];
            tensor<string, []> var_49296_equation_0 = const()[name = tensor<string, []>("op_49296_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49296_cast_fp16 = einsum(equation = var_49296_equation_0, values = (var_49144_cast_fp16, var_49261_cast_fp16))[name = tensor<string, []>("op_49296_cast_fp16")];
            tensor<string, []> var_49298_equation_0 = const()[name = tensor<string, []>("op_49298_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49298_cast_fp16 = einsum(equation = var_49298_equation_0, values = (var_49148_cast_fp16, var_49262_cast_fp16))[name = tensor<string, []>("op_49298_cast_fp16")];
            tensor<string, []> var_49300_equation_0 = const()[name = tensor<string, []>("op_49300_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49300_cast_fp16 = einsum(equation = var_49300_equation_0, values = (var_49152_cast_fp16, var_49263_cast_fp16))[name = tensor<string, []>("op_49300_cast_fp16")];
            tensor<string, []> var_49302_equation_0 = const()[name = tensor<string, []>("op_49302_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49302_cast_fp16 = einsum(equation = var_49302_equation_0, values = (var_49156_cast_fp16, var_49264_cast_fp16))[name = tensor<string, []>("op_49302_cast_fp16")];
            tensor<string, []> var_49304_equation_0 = const()[name = tensor<string, []>("op_49304_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49304_cast_fp16 = einsum(equation = var_49304_equation_0, values = (var_49160_cast_fp16, var_49265_cast_fp16))[name = tensor<string, []>("op_49304_cast_fp16")];
            tensor<string, []> var_49306_equation_0 = const()[name = tensor<string, []>("op_49306_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49306_cast_fp16 = einsum(equation = var_49306_equation_0, values = (var_49164_cast_fp16, var_49266_cast_fp16))[name = tensor<string, []>("op_49306_cast_fp16")];
            tensor<bool, []> input_635_interleave_0 = const()[name = tensor<string, []>("input_635_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_635_cast_fp16 = concat(axis = var_30355, interleave = input_635_interleave_0, values = (var_49268_cast_fp16, var_49270_cast_fp16, var_49272_cast_fp16, var_49274_cast_fp16, var_49276_cast_fp16, var_49278_cast_fp16, var_49280_cast_fp16, var_49282_cast_fp16, var_49284_cast_fp16, var_49286_cast_fp16, var_49288_cast_fp16, var_49290_cast_fp16, var_49292_cast_fp16, var_49294_cast_fp16, var_49296_cast_fp16, var_49298_cast_fp16, var_49300_cast_fp16, var_49302_cast_fp16, var_49304_cast_fp16, var_49306_cast_fp16))[name = tensor<string, []>("input_635_cast_fp16")];
            tensor<string, []> var_49316_pad_type_0 = const()[name = tensor<string, []>("op_49316_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_49316_strides_0 = const()[name = tensor<string, []>("op_49316_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_49316_pad_0 = const()[name = tensor<string, []>("op_49316_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_49316_dilations_0 = const()[name = tensor<string, []>("op_49316_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_49316_groups_0 = const()[name = tensor<string, []>("op_49316_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(601935936))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(603164800))), name = tensor<string, []>("up_blocks_0_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_0_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_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, []>(603164992)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_49316_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_49316_dilations_0, groups = var_49316_groups_0, pad = var_49316_pad_0, pad_type = var_49316_pad_type_0, strides = var_49316_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized, x = input_635_cast_fp16)[name = tensor<string, []>("op_49316_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_327_cast_fp16 = add(x = var_49316_cast_fp16, y = inputs_325_cast_fp16)[name = tensor<string, []>("inputs_327_cast_fp16")];
            tensor<int32, [1]> hidden_states_437_axes_0 = const()[name = tensor<string, []>("hidden_states_437_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_437_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_437_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(603167616)))];
            tensor<fp16, [1280]> hidden_states_437_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_437_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(603170240)))];
            tensor<fp16, []> var_49326_to_fp16 = const()[name = tensor<string, []>("op_49326_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_437_cast_fp16 = layer_norm(axes = hidden_states_437_axes_0, beta = hidden_states_437_beta_0_to_fp16, epsilon = var_49326_to_fp16, gamma = hidden_states_437_gamma_0_to_fp16, x = inputs_327_cast_fp16)[name = tensor<string, []>("hidden_states_437_cast_fp16")];
            tensor<string, []> q_219_pad_type_0 = const()[name = tensor<string, []>("q_219_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_219_strides_0 = const()[name = tensor<string, []>("q_219_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_219_pad_0 = const()[name = tensor<string, []>("q_219_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_219_dilations_0 = const()[name = tensor<string, []>("q_219_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_219_groups_0 = const()[name = tensor<string, []>("q_219_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(603172864))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(604401728))), name = tensor<string, []>("up_blocks_0_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, 1024]> q_219_cast_fp16 = conv(dilations = q_219_dilations_0, groups = q_219_groups_0, pad = q_219_pad_0, pad_type = q_219_pad_type_0, strides = q_219_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_437_cast_fp16)[name = tensor<string, []>("q_219_cast_fp16")];
            tensor<string, []> k_437_pad_type_0 = const()[name = tensor<string, []>("k_437_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_437_strides_0 = const()[name = tensor<string, []>("k_437_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_437_pad_0 = const()[name = tensor<string, []>("k_437_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_437_dilations_0 = const()[name = tensor<string, []>("k_437_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_437_groups_0 = const()[name = tensor<string, []>("k_437_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(604401920))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(606368064))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> k_437_cast_fp16 = conv(dilations = k_437_dilations_0, groups = k_437_groups_0, pad = k_437_pad_0, pad_type = k_437_pad_type_0, strides = k_437_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("k_437_cast_fp16")];
            tensor<string, []> v_219_pad_type_0 = const()[name = tensor<string, []>("v_219_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_219_strides_0 = const()[name = tensor<string, []>("v_219_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_219_pad_0 = const()[name = tensor<string, []>("v_219_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_219_dilations_0 = const()[name = tensor<string, []>("v_219_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_219_groups_0 = const()[name = tensor<string, []>("v_219_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(606368256))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(608334400))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> v_219_cast_fp16 = conv(dilations = v_219_dilations_0, groups = v_219_groups_0, pad = v_219_pad_0, pad_type = v_219_pad_type_0, strides = v_219_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("v_219_cast_fp16")];
            tensor<int32, [4]> var_49359_begin_0 = const()[name = tensor<string, []>("op_49359_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_49359_end_0 = const()[name = tensor<string, []>("op_49359_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_49359_end_mask_0 = const()[name = tensor<string, []>("op_49359_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49359_cast_fp16 = slice_by_index(begin = var_49359_begin_0, end = var_49359_end_0, end_mask = var_49359_end_mask_0, x = q_219_cast_fp16)[name = tensor<string, []>("op_49359_cast_fp16")];
            tensor<int32, [4]> var_49363_begin_0 = const()[name = tensor<string, []>("op_49363_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_49363_end_0 = const()[name = tensor<string, []>("op_49363_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_49363_end_mask_0 = const()[name = tensor<string, []>("op_49363_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49363_cast_fp16 = slice_by_index(begin = var_49363_begin_0, end = var_49363_end_0, end_mask = var_49363_end_mask_0, x = q_219_cast_fp16)[name = tensor<string, []>("op_49363_cast_fp16")];
            tensor<int32, [4]> var_49367_begin_0 = const()[name = tensor<string, []>("op_49367_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_49367_end_0 = const()[name = tensor<string, []>("op_49367_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_49367_end_mask_0 = const()[name = tensor<string, []>("op_49367_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49367_cast_fp16 = slice_by_index(begin = var_49367_begin_0, end = var_49367_end_0, end_mask = var_49367_end_mask_0, x = q_219_cast_fp16)[name = tensor<string, []>("op_49367_cast_fp16")];
            tensor<int32, [4]> var_49371_begin_0 = const()[name = tensor<string, []>("op_49371_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_49371_end_0 = const()[name = tensor<string, []>("op_49371_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_49371_end_mask_0 = const()[name = tensor<string, []>("op_49371_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49371_cast_fp16 = slice_by_index(begin = var_49371_begin_0, end = var_49371_end_0, end_mask = var_49371_end_mask_0, x = q_219_cast_fp16)[name = tensor<string, []>("op_49371_cast_fp16")];
            tensor<int32, [4]> var_49375_begin_0 = const()[name = tensor<string, []>("op_49375_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_49375_end_0 = const()[name = tensor<string, []>("op_49375_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_49375_end_mask_0 = const()[name = tensor<string, []>("op_49375_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49375_cast_fp16 = slice_by_index(begin = var_49375_begin_0, end = var_49375_end_0, end_mask = var_49375_end_mask_0, x = q_219_cast_fp16)[name = tensor<string, []>("op_49375_cast_fp16")];
            tensor<int32, [4]> var_49379_begin_0 = const()[name = tensor<string, []>("op_49379_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_49379_end_0 = const()[name = tensor<string, []>("op_49379_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_49379_end_mask_0 = const()[name = tensor<string, []>("op_49379_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49379_cast_fp16 = slice_by_index(begin = var_49379_begin_0, end = var_49379_end_0, end_mask = var_49379_end_mask_0, x = q_219_cast_fp16)[name = tensor<string, []>("op_49379_cast_fp16")];
            tensor<int32, [4]> var_49383_begin_0 = const()[name = tensor<string, []>("op_49383_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_49383_end_0 = const()[name = tensor<string, []>("op_49383_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_49383_end_mask_0 = const()[name = tensor<string, []>("op_49383_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49383_cast_fp16 = slice_by_index(begin = var_49383_begin_0, end = var_49383_end_0, end_mask = var_49383_end_mask_0, x = q_219_cast_fp16)[name = tensor<string, []>("op_49383_cast_fp16")];
            tensor<int32, [4]> var_49387_begin_0 = const()[name = tensor<string, []>("op_49387_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_49387_end_0 = const()[name = tensor<string, []>("op_49387_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_49387_end_mask_0 = const()[name = tensor<string, []>("op_49387_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49387_cast_fp16 = slice_by_index(begin = var_49387_begin_0, end = var_49387_end_0, end_mask = var_49387_end_mask_0, x = q_219_cast_fp16)[name = tensor<string, []>("op_49387_cast_fp16")];
            tensor<int32, [4]> var_49391_begin_0 = const()[name = tensor<string, []>("op_49391_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_49391_end_0 = const()[name = tensor<string, []>("op_49391_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_49391_end_mask_0 = const()[name = tensor<string, []>("op_49391_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49391_cast_fp16 = slice_by_index(begin = var_49391_begin_0, end = var_49391_end_0, end_mask = var_49391_end_mask_0, x = q_219_cast_fp16)[name = tensor<string, []>("op_49391_cast_fp16")];
            tensor<int32, [4]> var_49395_begin_0 = const()[name = tensor<string, []>("op_49395_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_49395_end_0 = const()[name = tensor<string, []>("op_49395_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_49395_end_mask_0 = const()[name = tensor<string, []>("op_49395_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49395_cast_fp16 = slice_by_index(begin = var_49395_begin_0, end = var_49395_end_0, end_mask = var_49395_end_mask_0, x = q_219_cast_fp16)[name = tensor<string, []>("op_49395_cast_fp16")];
            tensor<int32, [4]> var_49399_begin_0 = const()[name = tensor<string, []>("op_49399_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_49399_end_0 = const()[name = tensor<string, []>("op_49399_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_49399_end_mask_0 = const()[name = tensor<string, []>("op_49399_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49399_cast_fp16 = slice_by_index(begin = var_49399_begin_0, end = var_49399_end_0, end_mask = var_49399_end_mask_0, x = q_219_cast_fp16)[name = tensor<string, []>("op_49399_cast_fp16")];
            tensor<int32, [4]> var_49403_begin_0 = const()[name = tensor<string, []>("op_49403_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_49403_end_0 = const()[name = tensor<string, []>("op_49403_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_49403_end_mask_0 = const()[name = tensor<string, []>("op_49403_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49403_cast_fp16 = slice_by_index(begin = var_49403_begin_0, end = var_49403_end_0, end_mask = var_49403_end_mask_0, x = q_219_cast_fp16)[name = tensor<string, []>("op_49403_cast_fp16")];
            tensor<int32, [4]> var_49407_begin_0 = const()[name = tensor<string, []>("op_49407_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_49407_end_0 = const()[name = tensor<string, []>("op_49407_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_49407_end_mask_0 = const()[name = tensor<string, []>("op_49407_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49407_cast_fp16 = slice_by_index(begin = var_49407_begin_0, end = var_49407_end_0, end_mask = var_49407_end_mask_0, x = q_219_cast_fp16)[name = tensor<string, []>("op_49407_cast_fp16")];
            tensor<int32, [4]> var_49411_begin_0 = const()[name = tensor<string, []>("op_49411_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_49411_end_0 = const()[name = tensor<string, []>("op_49411_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_49411_end_mask_0 = const()[name = tensor<string, []>("op_49411_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49411_cast_fp16 = slice_by_index(begin = var_49411_begin_0, end = var_49411_end_0, end_mask = var_49411_end_mask_0, x = q_219_cast_fp16)[name = tensor<string, []>("op_49411_cast_fp16")];
            tensor<int32, [4]> var_49415_begin_0 = const()[name = tensor<string, []>("op_49415_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_49415_end_0 = const()[name = tensor<string, []>("op_49415_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_49415_end_mask_0 = const()[name = tensor<string, []>("op_49415_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49415_cast_fp16 = slice_by_index(begin = var_49415_begin_0, end = var_49415_end_0, end_mask = var_49415_end_mask_0, x = q_219_cast_fp16)[name = tensor<string, []>("op_49415_cast_fp16")];
            tensor<int32, [4]> var_49419_begin_0 = const()[name = tensor<string, []>("op_49419_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_49419_end_0 = const()[name = tensor<string, []>("op_49419_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_49419_end_mask_0 = const()[name = tensor<string, []>("op_49419_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49419_cast_fp16 = slice_by_index(begin = var_49419_begin_0, end = var_49419_end_0, end_mask = var_49419_end_mask_0, x = q_219_cast_fp16)[name = tensor<string, []>("op_49419_cast_fp16")];
            tensor<int32, [4]> var_49423_begin_0 = const()[name = tensor<string, []>("op_49423_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_49423_end_0 = const()[name = tensor<string, []>("op_49423_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_49423_end_mask_0 = const()[name = tensor<string, []>("op_49423_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49423_cast_fp16 = slice_by_index(begin = var_49423_begin_0, end = var_49423_end_0, end_mask = var_49423_end_mask_0, x = q_219_cast_fp16)[name = tensor<string, []>("op_49423_cast_fp16")];
            tensor<int32, [4]> var_49427_begin_0 = const()[name = tensor<string, []>("op_49427_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_49427_end_0 = const()[name = tensor<string, []>("op_49427_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_49427_end_mask_0 = const()[name = tensor<string, []>("op_49427_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49427_cast_fp16 = slice_by_index(begin = var_49427_begin_0, end = var_49427_end_0, end_mask = var_49427_end_mask_0, x = q_219_cast_fp16)[name = tensor<string, []>("op_49427_cast_fp16")];
            tensor<int32, [4]> var_49431_begin_0 = const()[name = tensor<string, []>("op_49431_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_49431_end_0 = const()[name = tensor<string, []>("op_49431_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_49431_end_mask_0 = const()[name = tensor<string, []>("op_49431_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49431_cast_fp16 = slice_by_index(begin = var_49431_begin_0, end = var_49431_end_0, end_mask = var_49431_end_mask_0, x = q_219_cast_fp16)[name = tensor<string, []>("op_49431_cast_fp16")];
            tensor<int32, [4]> var_49435_begin_0 = const()[name = tensor<string, []>("op_49435_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_49435_end_0 = const()[name = tensor<string, []>("op_49435_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_49435_end_mask_0 = const()[name = tensor<string, []>("op_49435_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49435_cast_fp16 = slice_by_index(begin = var_49435_begin_0, end = var_49435_end_0, end_mask = var_49435_end_mask_0, x = q_219_cast_fp16)[name = tensor<string, []>("op_49435_cast_fp16")];
            tensor<int32, [4]> k_439_perm_0 = const()[name = tensor<string, []>("k_439_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_49442_begin_0 = const()[name = tensor<string, []>("op_49442_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_49442_end_0 = const()[name = tensor<string, []>("op_49442_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_49442_end_mask_0 = const()[name = tensor<string, []>("op_49442_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> k_439_cast_fp16 = transpose(perm = k_439_perm_0, x = k_437_cast_fp16)[name = tensor<string, []>("transpose_30")];
            tensor<fp16, [2, 77, 1, 64]> var_49442_cast_fp16 = slice_by_index(begin = var_49442_begin_0, end = var_49442_end_0, end_mask = var_49442_end_mask_0, x = k_439_cast_fp16)[name = tensor<string, []>("op_49442_cast_fp16")];
            tensor<int32, [4]> var_49446_begin_0 = const()[name = tensor<string, []>("op_49446_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_49446_end_0 = const()[name = tensor<string, []>("op_49446_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_49446_end_mask_0 = const()[name = tensor<string, []>("op_49446_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_49446_cast_fp16 = slice_by_index(begin = var_49446_begin_0, end = var_49446_end_0, end_mask = var_49446_end_mask_0, x = k_439_cast_fp16)[name = tensor<string, []>("op_49446_cast_fp16")];
            tensor<int32, [4]> var_49450_begin_0 = const()[name = tensor<string, []>("op_49450_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_49450_end_0 = const()[name = tensor<string, []>("op_49450_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_49450_end_mask_0 = const()[name = tensor<string, []>("op_49450_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_49450_cast_fp16 = slice_by_index(begin = var_49450_begin_0, end = var_49450_end_0, end_mask = var_49450_end_mask_0, x = k_439_cast_fp16)[name = tensor<string, []>("op_49450_cast_fp16")];
            tensor<int32, [4]> var_49454_begin_0 = const()[name = tensor<string, []>("op_49454_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_49454_end_0 = const()[name = tensor<string, []>("op_49454_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_49454_end_mask_0 = const()[name = tensor<string, []>("op_49454_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_49454_cast_fp16 = slice_by_index(begin = var_49454_begin_0, end = var_49454_end_0, end_mask = var_49454_end_mask_0, x = k_439_cast_fp16)[name = tensor<string, []>("op_49454_cast_fp16")];
            tensor<int32, [4]> var_49458_begin_0 = const()[name = tensor<string, []>("op_49458_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_49458_end_0 = const()[name = tensor<string, []>("op_49458_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_49458_end_mask_0 = const()[name = tensor<string, []>("op_49458_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_49458_cast_fp16 = slice_by_index(begin = var_49458_begin_0, end = var_49458_end_0, end_mask = var_49458_end_mask_0, x = k_439_cast_fp16)[name = tensor<string, []>("op_49458_cast_fp16")];
            tensor<int32, [4]> var_49462_begin_0 = const()[name = tensor<string, []>("op_49462_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_49462_end_0 = const()[name = tensor<string, []>("op_49462_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_49462_end_mask_0 = const()[name = tensor<string, []>("op_49462_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_49462_cast_fp16 = slice_by_index(begin = var_49462_begin_0, end = var_49462_end_0, end_mask = var_49462_end_mask_0, x = k_439_cast_fp16)[name = tensor<string, []>("op_49462_cast_fp16")];
            tensor<int32, [4]> var_49466_begin_0 = const()[name = tensor<string, []>("op_49466_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_49466_end_0 = const()[name = tensor<string, []>("op_49466_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_49466_end_mask_0 = const()[name = tensor<string, []>("op_49466_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_49466_cast_fp16 = slice_by_index(begin = var_49466_begin_0, end = var_49466_end_0, end_mask = var_49466_end_mask_0, x = k_439_cast_fp16)[name = tensor<string, []>("op_49466_cast_fp16")];
            tensor<int32, [4]> var_49470_begin_0 = const()[name = tensor<string, []>("op_49470_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_49470_end_0 = const()[name = tensor<string, []>("op_49470_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_49470_end_mask_0 = const()[name = tensor<string, []>("op_49470_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_49470_cast_fp16 = slice_by_index(begin = var_49470_begin_0, end = var_49470_end_0, end_mask = var_49470_end_mask_0, x = k_439_cast_fp16)[name = tensor<string, []>("op_49470_cast_fp16")];
            tensor<int32, [4]> var_49474_begin_0 = const()[name = tensor<string, []>("op_49474_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_49474_end_0 = const()[name = tensor<string, []>("op_49474_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_49474_end_mask_0 = const()[name = tensor<string, []>("op_49474_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_49474_cast_fp16 = slice_by_index(begin = var_49474_begin_0, end = var_49474_end_0, end_mask = var_49474_end_mask_0, x = k_439_cast_fp16)[name = tensor<string, []>("op_49474_cast_fp16")];
            tensor<int32, [4]> var_49478_begin_0 = const()[name = tensor<string, []>("op_49478_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_49478_end_0 = const()[name = tensor<string, []>("op_49478_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_49478_end_mask_0 = const()[name = tensor<string, []>("op_49478_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_49478_cast_fp16 = slice_by_index(begin = var_49478_begin_0, end = var_49478_end_0, end_mask = var_49478_end_mask_0, x = k_439_cast_fp16)[name = tensor<string, []>("op_49478_cast_fp16")];
            tensor<int32, [4]> var_49482_begin_0 = const()[name = tensor<string, []>("op_49482_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_49482_end_0 = const()[name = tensor<string, []>("op_49482_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_49482_end_mask_0 = const()[name = tensor<string, []>("op_49482_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_49482_cast_fp16 = slice_by_index(begin = var_49482_begin_0, end = var_49482_end_0, end_mask = var_49482_end_mask_0, x = k_439_cast_fp16)[name = tensor<string, []>("op_49482_cast_fp16")];
            tensor<int32, [4]> var_49486_begin_0 = const()[name = tensor<string, []>("op_49486_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_49486_end_0 = const()[name = tensor<string, []>("op_49486_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_49486_end_mask_0 = const()[name = tensor<string, []>("op_49486_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_49486_cast_fp16 = slice_by_index(begin = var_49486_begin_0, end = var_49486_end_0, end_mask = var_49486_end_mask_0, x = k_439_cast_fp16)[name = tensor<string, []>("op_49486_cast_fp16")];
            tensor<int32, [4]> var_49490_begin_0 = const()[name = tensor<string, []>("op_49490_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_49490_end_0 = const()[name = tensor<string, []>("op_49490_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_49490_end_mask_0 = const()[name = tensor<string, []>("op_49490_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_49490_cast_fp16 = slice_by_index(begin = var_49490_begin_0, end = var_49490_end_0, end_mask = var_49490_end_mask_0, x = k_439_cast_fp16)[name = tensor<string, []>("op_49490_cast_fp16")];
            tensor<int32, [4]> var_49494_begin_0 = const()[name = tensor<string, []>("op_49494_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_49494_end_0 = const()[name = tensor<string, []>("op_49494_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_49494_end_mask_0 = const()[name = tensor<string, []>("op_49494_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_49494_cast_fp16 = slice_by_index(begin = var_49494_begin_0, end = var_49494_end_0, end_mask = var_49494_end_mask_0, x = k_439_cast_fp16)[name = tensor<string, []>("op_49494_cast_fp16")];
            tensor<int32, [4]> var_49498_begin_0 = const()[name = tensor<string, []>("op_49498_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_49498_end_0 = const()[name = tensor<string, []>("op_49498_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_49498_end_mask_0 = const()[name = tensor<string, []>("op_49498_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_49498_cast_fp16 = slice_by_index(begin = var_49498_begin_0, end = var_49498_end_0, end_mask = var_49498_end_mask_0, x = k_439_cast_fp16)[name = tensor<string, []>("op_49498_cast_fp16")];
            tensor<int32, [4]> var_49502_begin_0 = const()[name = tensor<string, []>("op_49502_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_49502_end_0 = const()[name = tensor<string, []>("op_49502_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_49502_end_mask_0 = const()[name = tensor<string, []>("op_49502_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_49502_cast_fp16 = slice_by_index(begin = var_49502_begin_0, end = var_49502_end_0, end_mask = var_49502_end_mask_0, x = k_439_cast_fp16)[name = tensor<string, []>("op_49502_cast_fp16")];
            tensor<int32, [4]> var_49506_begin_0 = const()[name = tensor<string, []>("op_49506_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_49506_end_0 = const()[name = tensor<string, []>("op_49506_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_49506_end_mask_0 = const()[name = tensor<string, []>("op_49506_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_49506_cast_fp16 = slice_by_index(begin = var_49506_begin_0, end = var_49506_end_0, end_mask = var_49506_end_mask_0, x = k_439_cast_fp16)[name = tensor<string, []>("op_49506_cast_fp16")];
            tensor<int32, [4]> var_49510_begin_0 = const()[name = tensor<string, []>("op_49510_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_49510_end_0 = const()[name = tensor<string, []>("op_49510_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_49510_end_mask_0 = const()[name = tensor<string, []>("op_49510_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_49510_cast_fp16 = slice_by_index(begin = var_49510_begin_0, end = var_49510_end_0, end_mask = var_49510_end_mask_0, x = k_439_cast_fp16)[name = tensor<string, []>("op_49510_cast_fp16")];
            tensor<int32, [4]> var_49514_begin_0 = const()[name = tensor<string, []>("op_49514_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_49514_end_0 = const()[name = tensor<string, []>("op_49514_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_49514_end_mask_0 = const()[name = tensor<string, []>("op_49514_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_49514_cast_fp16 = slice_by_index(begin = var_49514_begin_0, end = var_49514_end_0, end_mask = var_49514_end_mask_0, x = k_439_cast_fp16)[name = tensor<string, []>("op_49514_cast_fp16")];
            tensor<int32, [4]> var_49518_begin_0 = const()[name = tensor<string, []>("op_49518_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_49518_end_0 = const()[name = tensor<string, []>("op_49518_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_49518_end_mask_0 = const()[name = tensor<string, []>("op_49518_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_49518_cast_fp16 = slice_by_index(begin = var_49518_begin_0, end = var_49518_end_0, end_mask = var_49518_end_mask_0, x = k_439_cast_fp16)[name = tensor<string, []>("op_49518_cast_fp16")];
            tensor<int32, [4]> var_49520_begin_0 = const()[name = tensor<string, []>("op_49520_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_49520_end_0 = const()[name = tensor<string, []>("op_49520_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_49520_end_mask_0 = const()[name = tensor<string, []>("op_49520_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_49520_cast_fp16 = slice_by_index(begin = var_49520_begin_0, end = var_49520_end_0, end_mask = var_49520_end_mask_0, x = v_219_cast_fp16)[name = tensor<string, []>("op_49520_cast_fp16")];
            tensor<int32, [4]> var_49524_begin_0 = const()[name = tensor<string, []>("op_49524_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_49524_end_0 = const()[name = tensor<string, []>("op_49524_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_49524_end_mask_0 = const()[name = tensor<string, []>("op_49524_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_49524_cast_fp16 = slice_by_index(begin = var_49524_begin_0, end = var_49524_end_0, end_mask = var_49524_end_mask_0, x = v_219_cast_fp16)[name = tensor<string, []>("op_49524_cast_fp16")];
            tensor<int32, [4]> var_49528_begin_0 = const()[name = tensor<string, []>("op_49528_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_49528_end_0 = const()[name = tensor<string, []>("op_49528_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_49528_end_mask_0 = const()[name = tensor<string, []>("op_49528_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_49528_cast_fp16 = slice_by_index(begin = var_49528_begin_0, end = var_49528_end_0, end_mask = var_49528_end_mask_0, x = v_219_cast_fp16)[name = tensor<string, []>("op_49528_cast_fp16")];
            tensor<int32, [4]> var_49532_begin_0 = const()[name = tensor<string, []>("op_49532_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_49532_end_0 = const()[name = tensor<string, []>("op_49532_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_49532_end_mask_0 = const()[name = tensor<string, []>("op_49532_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_49532_cast_fp16 = slice_by_index(begin = var_49532_begin_0, end = var_49532_end_0, end_mask = var_49532_end_mask_0, x = v_219_cast_fp16)[name = tensor<string, []>("op_49532_cast_fp16")];
            tensor<int32, [4]> var_49536_begin_0 = const()[name = tensor<string, []>("op_49536_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_49536_end_0 = const()[name = tensor<string, []>("op_49536_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_49536_end_mask_0 = const()[name = tensor<string, []>("op_49536_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_49536_cast_fp16 = slice_by_index(begin = var_49536_begin_0, end = var_49536_end_0, end_mask = var_49536_end_mask_0, x = v_219_cast_fp16)[name = tensor<string, []>("op_49536_cast_fp16")];
            tensor<int32, [4]> var_49540_begin_0 = const()[name = tensor<string, []>("op_49540_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_49540_end_0 = const()[name = tensor<string, []>("op_49540_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_49540_end_mask_0 = const()[name = tensor<string, []>("op_49540_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_49540_cast_fp16 = slice_by_index(begin = var_49540_begin_0, end = var_49540_end_0, end_mask = var_49540_end_mask_0, x = v_219_cast_fp16)[name = tensor<string, []>("op_49540_cast_fp16")];
            tensor<int32, [4]> var_49544_begin_0 = const()[name = tensor<string, []>("op_49544_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_49544_end_0 = const()[name = tensor<string, []>("op_49544_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_49544_end_mask_0 = const()[name = tensor<string, []>("op_49544_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_49544_cast_fp16 = slice_by_index(begin = var_49544_begin_0, end = var_49544_end_0, end_mask = var_49544_end_mask_0, x = v_219_cast_fp16)[name = tensor<string, []>("op_49544_cast_fp16")];
            tensor<int32, [4]> var_49548_begin_0 = const()[name = tensor<string, []>("op_49548_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_49548_end_0 = const()[name = tensor<string, []>("op_49548_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_49548_end_mask_0 = const()[name = tensor<string, []>("op_49548_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_49548_cast_fp16 = slice_by_index(begin = var_49548_begin_0, end = var_49548_end_0, end_mask = var_49548_end_mask_0, x = v_219_cast_fp16)[name = tensor<string, []>("op_49548_cast_fp16")];
            tensor<int32, [4]> var_49552_begin_0 = const()[name = tensor<string, []>("op_49552_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_49552_end_0 = const()[name = tensor<string, []>("op_49552_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_49552_end_mask_0 = const()[name = tensor<string, []>("op_49552_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_49552_cast_fp16 = slice_by_index(begin = var_49552_begin_0, end = var_49552_end_0, end_mask = var_49552_end_mask_0, x = v_219_cast_fp16)[name = tensor<string, []>("op_49552_cast_fp16")];
            tensor<int32, [4]> var_49556_begin_0 = const()[name = tensor<string, []>("op_49556_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_49556_end_0 = const()[name = tensor<string, []>("op_49556_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_49556_end_mask_0 = const()[name = tensor<string, []>("op_49556_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_49556_cast_fp16 = slice_by_index(begin = var_49556_begin_0, end = var_49556_end_0, end_mask = var_49556_end_mask_0, x = v_219_cast_fp16)[name = tensor<string, []>("op_49556_cast_fp16")];
            tensor<int32, [4]> var_49560_begin_0 = const()[name = tensor<string, []>("op_49560_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_49560_end_0 = const()[name = tensor<string, []>("op_49560_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_49560_end_mask_0 = const()[name = tensor<string, []>("op_49560_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_49560_cast_fp16 = slice_by_index(begin = var_49560_begin_0, end = var_49560_end_0, end_mask = var_49560_end_mask_0, x = v_219_cast_fp16)[name = tensor<string, []>("op_49560_cast_fp16")];
            tensor<int32, [4]> var_49564_begin_0 = const()[name = tensor<string, []>("op_49564_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_49564_end_0 = const()[name = tensor<string, []>("op_49564_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_49564_end_mask_0 = const()[name = tensor<string, []>("op_49564_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_49564_cast_fp16 = slice_by_index(begin = var_49564_begin_0, end = var_49564_end_0, end_mask = var_49564_end_mask_0, x = v_219_cast_fp16)[name = tensor<string, []>("op_49564_cast_fp16")];
            tensor<int32, [4]> var_49568_begin_0 = const()[name = tensor<string, []>("op_49568_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_49568_end_0 = const()[name = tensor<string, []>("op_49568_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_49568_end_mask_0 = const()[name = tensor<string, []>("op_49568_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_49568_cast_fp16 = slice_by_index(begin = var_49568_begin_0, end = var_49568_end_0, end_mask = var_49568_end_mask_0, x = v_219_cast_fp16)[name = tensor<string, []>("op_49568_cast_fp16")];
            tensor<int32, [4]> var_49572_begin_0 = const()[name = tensor<string, []>("op_49572_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_49572_end_0 = const()[name = tensor<string, []>("op_49572_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_49572_end_mask_0 = const()[name = tensor<string, []>("op_49572_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_49572_cast_fp16 = slice_by_index(begin = var_49572_begin_0, end = var_49572_end_0, end_mask = var_49572_end_mask_0, x = v_219_cast_fp16)[name = tensor<string, []>("op_49572_cast_fp16")];
            tensor<int32, [4]> var_49576_begin_0 = const()[name = tensor<string, []>("op_49576_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_49576_end_0 = const()[name = tensor<string, []>("op_49576_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_49576_end_mask_0 = const()[name = tensor<string, []>("op_49576_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_49576_cast_fp16 = slice_by_index(begin = var_49576_begin_0, end = var_49576_end_0, end_mask = var_49576_end_mask_0, x = v_219_cast_fp16)[name = tensor<string, []>("op_49576_cast_fp16")];
            tensor<int32, [4]> var_49580_begin_0 = const()[name = tensor<string, []>("op_49580_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_49580_end_0 = const()[name = tensor<string, []>("op_49580_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_49580_end_mask_0 = const()[name = tensor<string, []>("op_49580_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_49580_cast_fp16 = slice_by_index(begin = var_49580_begin_0, end = var_49580_end_0, end_mask = var_49580_end_mask_0, x = v_219_cast_fp16)[name = tensor<string, []>("op_49580_cast_fp16")];
            tensor<int32, [4]> var_49584_begin_0 = const()[name = tensor<string, []>("op_49584_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_49584_end_0 = const()[name = tensor<string, []>("op_49584_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_49584_end_mask_0 = const()[name = tensor<string, []>("op_49584_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_49584_cast_fp16 = slice_by_index(begin = var_49584_begin_0, end = var_49584_end_0, end_mask = var_49584_end_mask_0, x = v_219_cast_fp16)[name = tensor<string, []>("op_49584_cast_fp16")];
            tensor<int32, [4]> var_49588_begin_0 = const()[name = tensor<string, []>("op_49588_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_49588_end_0 = const()[name = tensor<string, []>("op_49588_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_49588_end_mask_0 = const()[name = tensor<string, []>("op_49588_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_49588_cast_fp16 = slice_by_index(begin = var_49588_begin_0, end = var_49588_end_0, end_mask = var_49588_end_mask_0, x = v_219_cast_fp16)[name = tensor<string, []>("op_49588_cast_fp16")];
            tensor<int32, [4]> var_49592_begin_0 = const()[name = tensor<string, []>("op_49592_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_49592_end_0 = const()[name = tensor<string, []>("op_49592_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_49592_end_mask_0 = const()[name = tensor<string, []>("op_49592_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_49592_cast_fp16 = slice_by_index(begin = var_49592_begin_0, end = var_49592_end_0, end_mask = var_49592_end_mask_0, x = v_219_cast_fp16)[name = tensor<string, []>("op_49592_cast_fp16")];
            tensor<int32, [4]> var_49596_begin_0 = const()[name = tensor<string, []>("op_49596_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_49596_end_0 = const()[name = tensor<string, []>("op_49596_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_49596_end_mask_0 = const()[name = tensor<string, []>("op_49596_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_49596_cast_fp16 = slice_by_index(begin = var_49596_begin_0, end = var_49596_end_0, end_mask = var_49596_end_mask_0, x = v_219_cast_fp16)[name = tensor<string, []>("op_49596_cast_fp16")];
            tensor<string, []> var_49600_equation_0 = const()[name = tensor<string, []>("op_49600_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_49600_cast_fp16 = einsum(equation = var_49600_equation_0, values = (var_49442_cast_fp16, var_49359_cast_fp16))[name = tensor<string, []>("op_49600_cast_fp16")];
            tensor<fp16, []> var_49601_to_fp16 = const()[name = tensor<string, []>("op_49601_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4201_cast_fp16 = mul(x = var_49600_cast_fp16, y = var_49601_to_fp16)[name = tensor<string, []>("aw_4201_cast_fp16")];
            tensor<string, []> var_49604_equation_0 = const()[name = tensor<string, []>("op_49604_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_49604_cast_fp16 = einsum(equation = var_49604_equation_0, values = (var_49446_cast_fp16, var_49363_cast_fp16))[name = tensor<string, []>("op_49604_cast_fp16")];
            tensor<fp16, []> var_49605_to_fp16 = const()[name = tensor<string, []>("op_49605_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4203_cast_fp16 = mul(x = var_49604_cast_fp16, y = var_49605_to_fp16)[name = tensor<string, []>("aw_4203_cast_fp16")];
            tensor<string, []> var_49608_equation_0 = const()[name = tensor<string, []>("op_49608_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_49608_cast_fp16 = einsum(equation = var_49608_equation_0, values = (var_49450_cast_fp16, var_49367_cast_fp16))[name = tensor<string, []>("op_49608_cast_fp16")];
            tensor<fp16, []> var_49609_to_fp16 = const()[name = tensor<string, []>("op_49609_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4205_cast_fp16 = mul(x = var_49608_cast_fp16, y = var_49609_to_fp16)[name = tensor<string, []>("aw_4205_cast_fp16")];
            tensor<string, []> var_49612_equation_0 = const()[name = tensor<string, []>("op_49612_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_49612_cast_fp16 = einsum(equation = var_49612_equation_0, values = (var_49454_cast_fp16, var_49371_cast_fp16))[name = tensor<string, []>("op_49612_cast_fp16")];
            tensor<fp16, []> var_49613_to_fp16 = const()[name = tensor<string, []>("op_49613_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4207_cast_fp16 = mul(x = var_49612_cast_fp16, y = var_49613_to_fp16)[name = tensor<string, []>("aw_4207_cast_fp16")];
            tensor<string, []> var_49616_equation_0 = const()[name = tensor<string, []>("op_49616_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_49616_cast_fp16 = einsum(equation = var_49616_equation_0, values = (var_49458_cast_fp16, var_49375_cast_fp16))[name = tensor<string, []>("op_49616_cast_fp16")];
            tensor<fp16, []> var_49617_to_fp16 = const()[name = tensor<string, []>("op_49617_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4209_cast_fp16 = mul(x = var_49616_cast_fp16, y = var_49617_to_fp16)[name = tensor<string, []>("aw_4209_cast_fp16")];
            tensor<string, []> var_49620_equation_0 = const()[name = tensor<string, []>("op_49620_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_49620_cast_fp16 = einsum(equation = var_49620_equation_0, values = (var_49462_cast_fp16, var_49379_cast_fp16))[name = tensor<string, []>("op_49620_cast_fp16")];
            tensor<fp16, []> var_49621_to_fp16 = const()[name = tensor<string, []>("op_49621_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4211_cast_fp16 = mul(x = var_49620_cast_fp16, y = var_49621_to_fp16)[name = tensor<string, []>("aw_4211_cast_fp16")];
            tensor<string, []> var_49624_equation_0 = const()[name = tensor<string, []>("op_49624_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_49624_cast_fp16 = einsum(equation = var_49624_equation_0, values = (var_49466_cast_fp16, var_49383_cast_fp16))[name = tensor<string, []>("op_49624_cast_fp16")];
            tensor<fp16, []> var_49625_to_fp16 = const()[name = tensor<string, []>("op_49625_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4213_cast_fp16 = mul(x = var_49624_cast_fp16, y = var_49625_to_fp16)[name = tensor<string, []>("aw_4213_cast_fp16")];
            tensor<string, []> var_49628_equation_0 = const()[name = tensor<string, []>("op_49628_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_49628_cast_fp16 = einsum(equation = var_49628_equation_0, values = (var_49470_cast_fp16, var_49387_cast_fp16))[name = tensor<string, []>("op_49628_cast_fp16")];
            tensor<fp16, []> var_49629_to_fp16 = const()[name = tensor<string, []>("op_49629_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4215_cast_fp16 = mul(x = var_49628_cast_fp16, y = var_49629_to_fp16)[name = tensor<string, []>("aw_4215_cast_fp16")];
            tensor<string, []> var_49632_equation_0 = const()[name = tensor<string, []>("op_49632_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_49632_cast_fp16 = einsum(equation = var_49632_equation_0, values = (var_49474_cast_fp16, var_49391_cast_fp16))[name = tensor<string, []>("op_49632_cast_fp16")];
            tensor<fp16, []> var_49633_to_fp16 = const()[name = tensor<string, []>("op_49633_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4217_cast_fp16 = mul(x = var_49632_cast_fp16, y = var_49633_to_fp16)[name = tensor<string, []>("aw_4217_cast_fp16")];
            tensor<string, []> var_49636_equation_0 = const()[name = tensor<string, []>("op_49636_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_49636_cast_fp16 = einsum(equation = var_49636_equation_0, values = (var_49478_cast_fp16, var_49395_cast_fp16))[name = tensor<string, []>("op_49636_cast_fp16")];
            tensor<fp16, []> var_49637_to_fp16 = const()[name = tensor<string, []>("op_49637_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4219_cast_fp16 = mul(x = var_49636_cast_fp16, y = var_49637_to_fp16)[name = tensor<string, []>("aw_4219_cast_fp16")];
            tensor<string, []> var_49640_equation_0 = const()[name = tensor<string, []>("op_49640_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_49640_cast_fp16 = einsum(equation = var_49640_equation_0, values = (var_49482_cast_fp16, var_49399_cast_fp16))[name = tensor<string, []>("op_49640_cast_fp16")];
            tensor<fp16, []> var_49641_to_fp16 = const()[name = tensor<string, []>("op_49641_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4221_cast_fp16 = mul(x = var_49640_cast_fp16, y = var_49641_to_fp16)[name = tensor<string, []>("aw_4221_cast_fp16")];
            tensor<string, []> var_49644_equation_0 = const()[name = tensor<string, []>("op_49644_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_49644_cast_fp16 = einsum(equation = var_49644_equation_0, values = (var_49486_cast_fp16, var_49403_cast_fp16))[name = tensor<string, []>("op_49644_cast_fp16")];
            tensor<fp16, []> var_49645_to_fp16 = const()[name = tensor<string, []>("op_49645_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4223_cast_fp16 = mul(x = var_49644_cast_fp16, y = var_49645_to_fp16)[name = tensor<string, []>("aw_4223_cast_fp16")];
            tensor<string, []> var_49648_equation_0 = const()[name = tensor<string, []>("op_49648_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_49648_cast_fp16 = einsum(equation = var_49648_equation_0, values = (var_49490_cast_fp16, var_49407_cast_fp16))[name = tensor<string, []>("op_49648_cast_fp16")];
            tensor<fp16, []> var_49649_to_fp16 = const()[name = tensor<string, []>("op_49649_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4225_cast_fp16 = mul(x = var_49648_cast_fp16, y = var_49649_to_fp16)[name = tensor<string, []>("aw_4225_cast_fp16")];
            tensor<string, []> var_49652_equation_0 = const()[name = tensor<string, []>("op_49652_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_49652_cast_fp16 = einsum(equation = var_49652_equation_0, values = (var_49494_cast_fp16, var_49411_cast_fp16))[name = tensor<string, []>("op_49652_cast_fp16")];
            tensor<fp16, []> var_49653_to_fp16 = const()[name = tensor<string, []>("op_49653_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4227_cast_fp16 = mul(x = var_49652_cast_fp16, y = var_49653_to_fp16)[name = tensor<string, []>("aw_4227_cast_fp16")];
            tensor<string, []> var_49656_equation_0 = const()[name = tensor<string, []>("op_49656_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_49656_cast_fp16 = einsum(equation = var_49656_equation_0, values = (var_49498_cast_fp16, var_49415_cast_fp16))[name = tensor<string, []>("op_49656_cast_fp16")];
            tensor<fp16, []> var_49657_to_fp16 = const()[name = tensor<string, []>("op_49657_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4229_cast_fp16 = mul(x = var_49656_cast_fp16, y = var_49657_to_fp16)[name = tensor<string, []>("aw_4229_cast_fp16")];
            tensor<string, []> var_49660_equation_0 = const()[name = tensor<string, []>("op_49660_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_49660_cast_fp16 = einsum(equation = var_49660_equation_0, values = (var_49502_cast_fp16, var_49419_cast_fp16))[name = tensor<string, []>("op_49660_cast_fp16")];
            tensor<fp16, []> var_49661_to_fp16 = const()[name = tensor<string, []>("op_49661_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4231_cast_fp16 = mul(x = var_49660_cast_fp16, y = var_49661_to_fp16)[name = tensor<string, []>("aw_4231_cast_fp16")];
            tensor<string, []> var_49664_equation_0 = const()[name = tensor<string, []>("op_49664_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_49664_cast_fp16 = einsum(equation = var_49664_equation_0, values = (var_49506_cast_fp16, var_49423_cast_fp16))[name = tensor<string, []>("op_49664_cast_fp16")];
            tensor<fp16, []> var_49665_to_fp16 = const()[name = tensor<string, []>("op_49665_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4233_cast_fp16 = mul(x = var_49664_cast_fp16, y = var_49665_to_fp16)[name = tensor<string, []>("aw_4233_cast_fp16")];
            tensor<string, []> var_49668_equation_0 = const()[name = tensor<string, []>("op_49668_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_49668_cast_fp16 = einsum(equation = var_49668_equation_0, values = (var_49510_cast_fp16, var_49427_cast_fp16))[name = tensor<string, []>("op_49668_cast_fp16")];
            tensor<fp16, []> var_49669_to_fp16 = const()[name = tensor<string, []>("op_49669_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4235_cast_fp16 = mul(x = var_49668_cast_fp16, y = var_49669_to_fp16)[name = tensor<string, []>("aw_4235_cast_fp16")];
            tensor<string, []> var_49672_equation_0 = const()[name = tensor<string, []>("op_49672_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_49672_cast_fp16 = einsum(equation = var_49672_equation_0, values = (var_49514_cast_fp16, var_49431_cast_fp16))[name = tensor<string, []>("op_49672_cast_fp16")];
            tensor<fp16, []> var_49673_to_fp16 = const()[name = tensor<string, []>("op_49673_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4237_cast_fp16 = mul(x = var_49672_cast_fp16, y = var_49673_to_fp16)[name = tensor<string, []>("aw_4237_cast_fp16")];
            tensor<string, []> var_49676_equation_0 = const()[name = tensor<string, []>("op_49676_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_49676_cast_fp16 = einsum(equation = var_49676_equation_0, values = (var_49518_cast_fp16, var_49435_cast_fp16))[name = tensor<string, []>("op_49676_cast_fp16")];
            tensor<fp16, []> var_49677_to_fp16 = const()[name = tensor<string, []>("op_49677_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4239_cast_fp16 = mul(x = var_49676_cast_fp16, y = var_49677_to_fp16)[name = tensor<string, []>("aw_4239_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_49679_cast_fp16 = softmax(axis = var_30355, x = aw_4201_cast_fp16)[name = tensor<string, []>("op_49679_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_49680_cast_fp16 = softmax(axis = var_30355, x = aw_4203_cast_fp16)[name = tensor<string, []>("op_49680_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_49681_cast_fp16 = softmax(axis = var_30355, x = aw_4205_cast_fp16)[name = tensor<string, []>("op_49681_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_49682_cast_fp16 = softmax(axis = var_30355, x = aw_4207_cast_fp16)[name = tensor<string, []>("op_49682_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_49683_cast_fp16 = softmax(axis = var_30355, x = aw_4209_cast_fp16)[name = tensor<string, []>("op_49683_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_49684_cast_fp16 = softmax(axis = var_30355, x = aw_4211_cast_fp16)[name = tensor<string, []>("op_49684_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_49685_cast_fp16 = softmax(axis = var_30355, x = aw_4213_cast_fp16)[name = tensor<string, []>("op_49685_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_49686_cast_fp16 = softmax(axis = var_30355, x = aw_4215_cast_fp16)[name = tensor<string, []>("op_49686_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_49687_cast_fp16 = softmax(axis = var_30355, x = aw_4217_cast_fp16)[name = tensor<string, []>("op_49687_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_49688_cast_fp16 = softmax(axis = var_30355, x = aw_4219_cast_fp16)[name = tensor<string, []>("op_49688_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_49689_cast_fp16 = softmax(axis = var_30355, x = aw_4221_cast_fp16)[name = tensor<string, []>("op_49689_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_49690_cast_fp16 = softmax(axis = var_30355, x = aw_4223_cast_fp16)[name = tensor<string, []>("op_49690_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_49691_cast_fp16 = softmax(axis = var_30355, x = aw_4225_cast_fp16)[name = tensor<string, []>("op_49691_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_49692_cast_fp16 = softmax(axis = var_30355, x = aw_4227_cast_fp16)[name = tensor<string, []>("op_49692_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_49693_cast_fp16 = softmax(axis = var_30355, x = aw_4229_cast_fp16)[name = tensor<string, []>("op_49693_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_49694_cast_fp16 = softmax(axis = var_30355, x = aw_4231_cast_fp16)[name = tensor<string, []>("op_49694_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_49695_cast_fp16 = softmax(axis = var_30355, x = aw_4233_cast_fp16)[name = tensor<string, []>("op_49695_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_49696_cast_fp16 = softmax(axis = var_30355, x = aw_4235_cast_fp16)[name = tensor<string, []>("op_49696_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_49697_cast_fp16 = softmax(axis = var_30355, x = aw_4237_cast_fp16)[name = tensor<string, []>("op_49697_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_49698_cast_fp16 = softmax(axis = var_30355, x = aw_4239_cast_fp16)[name = tensor<string, []>("op_49698_cast_fp16")];
            tensor<string, []> var_49700_equation_0 = const()[name = tensor<string, []>("op_49700_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49700_cast_fp16 = einsum(equation = var_49700_equation_0, values = (var_49520_cast_fp16, var_49679_cast_fp16))[name = tensor<string, []>("op_49700_cast_fp16")];
            tensor<string, []> var_49702_equation_0 = const()[name = tensor<string, []>("op_49702_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49702_cast_fp16 = einsum(equation = var_49702_equation_0, values = (var_49524_cast_fp16, var_49680_cast_fp16))[name = tensor<string, []>("op_49702_cast_fp16")];
            tensor<string, []> var_49704_equation_0 = const()[name = tensor<string, []>("op_49704_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49704_cast_fp16 = einsum(equation = var_49704_equation_0, values = (var_49528_cast_fp16, var_49681_cast_fp16))[name = tensor<string, []>("op_49704_cast_fp16")];
            tensor<string, []> var_49706_equation_0 = const()[name = tensor<string, []>("op_49706_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49706_cast_fp16 = einsum(equation = var_49706_equation_0, values = (var_49532_cast_fp16, var_49682_cast_fp16))[name = tensor<string, []>("op_49706_cast_fp16")];
            tensor<string, []> var_49708_equation_0 = const()[name = tensor<string, []>("op_49708_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49708_cast_fp16 = einsum(equation = var_49708_equation_0, values = (var_49536_cast_fp16, var_49683_cast_fp16))[name = tensor<string, []>("op_49708_cast_fp16")];
            tensor<string, []> var_49710_equation_0 = const()[name = tensor<string, []>("op_49710_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49710_cast_fp16 = einsum(equation = var_49710_equation_0, values = (var_49540_cast_fp16, var_49684_cast_fp16))[name = tensor<string, []>("op_49710_cast_fp16")];
            tensor<string, []> var_49712_equation_0 = const()[name = tensor<string, []>("op_49712_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49712_cast_fp16 = einsum(equation = var_49712_equation_0, values = (var_49544_cast_fp16, var_49685_cast_fp16))[name = tensor<string, []>("op_49712_cast_fp16")];
            tensor<string, []> var_49714_equation_0 = const()[name = tensor<string, []>("op_49714_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49714_cast_fp16 = einsum(equation = var_49714_equation_0, values = (var_49548_cast_fp16, var_49686_cast_fp16))[name = tensor<string, []>("op_49714_cast_fp16")];
            tensor<string, []> var_49716_equation_0 = const()[name = tensor<string, []>("op_49716_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49716_cast_fp16 = einsum(equation = var_49716_equation_0, values = (var_49552_cast_fp16, var_49687_cast_fp16))[name = tensor<string, []>("op_49716_cast_fp16")];
            tensor<string, []> var_49718_equation_0 = const()[name = tensor<string, []>("op_49718_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49718_cast_fp16 = einsum(equation = var_49718_equation_0, values = (var_49556_cast_fp16, var_49688_cast_fp16))[name = tensor<string, []>("op_49718_cast_fp16")];
            tensor<string, []> var_49720_equation_0 = const()[name = tensor<string, []>("op_49720_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49720_cast_fp16 = einsum(equation = var_49720_equation_0, values = (var_49560_cast_fp16, var_49689_cast_fp16))[name = tensor<string, []>("op_49720_cast_fp16")];
            tensor<string, []> var_49722_equation_0 = const()[name = tensor<string, []>("op_49722_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49722_cast_fp16 = einsum(equation = var_49722_equation_0, values = (var_49564_cast_fp16, var_49690_cast_fp16))[name = tensor<string, []>("op_49722_cast_fp16")];
            tensor<string, []> var_49724_equation_0 = const()[name = tensor<string, []>("op_49724_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49724_cast_fp16 = einsum(equation = var_49724_equation_0, values = (var_49568_cast_fp16, var_49691_cast_fp16))[name = tensor<string, []>("op_49724_cast_fp16")];
            tensor<string, []> var_49726_equation_0 = const()[name = tensor<string, []>("op_49726_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49726_cast_fp16 = einsum(equation = var_49726_equation_0, values = (var_49572_cast_fp16, var_49692_cast_fp16))[name = tensor<string, []>("op_49726_cast_fp16")];
            tensor<string, []> var_49728_equation_0 = const()[name = tensor<string, []>("op_49728_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49728_cast_fp16 = einsum(equation = var_49728_equation_0, values = (var_49576_cast_fp16, var_49693_cast_fp16))[name = tensor<string, []>("op_49728_cast_fp16")];
            tensor<string, []> var_49730_equation_0 = const()[name = tensor<string, []>("op_49730_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49730_cast_fp16 = einsum(equation = var_49730_equation_0, values = (var_49580_cast_fp16, var_49694_cast_fp16))[name = tensor<string, []>("op_49730_cast_fp16")];
            tensor<string, []> var_49732_equation_0 = const()[name = tensor<string, []>("op_49732_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49732_cast_fp16 = einsum(equation = var_49732_equation_0, values = (var_49584_cast_fp16, var_49695_cast_fp16))[name = tensor<string, []>("op_49732_cast_fp16")];
            tensor<string, []> var_49734_equation_0 = const()[name = tensor<string, []>("op_49734_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49734_cast_fp16 = einsum(equation = var_49734_equation_0, values = (var_49588_cast_fp16, var_49696_cast_fp16))[name = tensor<string, []>("op_49734_cast_fp16")];
            tensor<string, []> var_49736_equation_0 = const()[name = tensor<string, []>("op_49736_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49736_cast_fp16 = einsum(equation = var_49736_equation_0, values = (var_49592_cast_fp16, var_49697_cast_fp16))[name = tensor<string, []>("op_49736_cast_fp16")];
            tensor<string, []> var_49738_equation_0 = const()[name = tensor<string, []>("op_49738_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49738_cast_fp16 = einsum(equation = var_49738_equation_0, values = (var_49596_cast_fp16, var_49698_cast_fp16))[name = tensor<string, []>("op_49738_cast_fp16")];
            tensor<bool, []> input_637_interleave_0 = const()[name = tensor<string, []>("input_637_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_637_cast_fp16 = concat(axis = var_30355, interleave = input_637_interleave_0, values = (var_49700_cast_fp16, var_49702_cast_fp16, var_49704_cast_fp16, var_49706_cast_fp16, var_49708_cast_fp16, var_49710_cast_fp16, var_49712_cast_fp16, var_49714_cast_fp16, var_49716_cast_fp16, var_49718_cast_fp16, var_49720_cast_fp16, var_49722_cast_fp16, var_49724_cast_fp16, var_49726_cast_fp16, var_49728_cast_fp16, var_49730_cast_fp16, var_49732_cast_fp16, var_49734_cast_fp16, var_49736_cast_fp16, var_49738_cast_fp16))[name = tensor<string, []>("input_637_cast_fp16")];
            tensor<string, []> var_49748_pad_type_0 = const()[name = tensor<string, []>("op_49748_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_49748_strides_0 = const()[name = tensor<string, []>("op_49748_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_49748_pad_0 = const()[name = tensor<string, []>("op_49748_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_49748_dilations_0 = const()[name = tensor<string, []>("op_49748_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_49748_groups_0 = const()[name = tensor<string, []>("op_49748_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(608334592))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(609563456))), name = tensor<string, []>("up_blocks_0_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_0_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_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, []>(609563648)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_49748_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_49748_dilations_0, groups = var_49748_groups_0, pad = var_49748_pad_0, pad_type = var_49748_pad_type_0, strides = var_49748_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized, x = input_637_cast_fp16)[name = tensor<string, []>("op_49748_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_329_cast_fp16 = add(x = var_49748_cast_fp16, y = inputs_327_cast_fp16)[name = tensor<string, []>("inputs_329_cast_fp16")];
            tensor<int32, [1]> input_639_axes_0 = const()[name = tensor<string, []>("input_639_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> input_639_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_639_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(609566272)))];
            tensor<fp16, [1280]> input_639_beta_0_to_fp16 = const()[name = tensor<string, []>("input_639_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(609568896)))];
            tensor<fp16, []> var_49758_to_fp16 = const()[name = tensor<string, []>("op_49758_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> input_639_cast_fp16 = layer_norm(axes = input_639_axes_0, beta = input_639_beta_0_to_fp16, epsilon = var_49758_to_fp16, gamma = input_639_gamma_0_to_fp16, x = inputs_329_cast_fp16)[name = tensor<string, []>("input_639_cast_fp16")];
            tensor<string, []> var_49778_pad_type_0 = const()[name = tensor<string, []>("op_49778_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_49778_strides_0 = const()[name = tensor<string, []>("op_49778_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_49778_pad_0 = const()[name = tensor<string, []>("op_49778_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_49778_dilations_0 = const()[name = tensor<string, []>("op_49778_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_49778_groups_0 = const()[name = tensor<string, []>("op_49778_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [9830400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(609571520))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(619401984))), name = tensor<string, []>("up_blocks_0_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_0_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(619402176)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_49778_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_49778_dilations_0, groups = var_49778_groups_0, pad = var_49778_pad_0, pad_type = var_49778_pad_type_0, strides = var_49778_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized, x = input_639_cast_fp16)[name = tensor<string, []>("op_49778_cast_fp16")];
            tensor<int32, [2]> var_49779_split_sizes_0 = const()[name = tensor<string, []>("op_49779_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_49779_axis_0 = const()[name = tensor<string, []>("op_49779_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_49779_cast_fp16_0, tensor<fp16, [2, 5120, 1, 1024]> var_49779_cast_fp16_1 = split(axis = var_49779_axis_0, split_sizes = var_49779_split_sizes_0, x = var_49778_cast_fp16)[name = tensor<string, []>("op_49779_cast_fp16")];
            tensor<string, []> var_49781_mode_0 = const()[name = tensor<string, []>("op_49781_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_49781_cast_fp16 = gelu(mode = var_49781_mode_0, x = var_49779_cast_fp16_1)[name = tensor<string, []>("op_49781_cast_fp16")];
            tensor<fp16, [2, 5120, 1, 1024]> input_641_cast_fp16 = mul(x = var_49779_cast_fp16_0, y = var_49781_cast_fp16)[name = tensor<string, []>("input_641_cast_fp16")];
            tensor<string, []> var_49789_pad_type_0 = const()[name = tensor<string, []>("op_49789_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_49789_strides_0 = const()[name = tensor<string, []>("op_49789_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_49789_pad_0 = const()[name = tensor<string, []>("op_49789_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_49789_dilations_0 = const()[name = tensor<string, []>("op_49789_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_49789_groups_0 = const()[name = tensor<string, []>("op_49789_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [4915200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(619422720))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(624337984))), name = tensor<string, []>("up_blocks_0_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_0_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_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, []>(624338176)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_49789_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_49789_dilations_0, groups = var_49789_groups_0, pad = var_49789_pad_0, pad_type = var_49789_pad_type_0, strides = var_49789_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_641_cast_fp16)[name = tensor<string, []>("op_49789_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_331_cast_fp16 = add(x = var_49789_cast_fp16, y = inputs_329_cast_fp16)[name = tensor<string, []>("inputs_331_cast_fp16")];
            tensor<int32, [1]> hidden_states_441_axes_0 = const()[name = tensor<string, []>("hidden_states_441_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_441_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_441_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(624340800)))];
            tensor<fp16, [1280]> hidden_states_441_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_441_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(624343424)))];
            tensor<fp16, []> var_49805_to_fp16 = const()[name = tensor<string, []>("op_49805_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_441_cast_fp16 = layer_norm(axes = hidden_states_441_axes_0, beta = hidden_states_441_beta_0_to_fp16, epsilon = var_49805_to_fp16, gamma = hidden_states_441_gamma_0_to_fp16, x = inputs_331_cast_fp16)[name = tensor<string, []>("hidden_states_441_cast_fp16")];
            tensor<string, []> q_221_pad_type_0 = const()[name = tensor<string, []>("q_221_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_221_strides_0 = const()[name = tensor<string, []>("q_221_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_221_pad_0 = const()[name = tensor<string, []>("q_221_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_221_dilations_0 = const()[name = tensor<string, []>("q_221_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_221_groups_0 = const()[name = tensor<string, []>("q_221_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_1_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(624346048))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(625574912))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_1_attn1_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_221_cast_fp16 = conv(dilations = q_221_dilations_0, groups = q_221_groups_0, pad = q_221_pad_0, pad_type = q_221_pad_type_0, strides = q_221_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_1_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_441_cast_fp16)[name = tensor<string, []>("q_221_cast_fp16")];
            tensor<string, []> k_441_pad_type_0 = const()[name = tensor<string, []>("k_441_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_441_strides_0 = const()[name = tensor<string, []>("k_441_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_441_pad_0 = const()[name = tensor<string, []>("k_441_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_441_dilations_0 = const()[name = tensor<string, []>("k_441_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_441_groups_0 = const()[name = tensor<string, []>("k_441_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_1_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(625575104))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(626803968))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_1_attn1_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> k_441_cast_fp16 = conv(dilations = k_441_dilations_0, groups = k_441_groups_0, pad = k_441_pad_0, pad_type = k_441_pad_type_0, strides = k_441_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_1_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_441_cast_fp16)[name = tensor<string, []>("k_441_cast_fp16")];
            tensor<string, []> v_221_pad_type_0 = const()[name = tensor<string, []>("v_221_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_221_strides_0 = const()[name = tensor<string, []>("v_221_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_221_pad_0 = const()[name = tensor<string, []>("v_221_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_221_dilations_0 = const()[name = tensor<string, []>("v_221_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_221_groups_0 = const()[name = tensor<string, []>("v_221_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_1_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(626804160))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(628033024))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_1_attn1_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> v_221_cast_fp16 = conv(dilations = v_221_dilations_0, groups = v_221_groups_0, pad = v_221_pad_0, pad_type = v_221_pad_type_0, strides = v_221_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_1_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_441_cast_fp16)[name = tensor<string, []>("v_221_cast_fp16")];
            tensor<int32, [4]> var_49838_begin_0 = const()[name = tensor<string, []>("op_49838_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_49838_end_0 = const()[name = tensor<string, []>("op_49838_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_49838_end_mask_0 = const()[name = tensor<string, []>("op_49838_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49838_cast_fp16 = slice_by_index(begin = var_49838_begin_0, end = var_49838_end_0, end_mask = var_49838_end_mask_0, x = q_221_cast_fp16)[name = tensor<string, []>("op_49838_cast_fp16")];
            tensor<int32, [4]> var_49842_begin_0 = const()[name = tensor<string, []>("op_49842_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_49842_end_0 = const()[name = tensor<string, []>("op_49842_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_49842_end_mask_0 = const()[name = tensor<string, []>("op_49842_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49842_cast_fp16 = slice_by_index(begin = var_49842_begin_0, end = var_49842_end_0, end_mask = var_49842_end_mask_0, x = q_221_cast_fp16)[name = tensor<string, []>("op_49842_cast_fp16")];
            tensor<int32, [4]> var_49846_begin_0 = const()[name = tensor<string, []>("op_49846_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_49846_end_0 = const()[name = tensor<string, []>("op_49846_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_49846_end_mask_0 = const()[name = tensor<string, []>("op_49846_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49846_cast_fp16 = slice_by_index(begin = var_49846_begin_0, end = var_49846_end_0, end_mask = var_49846_end_mask_0, x = q_221_cast_fp16)[name = tensor<string, []>("op_49846_cast_fp16")];
            tensor<int32, [4]> var_49850_begin_0 = const()[name = tensor<string, []>("op_49850_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_49850_end_0 = const()[name = tensor<string, []>("op_49850_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_49850_end_mask_0 = const()[name = tensor<string, []>("op_49850_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49850_cast_fp16 = slice_by_index(begin = var_49850_begin_0, end = var_49850_end_0, end_mask = var_49850_end_mask_0, x = q_221_cast_fp16)[name = tensor<string, []>("op_49850_cast_fp16")];
            tensor<int32, [4]> var_49854_begin_0 = const()[name = tensor<string, []>("op_49854_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_49854_end_0 = const()[name = tensor<string, []>("op_49854_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_49854_end_mask_0 = const()[name = tensor<string, []>("op_49854_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49854_cast_fp16 = slice_by_index(begin = var_49854_begin_0, end = var_49854_end_0, end_mask = var_49854_end_mask_0, x = q_221_cast_fp16)[name = tensor<string, []>("op_49854_cast_fp16")];
            tensor<int32, [4]> var_49858_begin_0 = const()[name = tensor<string, []>("op_49858_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_49858_end_0 = const()[name = tensor<string, []>("op_49858_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_49858_end_mask_0 = const()[name = tensor<string, []>("op_49858_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49858_cast_fp16 = slice_by_index(begin = var_49858_begin_0, end = var_49858_end_0, end_mask = var_49858_end_mask_0, x = q_221_cast_fp16)[name = tensor<string, []>("op_49858_cast_fp16")];
            tensor<int32, [4]> var_49862_begin_0 = const()[name = tensor<string, []>("op_49862_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_49862_end_0 = const()[name = tensor<string, []>("op_49862_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_49862_end_mask_0 = const()[name = tensor<string, []>("op_49862_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49862_cast_fp16 = slice_by_index(begin = var_49862_begin_0, end = var_49862_end_0, end_mask = var_49862_end_mask_0, x = q_221_cast_fp16)[name = tensor<string, []>("op_49862_cast_fp16")];
            tensor<int32, [4]> var_49866_begin_0 = const()[name = tensor<string, []>("op_49866_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_49866_end_0 = const()[name = tensor<string, []>("op_49866_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_49866_end_mask_0 = const()[name = tensor<string, []>("op_49866_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49866_cast_fp16 = slice_by_index(begin = var_49866_begin_0, end = var_49866_end_0, end_mask = var_49866_end_mask_0, x = q_221_cast_fp16)[name = tensor<string, []>("op_49866_cast_fp16")];
            tensor<int32, [4]> var_49870_begin_0 = const()[name = tensor<string, []>("op_49870_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_49870_end_0 = const()[name = tensor<string, []>("op_49870_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_49870_end_mask_0 = const()[name = tensor<string, []>("op_49870_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49870_cast_fp16 = slice_by_index(begin = var_49870_begin_0, end = var_49870_end_0, end_mask = var_49870_end_mask_0, x = q_221_cast_fp16)[name = tensor<string, []>("op_49870_cast_fp16")];
            tensor<int32, [4]> var_49874_begin_0 = const()[name = tensor<string, []>("op_49874_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_49874_end_0 = const()[name = tensor<string, []>("op_49874_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_49874_end_mask_0 = const()[name = tensor<string, []>("op_49874_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49874_cast_fp16 = slice_by_index(begin = var_49874_begin_0, end = var_49874_end_0, end_mask = var_49874_end_mask_0, x = q_221_cast_fp16)[name = tensor<string, []>("op_49874_cast_fp16")];
            tensor<int32, [4]> var_49878_begin_0 = const()[name = tensor<string, []>("op_49878_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_49878_end_0 = const()[name = tensor<string, []>("op_49878_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_49878_end_mask_0 = const()[name = tensor<string, []>("op_49878_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49878_cast_fp16 = slice_by_index(begin = var_49878_begin_0, end = var_49878_end_0, end_mask = var_49878_end_mask_0, x = q_221_cast_fp16)[name = tensor<string, []>("op_49878_cast_fp16")];
            tensor<int32, [4]> var_49882_begin_0 = const()[name = tensor<string, []>("op_49882_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_49882_end_0 = const()[name = tensor<string, []>("op_49882_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_49882_end_mask_0 = const()[name = tensor<string, []>("op_49882_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49882_cast_fp16 = slice_by_index(begin = var_49882_begin_0, end = var_49882_end_0, end_mask = var_49882_end_mask_0, x = q_221_cast_fp16)[name = tensor<string, []>("op_49882_cast_fp16")];
            tensor<int32, [4]> var_49886_begin_0 = const()[name = tensor<string, []>("op_49886_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_49886_end_0 = const()[name = tensor<string, []>("op_49886_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_49886_end_mask_0 = const()[name = tensor<string, []>("op_49886_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49886_cast_fp16 = slice_by_index(begin = var_49886_begin_0, end = var_49886_end_0, end_mask = var_49886_end_mask_0, x = q_221_cast_fp16)[name = tensor<string, []>("op_49886_cast_fp16")];
            tensor<int32, [4]> var_49890_begin_0 = const()[name = tensor<string, []>("op_49890_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_49890_end_0 = const()[name = tensor<string, []>("op_49890_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_49890_end_mask_0 = const()[name = tensor<string, []>("op_49890_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49890_cast_fp16 = slice_by_index(begin = var_49890_begin_0, end = var_49890_end_0, end_mask = var_49890_end_mask_0, x = q_221_cast_fp16)[name = tensor<string, []>("op_49890_cast_fp16")];
            tensor<int32, [4]> var_49894_begin_0 = const()[name = tensor<string, []>("op_49894_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_49894_end_0 = const()[name = tensor<string, []>("op_49894_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_49894_end_mask_0 = const()[name = tensor<string, []>("op_49894_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49894_cast_fp16 = slice_by_index(begin = var_49894_begin_0, end = var_49894_end_0, end_mask = var_49894_end_mask_0, x = q_221_cast_fp16)[name = tensor<string, []>("op_49894_cast_fp16")];
            tensor<int32, [4]> var_49898_begin_0 = const()[name = tensor<string, []>("op_49898_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_49898_end_0 = const()[name = tensor<string, []>("op_49898_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_49898_end_mask_0 = const()[name = tensor<string, []>("op_49898_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49898_cast_fp16 = slice_by_index(begin = var_49898_begin_0, end = var_49898_end_0, end_mask = var_49898_end_mask_0, x = q_221_cast_fp16)[name = tensor<string, []>("op_49898_cast_fp16")];
            tensor<int32, [4]> var_49902_begin_0 = const()[name = tensor<string, []>("op_49902_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_49902_end_0 = const()[name = tensor<string, []>("op_49902_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_49902_end_mask_0 = const()[name = tensor<string, []>("op_49902_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49902_cast_fp16 = slice_by_index(begin = var_49902_begin_0, end = var_49902_end_0, end_mask = var_49902_end_mask_0, x = q_221_cast_fp16)[name = tensor<string, []>("op_49902_cast_fp16")];
            tensor<int32, [4]> var_49906_begin_0 = const()[name = tensor<string, []>("op_49906_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_49906_end_0 = const()[name = tensor<string, []>("op_49906_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_49906_end_mask_0 = const()[name = tensor<string, []>("op_49906_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49906_cast_fp16 = slice_by_index(begin = var_49906_begin_0, end = var_49906_end_0, end_mask = var_49906_end_mask_0, x = q_221_cast_fp16)[name = tensor<string, []>("op_49906_cast_fp16")];
            tensor<int32, [4]> var_49910_begin_0 = const()[name = tensor<string, []>("op_49910_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_49910_end_0 = const()[name = tensor<string, []>("op_49910_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_49910_end_mask_0 = const()[name = tensor<string, []>("op_49910_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49910_cast_fp16 = slice_by_index(begin = var_49910_begin_0, end = var_49910_end_0, end_mask = var_49910_end_mask_0, x = q_221_cast_fp16)[name = tensor<string, []>("op_49910_cast_fp16")];
            tensor<int32, [4]> var_49914_begin_0 = const()[name = tensor<string, []>("op_49914_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_49914_end_0 = const()[name = tensor<string, []>("op_49914_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_49914_end_mask_0 = const()[name = tensor<string, []>("op_49914_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49914_cast_fp16 = slice_by_index(begin = var_49914_begin_0, end = var_49914_end_0, end_mask = var_49914_end_mask_0, x = q_221_cast_fp16)[name = tensor<string, []>("op_49914_cast_fp16")];
            tensor<int32, [4]> k_443_perm_0 = const()[name = tensor<string, []>("k_443_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_49921_begin_0 = const()[name = tensor<string, []>("op_49921_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_49921_end_0 = const()[name = tensor<string, []>("op_49921_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_49921_end_mask_0 = const()[name = tensor<string, []>("op_49921_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> k_443_cast_fp16 = transpose(perm = k_443_perm_0, x = k_441_cast_fp16)[name = tensor<string, []>("transpose_29")];
            tensor<fp16, [2, 1024, 1, 64]> var_49921_cast_fp16 = slice_by_index(begin = var_49921_begin_0, end = var_49921_end_0, end_mask = var_49921_end_mask_0, x = k_443_cast_fp16)[name = tensor<string, []>("op_49921_cast_fp16")];
            tensor<int32, [4]> var_49925_begin_0 = const()[name = tensor<string, []>("op_49925_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_49925_end_0 = const()[name = tensor<string, []>("op_49925_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_49925_end_mask_0 = const()[name = tensor<string, []>("op_49925_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49925_cast_fp16 = slice_by_index(begin = var_49925_begin_0, end = var_49925_end_0, end_mask = var_49925_end_mask_0, x = k_443_cast_fp16)[name = tensor<string, []>("op_49925_cast_fp16")];
            tensor<int32, [4]> var_49929_begin_0 = const()[name = tensor<string, []>("op_49929_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_49929_end_0 = const()[name = tensor<string, []>("op_49929_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_49929_end_mask_0 = const()[name = tensor<string, []>("op_49929_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49929_cast_fp16 = slice_by_index(begin = var_49929_begin_0, end = var_49929_end_0, end_mask = var_49929_end_mask_0, x = k_443_cast_fp16)[name = tensor<string, []>("op_49929_cast_fp16")];
            tensor<int32, [4]> var_49933_begin_0 = const()[name = tensor<string, []>("op_49933_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_49933_end_0 = const()[name = tensor<string, []>("op_49933_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_49933_end_mask_0 = const()[name = tensor<string, []>("op_49933_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49933_cast_fp16 = slice_by_index(begin = var_49933_begin_0, end = var_49933_end_0, end_mask = var_49933_end_mask_0, x = k_443_cast_fp16)[name = tensor<string, []>("op_49933_cast_fp16")];
            tensor<int32, [4]> var_49937_begin_0 = const()[name = tensor<string, []>("op_49937_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_49937_end_0 = const()[name = tensor<string, []>("op_49937_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_49937_end_mask_0 = const()[name = tensor<string, []>("op_49937_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49937_cast_fp16 = slice_by_index(begin = var_49937_begin_0, end = var_49937_end_0, end_mask = var_49937_end_mask_0, x = k_443_cast_fp16)[name = tensor<string, []>("op_49937_cast_fp16")];
            tensor<int32, [4]> var_49941_begin_0 = const()[name = tensor<string, []>("op_49941_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_49941_end_0 = const()[name = tensor<string, []>("op_49941_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_49941_end_mask_0 = const()[name = tensor<string, []>("op_49941_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49941_cast_fp16 = slice_by_index(begin = var_49941_begin_0, end = var_49941_end_0, end_mask = var_49941_end_mask_0, x = k_443_cast_fp16)[name = tensor<string, []>("op_49941_cast_fp16")];
            tensor<int32, [4]> var_49945_begin_0 = const()[name = tensor<string, []>("op_49945_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_49945_end_0 = const()[name = tensor<string, []>("op_49945_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_49945_end_mask_0 = const()[name = tensor<string, []>("op_49945_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49945_cast_fp16 = slice_by_index(begin = var_49945_begin_0, end = var_49945_end_0, end_mask = var_49945_end_mask_0, x = k_443_cast_fp16)[name = tensor<string, []>("op_49945_cast_fp16")];
            tensor<int32, [4]> var_49949_begin_0 = const()[name = tensor<string, []>("op_49949_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_49949_end_0 = const()[name = tensor<string, []>("op_49949_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_49949_end_mask_0 = const()[name = tensor<string, []>("op_49949_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49949_cast_fp16 = slice_by_index(begin = var_49949_begin_0, end = var_49949_end_0, end_mask = var_49949_end_mask_0, x = k_443_cast_fp16)[name = tensor<string, []>("op_49949_cast_fp16")];
            tensor<int32, [4]> var_49953_begin_0 = const()[name = tensor<string, []>("op_49953_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_49953_end_0 = const()[name = tensor<string, []>("op_49953_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_49953_end_mask_0 = const()[name = tensor<string, []>("op_49953_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49953_cast_fp16 = slice_by_index(begin = var_49953_begin_0, end = var_49953_end_0, end_mask = var_49953_end_mask_0, x = k_443_cast_fp16)[name = tensor<string, []>("op_49953_cast_fp16")];
            tensor<int32, [4]> var_49957_begin_0 = const()[name = tensor<string, []>("op_49957_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_49957_end_0 = const()[name = tensor<string, []>("op_49957_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_49957_end_mask_0 = const()[name = tensor<string, []>("op_49957_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49957_cast_fp16 = slice_by_index(begin = var_49957_begin_0, end = var_49957_end_0, end_mask = var_49957_end_mask_0, x = k_443_cast_fp16)[name = tensor<string, []>("op_49957_cast_fp16")];
            tensor<int32, [4]> var_49961_begin_0 = const()[name = tensor<string, []>("op_49961_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_49961_end_0 = const()[name = tensor<string, []>("op_49961_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_49961_end_mask_0 = const()[name = tensor<string, []>("op_49961_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49961_cast_fp16 = slice_by_index(begin = var_49961_begin_0, end = var_49961_end_0, end_mask = var_49961_end_mask_0, x = k_443_cast_fp16)[name = tensor<string, []>("op_49961_cast_fp16")];
            tensor<int32, [4]> var_49965_begin_0 = const()[name = tensor<string, []>("op_49965_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_49965_end_0 = const()[name = tensor<string, []>("op_49965_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_49965_end_mask_0 = const()[name = tensor<string, []>("op_49965_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49965_cast_fp16 = slice_by_index(begin = var_49965_begin_0, end = var_49965_end_0, end_mask = var_49965_end_mask_0, x = k_443_cast_fp16)[name = tensor<string, []>("op_49965_cast_fp16")];
            tensor<int32, [4]> var_49969_begin_0 = const()[name = tensor<string, []>("op_49969_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_49969_end_0 = const()[name = tensor<string, []>("op_49969_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_49969_end_mask_0 = const()[name = tensor<string, []>("op_49969_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49969_cast_fp16 = slice_by_index(begin = var_49969_begin_0, end = var_49969_end_0, end_mask = var_49969_end_mask_0, x = k_443_cast_fp16)[name = tensor<string, []>("op_49969_cast_fp16")];
            tensor<int32, [4]> var_49973_begin_0 = const()[name = tensor<string, []>("op_49973_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_49973_end_0 = const()[name = tensor<string, []>("op_49973_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_49973_end_mask_0 = const()[name = tensor<string, []>("op_49973_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49973_cast_fp16 = slice_by_index(begin = var_49973_begin_0, end = var_49973_end_0, end_mask = var_49973_end_mask_0, x = k_443_cast_fp16)[name = tensor<string, []>("op_49973_cast_fp16")];
            tensor<int32, [4]> var_49977_begin_0 = const()[name = tensor<string, []>("op_49977_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_49977_end_0 = const()[name = tensor<string, []>("op_49977_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_49977_end_mask_0 = const()[name = tensor<string, []>("op_49977_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49977_cast_fp16 = slice_by_index(begin = var_49977_begin_0, end = var_49977_end_0, end_mask = var_49977_end_mask_0, x = k_443_cast_fp16)[name = tensor<string, []>("op_49977_cast_fp16")];
            tensor<int32, [4]> var_49981_begin_0 = const()[name = tensor<string, []>("op_49981_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_49981_end_0 = const()[name = tensor<string, []>("op_49981_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_49981_end_mask_0 = const()[name = tensor<string, []>("op_49981_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49981_cast_fp16 = slice_by_index(begin = var_49981_begin_0, end = var_49981_end_0, end_mask = var_49981_end_mask_0, x = k_443_cast_fp16)[name = tensor<string, []>("op_49981_cast_fp16")];
            tensor<int32, [4]> var_49985_begin_0 = const()[name = tensor<string, []>("op_49985_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_49985_end_0 = const()[name = tensor<string, []>("op_49985_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_49985_end_mask_0 = const()[name = tensor<string, []>("op_49985_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49985_cast_fp16 = slice_by_index(begin = var_49985_begin_0, end = var_49985_end_0, end_mask = var_49985_end_mask_0, x = k_443_cast_fp16)[name = tensor<string, []>("op_49985_cast_fp16")];
            tensor<int32, [4]> var_49989_begin_0 = const()[name = tensor<string, []>("op_49989_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_49989_end_0 = const()[name = tensor<string, []>("op_49989_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_49989_end_mask_0 = const()[name = tensor<string, []>("op_49989_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49989_cast_fp16 = slice_by_index(begin = var_49989_begin_0, end = var_49989_end_0, end_mask = var_49989_end_mask_0, x = k_443_cast_fp16)[name = tensor<string, []>("op_49989_cast_fp16")];
            tensor<int32, [4]> var_49993_begin_0 = const()[name = tensor<string, []>("op_49993_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_49993_end_0 = const()[name = tensor<string, []>("op_49993_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_49993_end_mask_0 = const()[name = tensor<string, []>("op_49993_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49993_cast_fp16 = slice_by_index(begin = var_49993_begin_0, end = var_49993_end_0, end_mask = var_49993_end_mask_0, x = k_443_cast_fp16)[name = tensor<string, []>("op_49993_cast_fp16")];
            tensor<int32, [4]> var_49997_begin_0 = const()[name = tensor<string, []>("op_49997_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_49997_end_0 = const()[name = tensor<string, []>("op_49997_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_49997_end_mask_0 = const()[name = tensor<string, []>("op_49997_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49997_cast_fp16 = slice_by_index(begin = var_49997_begin_0, end = var_49997_end_0, end_mask = var_49997_end_mask_0, x = k_443_cast_fp16)[name = tensor<string, []>("op_49997_cast_fp16")];
            tensor<int32, [4]> var_49999_begin_0 = const()[name = tensor<string, []>("op_49999_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_49999_end_0 = const()[name = tensor<string, []>("op_49999_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_49999_end_mask_0 = const()[name = tensor<string, []>("op_49999_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49999_cast_fp16 = slice_by_index(begin = var_49999_begin_0, end = var_49999_end_0, end_mask = var_49999_end_mask_0, x = v_221_cast_fp16)[name = tensor<string, []>("op_49999_cast_fp16")];
            tensor<int32, [4]> var_50003_begin_0 = const()[name = tensor<string, []>("op_50003_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_50003_end_0 = const()[name = tensor<string, []>("op_50003_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_50003_end_mask_0 = const()[name = tensor<string, []>("op_50003_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50003_cast_fp16 = slice_by_index(begin = var_50003_begin_0, end = var_50003_end_0, end_mask = var_50003_end_mask_0, x = v_221_cast_fp16)[name = tensor<string, []>("op_50003_cast_fp16")];
            tensor<int32, [4]> var_50007_begin_0 = const()[name = tensor<string, []>("op_50007_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_50007_end_0 = const()[name = tensor<string, []>("op_50007_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_50007_end_mask_0 = const()[name = tensor<string, []>("op_50007_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50007_cast_fp16 = slice_by_index(begin = var_50007_begin_0, end = var_50007_end_0, end_mask = var_50007_end_mask_0, x = v_221_cast_fp16)[name = tensor<string, []>("op_50007_cast_fp16")];
            tensor<int32, [4]> var_50011_begin_0 = const()[name = tensor<string, []>("op_50011_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_50011_end_0 = const()[name = tensor<string, []>("op_50011_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_50011_end_mask_0 = const()[name = tensor<string, []>("op_50011_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50011_cast_fp16 = slice_by_index(begin = var_50011_begin_0, end = var_50011_end_0, end_mask = var_50011_end_mask_0, x = v_221_cast_fp16)[name = tensor<string, []>("op_50011_cast_fp16")];
            tensor<int32, [4]> var_50015_begin_0 = const()[name = tensor<string, []>("op_50015_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_50015_end_0 = const()[name = tensor<string, []>("op_50015_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_50015_end_mask_0 = const()[name = tensor<string, []>("op_50015_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50015_cast_fp16 = slice_by_index(begin = var_50015_begin_0, end = var_50015_end_0, end_mask = var_50015_end_mask_0, x = v_221_cast_fp16)[name = tensor<string, []>("op_50015_cast_fp16")];
            tensor<int32, [4]> var_50019_begin_0 = const()[name = tensor<string, []>("op_50019_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_50019_end_0 = const()[name = tensor<string, []>("op_50019_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_50019_end_mask_0 = const()[name = tensor<string, []>("op_50019_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50019_cast_fp16 = slice_by_index(begin = var_50019_begin_0, end = var_50019_end_0, end_mask = var_50019_end_mask_0, x = v_221_cast_fp16)[name = tensor<string, []>("op_50019_cast_fp16")];
            tensor<int32, [4]> var_50023_begin_0 = const()[name = tensor<string, []>("op_50023_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_50023_end_0 = const()[name = tensor<string, []>("op_50023_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_50023_end_mask_0 = const()[name = tensor<string, []>("op_50023_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50023_cast_fp16 = slice_by_index(begin = var_50023_begin_0, end = var_50023_end_0, end_mask = var_50023_end_mask_0, x = v_221_cast_fp16)[name = tensor<string, []>("op_50023_cast_fp16")];
            tensor<int32, [4]> var_50027_begin_0 = const()[name = tensor<string, []>("op_50027_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_50027_end_0 = const()[name = tensor<string, []>("op_50027_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_50027_end_mask_0 = const()[name = tensor<string, []>("op_50027_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50027_cast_fp16 = slice_by_index(begin = var_50027_begin_0, end = var_50027_end_0, end_mask = var_50027_end_mask_0, x = v_221_cast_fp16)[name = tensor<string, []>("op_50027_cast_fp16")];
            tensor<int32, [4]> var_50031_begin_0 = const()[name = tensor<string, []>("op_50031_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_50031_end_0 = const()[name = tensor<string, []>("op_50031_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_50031_end_mask_0 = const()[name = tensor<string, []>("op_50031_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50031_cast_fp16 = slice_by_index(begin = var_50031_begin_0, end = var_50031_end_0, end_mask = var_50031_end_mask_0, x = v_221_cast_fp16)[name = tensor<string, []>("op_50031_cast_fp16")];
            tensor<int32, [4]> var_50035_begin_0 = const()[name = tensor<string, []>("op_50035_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_50035_end_0 = const()[name = tensor<string, []>("op_50035_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_50035_end_mask_0 = const()[name = tensor<string, []>("op_50035_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50035_cast_fp16 = slice_by_index(begin = var_50035_begin_0, end = var_50035_end_0, end_mask = var_50035_end_mask_0, x = v_221_cast_fp16)[name = tensor<string, []>("op_50035_cast_fp16")];
            tensor<int32, [4]> var_50039_begin_0 = const()[name = tensor<string, []>("op_50039_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_50039_end_0 = const()[name = tensor<string, []>("op_50039_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_50039_end_mask_0 = const()[name = tensor<string, []>("op_50039_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50039_cast_fp16 = slice_by_index(begin = var_50039_begin_0, end = var_50039_end_0, end_mask = var_50039_end_mask_0, x = v_221_cast_fp16)[name = tensor<string, []>("op_50039_cast_fp16")];
            tensor<int32, [4]> var_50043_begin_0 = const()[name = tensor<string, []>("op_50043_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_50043_end_0 = const()[name = tensor<string, []>("op_50043_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_50043_end_mask_0 = const()[name = tensor<string, []>("op_50043_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50043_cast_fp16 = slice_by_index(begin = var_50043_begin_0, end = var_50043_end_0, end_mask = var_50043_end_mask_0, x = v_221_cast_fp16)[name = tensor<string, []>("op_50043_cast_fp16")];
            tensor<int32, [4]> var_50047_begin_0 = const()[name = tensor<string, []>("op_50047_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_50047_end_0 = const()[name = tensor<string, []>("op_50047_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_50047_end_mask_0 = const()[name = tensor<string, []>("op_50047_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50047_cast_fp16 = slice_by_index(begin = var_50047_begin_0, end = var_50047_end_0, end_mask = var_50047_end_mask_0, x = v_221_cast_fp16)[name = tensor<string, []>("op_50047_cast_fp16")];
            tensor<int32, [4]> var_50051_begin_0 = const()[name = tensor<string, []>("op_50051_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_50051_end_0 = const()[name = tensor<string, []>("op_50051_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_50051_end_mask_0 = const()[name = tensor<string, []>("op_50051_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50051_cast_fp16 = slice_by_index(begin = var_50051_begin_0, end = var_50051_end_0, end_mask = var_50051_end_mask_0, x = v_221_cast_fp16)[name = tensor<string, []>("op_50051_cast_fp16")];
            tensor<int32, [4]> var_50055_begin_0 = const()[name = tensor<string, []>("op_50055_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_50055_end_0 = const()[name = tensor<string, []>("op_50055_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_50055_end_mask_0 = const()[name = tensor<string, []>("op_50055_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50055_cast_fp16 = slice_by_index(begin = var_50055_begin_0, end = var_50055_end_0, end_mask = var_50055_end_mask_0, x = v_221_cast_fp16)[name = tensor<string, []>("op_50055_cast_fp16")];
            tensor<int32, [4]> var_50059_begin_0 = const()[name = tensor<string, []>("op_50059_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_50059_end_0 = const()[name = tensor<string, []>("op_50059_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_50059_end_mask_0 = const()[name = tensor<string, []>("op_50059_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50059_cast_fp16 = slice_by_index(begin = var_50059_begin_0, end = var_50059_end_0, end_mask = var_50059_end_mask_0, x = v_221_cast_fp16)[name = tensor<string, []>("op_50059_cast_fp16")];
            tensor<int32, [4]> var_50063_begin_0 = const()[name = tensor<string, []>("op_50063_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_50063_end_0 = const()[name = tensor<string, []>("op_50063_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_50063_end_mask_0 = const()[name = tensor<string, []>("op_50063_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50063_cast_fp16 = slice_by_index(begin = var_50063_begin_0, end = var_50063_end_0, end_mask = var_50063_end_mask_0, x = v_221_cast_fp16)[name = tensor<string, []>("op_50063_cast_fp16")];
            tensor<int32, [4]> var_50067_begin_0 = const()[name = tensor<string, []>("op_50067_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_50067_end_0 = const()[name = tensor<string, []>("op_50067_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_50067_end_mask_0 = const()[name = tensor<string, []>("op_50067_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50067_cast_fp16 = slice_by_index(begin = var_50067_begin_0, end = var_50067_end_0, end_mask = var_50067_end_mask_0, x = v_221_cast_fp16)[name = tensor<string, []>("op_50067_cast_fp16")];
            tensor<int32, [4]> var_50071_begin_0 = const()[name = tensor<string, []>("op_50071_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_50071_end_0 = const()[name = tensor<string, []>("op_50071_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_50071_end_mask_0 = const()[name = tensor<string, []>("op_50071_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50071_cast_fp16 = slice_by_index(begin = var_50071_begin_0, end = var_50071_end_0, end_mask = var_50071_end_mask_0, x = v_221_cast_fp16)[name = tensor<string, []>("op_50071_cast_fp16")];
            tensor<int32, [4]> var_50075_begin_0 = const()[name = tensor<string, []>("op_50075_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_50075_end_0 = const()[name = tensor<string, []>("op_50075_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_50075_end_mask_0 = const()[name = tensor<string, []>("op_50075_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50075_cast_fp16 = slice_by_index(begin = var_50075_begin_0, end = var_50075_end_0, end_mask = var_50075_end_mask_0, x = v_221_cast_fp16)[name = tensor<string, []>("op_50075_cast_fp16")];
            tensor<string, []> var_50079_equation_0 = const()[name = tensor<string, []>("op_50079_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50079_cast_fp16 = einsum(equation = var_50079_equation_0, values = (var_49921_cast_fp16, var_49838_cast_fp16))[name = tensor<string, []>("op_50079_cast_fp16")];
            tensor<fp16, []> var_50080_to_fp16 = const()[name = tensor<string, []>("op_50080_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4241_cast_fp16 = mul(x = var_50079_cast_fp16, y = var_50080_to_fp16)[name = tensor<string, []>("aw_4241_cast_fp16")];
            tensor<string, []> var_50083_equation_0 = const()[name = tensor<string, []>("op_50083_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50083_cast_fp16 = einsum(equation = var_50083_equation_0, values = (var_49925_cast_fp16, var_49842_cast_fp16))[name = tensor<string, []>("op_50083_cast_fp16")];
            tensor<fp16, []> var_50084_to_fp16 = const()[name = tensor<string, []>("op_50084_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4243_cast_fp16 = mul(x = var_50083_cast_fp16, y = var_50084_to_fp16)[name = tensor<string, []>("aw_4243_cast_fp16")];
            tensor<string, []> var_50087_equation_0 = const()[name = tensor<string, []>("op_50087_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50087_cast_fp16 = einsum(equation = var_50087_equation_0, values = (var_49929_cast_fp16, var_49846_cast_fp16))[name = tensor<string, []>("op_50087_cast_fp16")];
            tensor<fp16, []> var_50088_to_fp16 = const()[name = tensor<string, []>("op_50088_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4245_cast_fp16 = mul(x = var_50087_cast_fp16, y = var_50088_to_fp16)[name = tensor<string, []>("aw_4245_cast_fp16")];
            tensor<string, []> var_50091_equation_0 = const()[name = tensor<string, []>("op_50091_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50091_cast_fp16 = einsum(equation = var_50091_equation_0, values = (var_49933_cast_fp16, var_49850_cast_fp16))[name = tensor<string, []>("op_50091_cast_fp16")];
            tensor<fp16, []> var_50092_to_fp16 = const()[name = tensor<string, []>("op_50092_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4247_cast_fp16 = mul(x = var_50091_cast_fp16, y = var_50092_to_fp16)[name = tensor<string, []>("aw_4247_cast_fp16")];
            tensor<string, []> var_50095_equation_0 = const()[name = tensor<string, []>("op_50095_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50095_cast_fp16 = einsum(equation = var_50095_equation_0, values = (var_49937_cast_fp16, var_49854_cast_fp16))[name = tensor<string, []>("op_50095_cast_fp16")];
            tensor<fp16, []> var_50096_to_fp16 = const()[name = tensor<string, []>("op_50096_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4249_cast_fp16 = mul(x = var_50095_cast_fp16, y = var_50096_to_fp16)[name = tensor<string, []>("aw_4249_cast_fp16")];
            tensor<string, []> var_50099_equation_0 = const()[name = tensor<string, []>("op_50099_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50099_cast_fp16 = einsum(equation = var_50099_equation_0, values = (var_49941_cast_fp16, var_49858_cast_fp16))[name = tensor<string, []>("op_50099_cast_fp16")];
            tensor<fp16, []> var_50100_to_fp16 = const()[name = tensor<string, []>("op_50100_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4251_cast_fp16 = mul(x = var_50099_cast_fp16, y = var_50100_to_fp16)[name = tensor<string, []>("aw_4251_cast_fp16")];
            tensor<string, []> var_50103_equation_0 = const()[name = tensor<string, []>("op_50103_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50103_cast_fp16 = einsum(equation = var_50103_equation_0, values = (var_49945_cast_fp16, var_49862_cast_fp16))[name = tensor<string, []>("op_50103_cast_fp16")];
            tensor<fp16, []> var_50104_to_fp16 = const()[name = tensor<string, []>("op_50104_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4253_cast_fp16 = mul(x = var_50103_cast_fp16, y = var_50104_to_fp16)[name = tensor<string, []>("aw_4253_cast_fp16")];
            tensor<string, []> var_50107_equation_0 = const()[name = tensor<string, []>("op_50107_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50107_cast_fp16 = einsum(equation = var_50107_equation_0, values = (var_49949_cast_fp16, var_49866_cast_fp16))[name = tensor<string, []>("op_50107_cast_fp16")];
            tensor<fp16, []> var_50108_to_fp16 = const()[name = tensor<string, []>("op_50108_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4255_cast_fp16 = mul(x = var_50107_cast_fp16, y = var_50108_to_fp16)[name = tensor<string, []>("aw_4255_cast_fp16")];
            tensor<string, []> var_50111_equation_0 = const()[name = tensor<string, []>("op_50111_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50111_cast_fp16 = einsum(equation = var_50111_equation_0, values = (var_49953_cast_fp16, var_49870_cast_fp16))[name = tensor<string, []>("op_50111_cast_fp16")];
            tensor<fp16, []> var_50112_to_fp16 = const()[name = tensor<string, []>("op_50112_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4257_cast_fp16 = mul(x = var_50111_cast_fp16, y = var_50112_to_fp16)[name = tensor<string, []>("aw_4257_cast_fp16")];
            tensor<string, []> var_50115_equation_0 = const()[name = tensor<string, []>("op_50115_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50115_cast_fp16 = einsum(equation = var_50115_equation_0, values = (var_49957_cast_fp16, var_49874_cast_fp16))[name = tensor<string, []>("op_50115_cast_fp16")];
            tensor<fp16, []> var_50116_to_fp16 = const()[name = tensor<string, []>("op_50116_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4259_cast_fp16 = mul(x = var_50115_cast_fp16, y = var_50116_to_fp16)[name = tensor<string, []>("aw_4259_cast_fp16")];
            tensor<string, []> var_50119_equation_0 = const()[name = tensor<string, []>("op_50119_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50119_cast_fp16 = einsum(equation = var_50119_equation_0, values = (var_49961_cast_fp16, var_49878_cast_fp16))[name = tensor<string, []>("op_50119_cast_fp16")];
            tensor<fp16, []> var_50120_to_fp16 = const()[name = tensor<string, []>("op_50120_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4261_cast_fp16 = mul(x = var_50119_cast_fp16, y = var_50120_to_fp16)[name = tensor<string, []>("aw_4261_cast_fp16")];
            tensor<string, []> var_50123_equation_0 = const()[name = tensor<string, []>("op_50123_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50123_cast_fp16 = einsum(equation = var_50123_equation_0, values = (var_49965_cast_fp16, var_49882_cast_fp16))[name = tensor<string, []>("op_50123_cast_fp16")];
            tensor<fp16, []> var_50124_to_fp16 = const()[name = tensor<string, []>("op_50124_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4263_cast_fp16 = mul(x = var_50123_cast_fp16, y = var_50124_to_fp16)[name = tensor<string, []>("aw_4263_cast_fp16")];
            tensor<string, []> var_50127_equation_0 = const()[name = tensor<string, []>("op_50127_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50127_cast_fp16 = einsum(equation = var_50127_equation_0, values = (var_49969_cast_fp16, var_49886_cast_fp16))[name = tensor<string, []>("op_50127_cast_fp16")];
            tensor<fp16, []> var_50128_to_fp16 = const()[name = tensor<string, []>("op_50128_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4265_cast_fp16 = mul(x = var_50127_cast_fp16, y = var_50128_to_fp16)[name = tensor<string, []>("aw_4265_cast_fp16")];
            tensor<string, []> var_50131_equation_0 = const()[name = tensor<string, []>("op_50131_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50131_cast_fp16 = einsum(equation = var_50131_equation_0, values = (var_49973_cast_fp16, var_49890_cast_fp16))[name = tensor<string, []>("op_50131_cast_fp16")];
            tensor<fp16, []> var_50132_to_fp16 = const()[name = tensor<string, []>("op_50132_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4267_cast_fp16 = mul(x = var_50131_cast_fp16, y = var_50132_to_fp16)[name = tensor<string, []>("aw_4267_cast_fp16")];
            tensor<string, []> var_50135_equation_0 = const()[name = tensor<string, []>("op_50135_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50135_cast_fp16 = einsum(equation = var_50135_equation_0, values = (var_49977_cast_fp16, var_49894_cast_fp16))[name = tensor<string, []>("op_50135_cast_fp16")];
            tensor<fp16, []> var_50136_to_fp16 = const()[name = tensor<string, []>("op_50136_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4269_cast_fp16 = mul(x = var_50135_cast_fp16, y = var_50136_to_fp16)[name = tensor<string, []>("aw_4269_cast_fp16")];
            tensor<string, []> var_50139_equation_0 = const()[name = tensor<string, []>("op_50139_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50139_cast_fp16 = einsum(equation = var_50139_equation_0, values = (var_49981_cast_fp16, var_49898_cast_fp16))[name = tensor<string, []>("op_50139_cast_fp16")];
            tensor<fp16, []> var_50140_to_fp16 = const()[name = tensor<string, []>("op_50140_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4271_cast_fp16 = mul(x = var_50139_cast_fp16, y = var_50140_to_fp16)[name = tensor<string, []>("aw_4271_cast_fp16")];
            tensor<string, []> var_50143_equation_0 = const()[name = tensor<string, []>("op_50143_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50143_cast_fp16 = einsum(equation = var_50143_equation_0, values = (var_49985_cast_fp16, var_49902_cast_fp16))[name = tensor<string, []>("op_50143_cast_fp16")];
            tensor<fp16, []> var_50144_to_fp16 = const()[name = tensor<string, []>("op_50144_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4273_cast_fp16 = mul(x = var_50143_cast_fp16, y = var_50144_to_fp16)[name = tensor<string, []>("aw_4273_cast_fp16")];
            tensor<string, []> var_50147_equation_0 = const()[name = tensor<string, []>("op_50147_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50147_cast_fp16 = einsum(equation = var_50147_equation_0, values = (var_49989_cast_fp16, var_49906_cast_fp16))[name = tensor<string, []>("op_50147_cast_fp16")];
            tensor<fp16, []> var_50148_to_fp16 = const()[name = tensor<string, []>("op_50148_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4275_cast_fp16 = mul(x = var_50147_cast_fp16, y = var_50148_to_fp16)[name = tensor<string, []>("aw_4275_cast_fp16")];
            tensor<string, []> var_50151_equation_0 = const()[name = tensor<string, []>("op_50151_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50151_cast_fp16 = einsum(equation = var_50151_equation_0, values = (var_49993_cast_fp16, var_49910_cast_fp16))[name = tensor<string, []>("op_50151_cast_fp16")];
            tensor<fp16, []> var_50152_to_fp16 = const()[name = tensor<string, []>("op_50152_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4277_cast_fp16 = mul(x = var_50151_cast_fp16, y = var_50152_to_fp16)[name = tensor<string, []>("aw_4277_cast_fp16")];
            tensor<string, []> var_50155_equation_0 = const()[name = tensor<string, []>("op_50155_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50155_cast_fp16 = einsum(equation = var_50155_equation_0, values = (var_49997_cast_fp16, var_49914_cast_fp16))[name = tensor<string, []>("op_50155_cast_fp16")];
            tensor<fp16, []> var_50156_to_fp16 = const()[name = tensor<string, []>("op_50156_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4279_cast_fp16 = mul(x = var_50155_cast_fp16, y = var_50156_to_fp16)[name = tensor<string, []>("aw_4279_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50158_cast_fp16 = softmax(axis = var_30355, x = aw_4241_cast_fp16)[name = tensor<string, []>("op_50158_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50159_cast_fp16 = softmax(axis = var_30355, x = aw_4243_cast_fp16)[name = tensor<string, []>("op_50159_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50160_cast_fp16 = softmax(axis = var_30355, x = aw_4245_cast_fp16)[name = tensor<string, []>("op_50160_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50161_cast_fp16 = softmax(axis = var_30355, x = aw_4247_cast_fp16)[name = tensor<string, []>("op_50161_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50162_cast_fp16 = softmax(axis = var_30355, x = aw_4249_cast_fp16)[name = tensor<string, []>("op_50162_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50163_cast_fp16 = softmax(axis = var_30355, x = aw_4251_cast_fp16)[name = tensor<string, []>("op_50163_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50164_cast_fp16 = softmax(axis = var_30355, x = aw_4253_cast_fp16)[name = tensor<string, []>("op_50164_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50165_cast_fp16 = softmax(axis = var_30355, x = aw_4255_cast_fp16)[name = tensor<string, []>("op_50165_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50166_cast_fp16 = softmax(axis = var_30355, x = aw_4257_cast_fp16)[name = tensor<string, []>("op_50166_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50167_cast_fp16 = softmax(axis = var_30355, x = aw_4259_cast_fp16)[name = tensor<string, []>("op_50167_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50168_cast_fp16 = softmax(axis = var_30355, x = aw_4261_cast_fp16)[name = tensor<string, []>("op_50168_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50169_cast_fp16 = softmax(axis = var_30355, x = aw_4263_cast_fp16)[name = tensor<string, []>("op_50169_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50170_cast_fp16 = softmax(axis = var_30355, x = aw_4265_cast_fp16)[name = tensor<string, []>("op_50170_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50171_cast_fp16 = softmax(axis = var_30355, x = aw_4267_cast_fp16)[name = tensor<string, []>("op_50171_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50172_cast_fp16 = softmax(axis = var_30355, x = aw_4269_cast_fp16)[name = tensor<string, []>("op_50172_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50173_cast_fp16 = softmax(axis = var_30355, x = aw_4271_cast_fp16)[name = tensor<string, []>("op_50173_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50174_cast_fp16 = softmax(axis = var_30355, x = aw_4273_cast_fp16)[name = tensor<string, []>("op_50174_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50175_cast_fp16 = softmax(axis = var_30355, x = aw_4275_cast_fp16)[name = tensor<string, []>("op_50175_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50176_cast_fp16 = softmax(axis = var_30355, x = aw_4277_cast_fp16)[name = tensor<string, []>("op_50176_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50177_cast_fp16 = softmax(axis = var_30355, x = aw_4279_cast_fp16)[name = tensor<string, []>("op_50177_cast_fp16")];
            tensor<string, []> var_50179_equation_0 = const()[name = tensor<string, []>("op_50179_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50179_cast_fp16 = einsum(equation = var_50179_equation_0, values = (var_49999_cast_fp16, var_50158_cast_fp16))[name = tensor<string, []>("op_50179_cast_fp16")];
            tensor<string, []> var_50181_equation_0 = const()[name = tensor<string, []>("op_50181_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50181_cast_fp16 = einsum(equation = var_50181_equation_0, values = (var_50003_cast_fp16, var_50159_cast_fp16))[name = tensor<string, []>("op_50181_cast_fp16")];
            tensor<string, []> var_50183_equation_0 = const()[name = tensor<string, []>("op_50183_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50183_cast_fp16 = einsum(equation = var_50183_equation_0, values = (var_50007_cast_fp16, var_50160_cast_fp16))[name = tensor<string, []>("op_50183_cast_fp16")];
            tensor<string, []> var_50185_equation_0 = const()[name = tensor<string, []>("op_50185_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50185_cast_fp16 = einsum(equation = var_50185_equation_0, values = (var_50011_cast_fp16, var_50161_cast_fp16))[name = tensor<string, []>("op_50185_cast_fp16")];
            tensor<string, []> var_50187_equation_0 = const()[name = tensor<string, []>("op_50187_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50187_cast_fp16 = einsum(equation = var_50187_equation_0, values = (var_50015_cast_fp16, var_50162_cast_fp16))[name = tensor<string, []>("op_50187_cast_fp16")];
            tensor<string, []> var_50189_equation_0 = const()[name = tensor<string, []>("op_50189_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50189_cast_fp16 = einsum(equation = var_50189_equation_0, values = (var_50019_cast_fp16, var_50163_cast_fp16))[name = tensor<string, []>("op_50189_cast_fp16")];
            tensor<string, []> var_50191_equation_0 = const()[name = tensor<string, []>("op_50191_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50191_cast_fp16 = einsum(equation = var_50191_equation_0, values = (var_50023_cast_fp16, var_50164_cast_fp16))[name = tensor<string, []>("op_50191_cast_fp16")];
            tensor<string, []> var_50193_equation_0 = const()[name = tensor<string, []>("op_50193_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50193_cast_fp16 = einsum(equation = var_50193_equation_0, values = (var_50027_cast_fp16, var_50165_cast_fp16))[name = tensor<string, []>("op_50193_cast_fp16")];
            tensor<string, []> var_50195_equation_0 = const()[name = tensor<string, []>("op_50195_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50195_cast_fp16 = einsum(equation = var_50195_equation_0, values = (var_50031_cast_fp16, var_50166_cast_fp16))[name = tensor<string, []>("op_50195_cast_fp16")];
            tensor<string, []> var_50197_equation_0 = const()[name = tensor<string, []>("op_50197_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50197_cast_fp16 = einsum(equation = var_50197_equation_0, values = (var_50035_cast_fp16, var_50167_cast_fp16))[name = tensor<string, []>("op_50197_cast_fp16")];
            tensor<string, []> var_50199_equation_0 = const()[name = tensor<string, []>("op_50199_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50199_cast_fp16 = einsum(equation = var_50199_equation_0, values = (var_50039_cast_fp16, var_50168_cast_fp16))[name = tensor<string, []>("op_50199_cast_fp16")];
            tensor<string, []> var_50201_equation_0 = const()[name = tensor<string, []>("op_50201_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50201_cast_fp16 = einsum(equation = var_50201_equation_0, values = (var_50043_cast_fp16, var_50169_cast_fp16))[name = tensor<string, []>("op_50201_cast_fp16")];
            tensor<string, []> var_50203_equation_0 = const()[name = tensor<string, []>("op_50203_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50203_cast_fp16 = einsum(equation = var_50203_equation_0, values = (var_50047_cast_fp16, var_50170_cast_fp16))[name = tensor<string, []>("op_50203_cast_fp16")];
            tensor<string, []> var_50205_equation_0 = const()[name = tensor<string, []>("op_50205_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50205_cast_fp16 = einsum(equation = var_50205_equation_0, values = (var_50051_cast_fp16, var_50171_cast_fp16))[name = tensor<string, []>("op_50205_cast_fp16")];
            tensor<string, []> var_50207_equation_0 = const()[name = tensor<string, []>("op_50207_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50207_cast_fp16 = einsum(equation = var_50207_equation_0, values = (var_50055_cast_fp16, var_50172_cast_fp16))[name = tensor<string, []>("op_50207_cast_fp16")];
            tensor<string, []> var_50209_equation_0 = const()[name = tensor<string, []>("op_50209_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50209_cast_fp16 = einsum(equation = var_50209_equation_0, values = (var_50059_cast_fp16, var_50173_cast_fp16))[name = tensor<string, []>("op_50209_cast_fp16")];
            tensor<string, []> var_50211_equation_0 = const()[name = tensor<string, []>("op_50211_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50211_cast_fp16 = einsum(equation = var_50211_equation_0, values = (var_50063_cast_fp16, var_50174_cast_fp16))[name = tensor<string, []>("op_50211_cast_fp16")];
            tensor<string, []> var_50213_equation_0 = const()[name = tensor<string, []>("op_50213_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50213_cast_fp16 = einsum(equation = var_50213_equation_0, values = (var_50067_cast_fp16, var_50175_cast_fp16))[name = tensor<string, []>("op_50213_cast_fp16")];
            tensor<string, []> var_50215_equation_0 = const()[name = tensor<string, []>("op_50215_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50215_cast_fp16 = einsum(equation = var_50215_equation_0, values = (var_50071_cast_fp16, var_50176_cast_fp16))[name = tensor<string, []>("op_50215_cast_fp16")];
            tensor<string, []> var_50217_equation_0 = const()[name = tensor<string, []>("op_50217_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50217_cast_fp16 = einsum(equation = var_50217_equation_0, values = (var_50075_cast_fp16, var_50177_cast_fp16))[name = tensor<string, []>("op_50217_cast_fp16")];
            tensor<bool, []> input_643_interleave_0 = const()[name = tensor<string, []>("input_643_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_643_cast_fp16 = concat(axis = var_30355, interleave = input_643_interleave_0, values = (var_50179_cast_fp16, var_50181_cast_fp16, var_50183_cast_fp16, var_50185_cast_fp16, var_50187_cast_fp16, var_50189_cast_fp16, var_50191_cast_fp16, var_50193_cast_fp16, var_50195_cast_fp16, var_50197_cast_fp16, var_50199_cast_fp16, var_50201_cast_fp16, var_50203_cast_fp16, var_50205_cast_fp16, var_50207_cast_fp16, var_50209_cast_fp16, var_50211_cast_fp16, var_50213_cast_fp16, var_50215_cast_fp16, var_50217_cast_fp16))[name = tensor<string, []>("input_643_cast_fp16")];
            tensor<string, []> var_50227_pad_type_0 = const()[name = tensor<string, []>("op_50227_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_50227_strides_0 = const()[name = tensor<string, []>("op_50227_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_50227_pad_0 = const()[name = tensor<string, []>("op_50227_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_50227_dilations_0 = const()[name = tensor<string, []>("op_50227_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_50227_groups_0 = const()[name = tensor<string, []>("op_50227_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_1_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(628033216))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(629262080))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_1_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_1_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_1_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(629262272)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_50227_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_1_attn1_to_out_0_bias_to_fp16, dilations = var_50227_dilations_0, groups = var_50227_groups_0, pad = var_50227_pad_0, pad_type = var_50227_pad_type_0, strides = var_50227_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_1_attn1_to_out_0_weight_to_fp16_palettized, x = input_643_cast_fp16)[name = tensor<string, []>("op_50227_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_333_cast_fp16 = add(x = var_50227_cast_fp16, y = inputs_331_cast_fp16)[name = tensor<string, []>("inputs_333_cast_fp16")];
            tensor<int32, [1]> hidden_states_443_axes_0 = const()[name = tensor<string, []>("hidden_states_443_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_443_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_443_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(629264896)))];
            tensor<fp16, [1280]> hidden_states_443_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_443_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(629267520)))];
            tensor<fp16, []> var_50237_to_fp16 = const()[name = tensor<string, []>("op_50237_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_443_cast_fp16 = layer_norm(axes = hidden_states_443_axes_0, beta = hidden_states_443_beta_0_to_fp16, epsilon = var_50237_to_fp16, gamma = hidden_states_443_gamma_0_to_fp16, x = inputs_333_cast_fp16)[name = tensor<string, []>("hidden_states_443_cast_fp16")];
            tensor<string, []> q_223_pad_type_0 = const()[name = tensor<string, []>("q_223_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_223_strides_0 = const()[name = tensor<string, []>("q_223_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_223_pad_0 = const()[name = tensor<string, []>("q_223_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_223_dilations_0 = const()[name = tensor<string, []>("q_223_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_223_groups_0 = const()[name = tensor<string, []>("q_223_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_1_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(629270144))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(630499008))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_1_attn2_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_223_cast_fp16 = conv(dilations = q_223_dilations_0, groups = q_223_groups_0, pad = q_223_pad_0, pad_type = q_223_pad_type_0, strides = q_223_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_1_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_443_cast_fp16)[name = tensor<string, []>("q_223_cast_fp16")];
            tensor<string, []> k_445_pad_type_0 = const()[name = tensor<string, []>("k_445_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_445_strides_0 = const()[name = tensor<string, []>("k_445_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_445_pad_0 = const()[name = tensor<string, []>("k_445_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_445_dilations_0 = const()[name = tensor<string, []>("k_445_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_445_groups_0 = const()[name = tensor<string, []>("k_445_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_1_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(630499200))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(632465344))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_1_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> k_445_cast_fp16 = conv(dilations = k_445_dilations_0, groups = k_445_groups_0, pad = k_445_pad_0, pad_type = k_445_pad_type_0, strides = k_445_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_1_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("k_445_cast_fp16")];
            tensor<string, []> v_223_pad_type_0 = const()[name = tensor<string, []>("v_223_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_223_strides_0 = const()[name = tensor<string, []>("v_223_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_223_pad_0 = const()[name = tensor<string, []>("v_223_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_223_dilations_0 = const()[name = tensor<string, []>("v_223_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_223_groups_0 = const()[name = tensor<string, []>("v_223_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_1_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(632465536))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(634431680))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_1_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> v_223_cast_fp16 = conv(dilations = v_223_dilations_0, groups = v_223_groups_0, pad = v_223_pad_0, pad_type = v_223_pad_type_0, strides = v_223_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_1_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("v_223_cast_fp16")];
            tensor<int32, [4]> var_50270_begin_0 = const()[name = tensor<string, []>("op_50270_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_50270_end_0 = const()[name = tensor<string, []>("op_50270_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_50270_end_mask_0 = const()[name = tensor<string, []>("op_50270_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50270_cast_fp16 = slice_by_index(begin = var_50270_begin_0, end = var_50270_end_0, end_mask = var_50270_end_mask_0, x = q_223_cast_fp16)[name = tensor<string, []>("op_50270_cast_fp16")];
            tensor<int32, [4]> var_50274_begin_0 = const()[name = tensor<string, []>("op_50274_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_50274_end_0 = const()[name = tensor<string, []>("op_50274_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_50274_end_mask_0 = const()[name = tensor<string, []>("op_50274_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50274_cast_fp16 = slice_by_index(begin = var_50274_begin_0, end = var_50274_end_0, end_mask = var_50274_end_mask_0, x = q_223_cast_fp16)[name = tensor<string, []>("op_50274_cast_fp16")];
            tensor<int32, [4]> var_50278_begin_0 = const()[name = tensor<string, []>("op_50278_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_50278_end_0 = const()[name = tensor<string, []>("op_50278_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_50278_end_mask_0 = const()[name = tensor<string, []>("op_50278_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50278_cast_fp16 = slice_by_index(begin = var_50278_begin_0, end = var_50278_end_0, end_mask = var_50278_end_mask_0, x = q_223_cast_fp16)[name = tensor<string, []>("op_50278_cast_fp16")];
            tensor<int32, [4]> var_50282_begin_0 = const()[name = tensor<string, []>("op_50282_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_50282_end_0 = const()[name = tensor<string, []>("op_50282_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_50282_end_mask_0 = const()[name = tensor<string, []>("op_50282_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50282_cast_fp16 = slice_by_index(begin = var_50282_begin_0, end = var_50282_end_0, end_mask = var_50282_end_mask_0, x = q_223_cast_fp16)[name = tensor<string, []>("op_50282_cast_fp16")];
            tensor<int32, [4]> var_50286_begin_0 = const()[name = tensor<string, []>("op_50286_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_50286_end_0 = const()[name = tensor<string, []>("op_50286_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_50286_end_mask_0 = const()[name = tensor<string, []>("op_50286_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50286_cast_fp16 = slice_by_index(begin = var_50286_begin_0, end = var_50286_end_0, end_mask = var_50286_end_mask_0, x = q_223_cast_fp16)[name = tensor<string, []>("op_50286_cast_fp16")];
            tensor<int32, [4]> var_50290_begin_0 = const()[name = tensor<string, []>("op_50290_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_50290_end_0 = const()[name = tensor<string, []>("op_50290_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_50290_end_mask_0 = const()[name = tensor<string, []>("op_50290_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50290_cast_fp16 = slice_by_index(begin = var_50290_begin_0, end = var_50290_end_0, end_mask = var_50290_end_mask_0, x = q_223_cast_fp16)[name = tensor<string, []>("op_50290_cast_fp16")];
            tensor<int32, [4]> var_50294_begin_0 = const()[name = tensor<string, []>("op_50294_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_50294_end_0 = const()[name = tensor<string, []>("op_50294_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_50294_end_mask_0 = const()[name = tensor<string, []>("op_50294_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50294_cast_fp16 = slice_by_index(begin = var_50294_begin_0, end = var_50294_end_0, end_mask = var_50294_end_mask_0, x = q_223_cast_fp16)[name = tensor<string, []>("op_50294_cast_fp16")];
            tensor<int32, [4]> var_50298_begin_0 = const()[name = tensor<string, []>("op_50298_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_50298_end_0 = const()[name = tensor<string, []>("op_50298_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_50298_end_mask_0 = const()[name = tensor<string, []>("op_50298_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50298_cast_fp16 = slice_by_index(begin = var_50298_begin_0, end = var_50298_end_0, end_mask = var_50298_end_mask_0, x = q_223_cast_fp16)[name = tensor<string, []>("op_50298_cast_fp16")];
            tensor<int32, [4]> var_50302_begin_0 = const()[name = tensor<string, []>("op_50302_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_50302_end_0 = const()[name = tensor<string, []>("op_50302_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_50302_end_mask_0 = const()[name = tensor<string, []>("op_50302_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50302_cast_fp16 = slice_by_index(begin = var_50302_begin_0, end = var_50302_end_0, end_mask = var_50302_end_mask_0, x = q_223_cast_fp16)[name = tensor<string, []>("op_50302_cast_fp16")];
            tensor<int32, [4]> var_50306_begin_0 = const()[name = tensor<string, []>("op_50306_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_50306_end_0 = const()[name = tensor<string, []>("op_50306_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_50306_end_mask_0 = const()[name = tensor<string, []>("op_50306_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50306_cast_fp16 = slice_by_index(begin = var_50306_begin_0, end = var_50306_end_0, end_mask = var_50306_end_mask_0, x = q_223_cast_fp16)[name = tensor<string, []>("op_50306_cast_fp16")];
            tensor<int32, [4]> var_50310_begin_0 = const()[name = tensor<string, []>("op_50310_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_50310_end_0 = const()[name = tensor<string, []>("op_50310_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_50310_end_mask_0 = const()[name = tensor<string, []>("op_50310_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50310_cast_fp16 = slice_by_index(begin = var_50310_begin_0, end = var_50310_end_0, end_mask = var_50310_end_mask_0, x = q_223_cast_fp16)[name = tensor<string, []>("op_50310_cast_fp16")];
            tensor<int32, [4]> var_50314_begin_0 = const()[name = tensor<string, []>("op_50314_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_50314_end_0 = const()[name = tensor<string, []>("op_50314_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_50314_end_mask_0 = const()[name = tensor<string, []>("op_50314_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50314_cast_fp16 = slice_by_index(begin = var_50314_begin_0, end = var_50314_end_0, end_mask = var_50314_end_mask_0, x = q_223_cast_fp16)[name = tensor<string, []>("op_50314_cast_fp16")];
            tensor<int32, [4]> var_50318_begin_0 = const()[name = tensor<string, []>("op_50318_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_50318_end_0 = const()[name = tensor<string, []>("op_50318_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_50318_end_mask_0 = const()[name = tensor<string, []>("op_50318_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50318_cast_fp16 = slice_by_index(begin = var_50318_begin_0, end = var_50318_end_0, end_mask = var_50318_end_mask_0, x = q_223_cast_fp16)[name = tensor<string, []>("op_50318_cast_fp16")];
            tensor<int32, [4]> var_50322_begin_0 = const()[name = tensor<string, []>("op_50322_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_50322_end_0 = const()[name = tensor<string, []>("op_50322_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_50322_end_mask_0 = const()[name = tensor<string, []>("op_50322_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50322_cast_fp16 = slice_by_index(begin = var_50322_begin_0, end = var_50322_end_0, end_mask = var_50322_end_mask_0, x = q_223_cast_fp16)[name = tensor<string, []>("op_50322_cast_fp16")];
            tensor<int32, [4]> var_50326_begin_0 = const()[name = tensor<string, []>("op_50326_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_50326_end_0 = const()[name = tensor<string, []>("op_50326_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_50326_end_mask_0 = const()[name = tensor<string, []>("op_50326_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50326_cast_fp16 = slice_by_index(begin = var_50326_begin_0, end = var_50326_end_0, end_mask = var_50326_end_mask_0, x = q_223_cast_fp16)[name = tensor<string, []>("op_50326_cast_fp16")];
            tensor<int32, [4]> var_50330_begin_0 = const()[name = tensor<string, []>("op_50330_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_50330_end_0 = const()[name = tensor<string, []>("op_50330_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_50330_end_mask_0 = const()[name = tensor<string, []>("op_50330_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50330_cast_fp16 = slice_by_index(begin = var_50330_begin_0, end = var_50330_end_0, end_mask = var_50330_end_mask_0, x = q_223_cast_fp16)[name = tensor<string, []>("op_50330_cast_fp16")];
            tensor<int32, [4]> var_50334_begin_0 = const()[name = tensor<string, []>("op_50334_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_50334_end_0 = const()[name = tensor<string, []>("op_50334_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_50334_end_mask_0 = const()[name = tensor<string, []>("op_50334_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50334_cast_fp16 = slice_by_index(begin = var_50334_begin_0, end = var_50334_end_0, end_mask = var_50334_end_mask_0, x = q_223_cast_fp16)[name = tensor<string, []>("op_50334_cast_fp16")];
            tensor<int32, [4]> var_50338_begin_0 = const()[name = tensor<string, []>("op_50338_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_50338_end_0 = const()[name = tensor<string, []>("op_50338_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_50338_end_mask_0 = const()[name = tensor<string, []>("op_50338_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50338_cast_fp16 = slice_by_index(begin = var_50338_begin_0, end = var_50338_end_0, end_mask = var_50338_end_mask_0, x = q_223_cast_fp16)[name = tensor<string, []>("op_50338_cast_fp16")];
            tensor<int32, [4]> var_50342_begin_0 = const()[name = tensor<string, []>("op_50342_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_50342_end_0 = const()[name = tensor<string, []>("op_50342_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_50342_end_mask_0 = const()[name = tensor<string, []>("op_50342_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50342_cast_fp16 = slice_by_index(begin = var_50342_begin_0, end = var_50342_end_0, end_mask = var_50342_end_mask_0, x = q_223_cast_fp16)[name = tensor<string, []>("op_50342_cast_fp16")];
            tensor<int32, [4]> var_50346_begin_0 = const()[name = tensor<string, []>("op_50346_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_50346_end_0 = const()[name = tensor<string, []>("op_50346_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_50346_end_mask_0 = const()[name = tensor<string, []>("op_50346_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50346_cast_fp16 = slice_by_index(begin = var_50346_begin_0, end = var_50346_end_0, end_mask = var_50346_end_mask_0, x = q_223_cast_fp16)[name = tensor<string, []>("op_50346_cast_fp16")];
            tensor<int32, [4]> k_447_perm_0 = const()[name = tensor<string, []>("k_447_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_50353_begin_0 = const()[name = tensor<string, []>("op_50353_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_50353_end_0 = const()[name = tensor<string, []>("op_50353_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_50353_end_mask_0 = const()[name = tensor<string, []>("op_50353_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> k_447_cast_fp16 = transpose(perm = k_447_perm_0, x = k_445_cast_fp16)[name = tensor<string, []>("transpose_28")];
            tensor<fp16, [2, 77, 1, 64]> var_50353_cast_fp16 = slice_by_index(begin = var_50353_begin_0, end = var_50353_end_0, end_mask = var_50353_end_mask_0, x = k_447_cast_fp16)[name = tensor<string, []>("op_50353_cast_fp16")];
            tensor<int32, [4]> var_50357_begin_0 = const()[name = tensor<string, []>("op_50357_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_50357_end_0 = const()[name = tensor<string, []>("op_50357_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_50357_end_mask_0 = const()[name = tensor<string, []>("op_50357_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_50357_cast_fp16 = slice_by_index(begin = var_50357_begin_0, end = var_50357_end_0, end_mask = var_50357_end_mask_0, x = k_447_cast_fp16)[name = tensor<string, []>("op_50357_cast_fp16")];
            tensor<int32, [4]> var_50361_begin_0 = const()[name = tensor<string, []>("op_50361_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_50361_end_0 = const()[name = tensor<string, []>("op_50361_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_50361_end_mask_0 = const()[name = tensor<string, []>("op_50361_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_50361_cast_fp16 = slice_by_index(begin = var_50361_begin_0, end = var_50361_end_0, end_mask = var_50361_end_mask_0, x = k_447_cast_fp16)[name = tensor<string, []>("op_50361_cast_fp16")];
            tensor<int32, [4]> var_50365_begin_0 = const()[name = tensor<string, []>("op_50365_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_50365_end_0 = const()[name = tensor<string, []>("op_50365_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_50365_end_mask_0 = const()[name = tensor<string, []>("op_50365_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_50365_cast_fp16 = slice_by_index(begin = var_50365_begin_0, end = var_50365_end_0, end_mask = var_50365_end_mask_0, x = k_447_cast_fp16)[name = tensor<string, []>("op_50365_cast_fp16")];
            tensor<int32, [4]> var_50369_begin_0 = const()[name = tensor<string, []>("op_50369_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_50369_end_0 = const()[name = tensor<string, []>("op_50369_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_50369_end_mask_0 = const()[name = tensor<string, []>("op_50369_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_50369_cast_fp16 = slice_by_index(begin = var_50369_begin_0, end = var_50369_end_0, end_mask = var_50369_end_mask_0, x = k_447_cast_fp16)[name = tensor<string, []>("op_50369_cast_fp16")];
            tensor<int32, [4]> var_50373_begin_0 = const()[name = tensor<string, []>("op_50373_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_50373_end_0 = const()[name = tensor<string, []>("op_50373_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_50373_end_mask_0 = const()[name = tensor<string, []>("op_50373_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_50373_cast_fp16 = slice_by_index(begin = var_50373_begin_0, end = var_50373_end_0, end_mask = var_50373_end_mask_0, x = k_447_cast_fp16)[name = tensor<string, []>("op_50373_cast_fp16")];
            tensor<int32, [4]> var_50377_begin_0 = const()[name = tensor<string, []>("op_50377_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_50377_end_0 = const()[name = tensor<string, []>("op_50377_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_50377_end_mask_0 = const()[name = tensor<string, []>("op_50377_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_50377_cast_fp16 = slice_by_index(begin = var_50377_begin_0, end = var_50377_end_0, end_mask = var_50377_end_mask_0, x = k_447_cast_fp16)[name = tensor<string, []>("op_50377_cast_fp16")];
            tensor<int32, [4]> var_50381_begin_0 = const()[name = tensor<string, []>("op_50381_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_50381_end_0 = const()[name = tensor<string, []>("op_50381_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_50381_end_mask_0 = const()[name = tensor<string, []>("op_50381_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_50381_cast_fp16 = slice_by_index(begin = var_50381_begin_0, end = var_50381_end_0, end_mask = var_50381_end_mask_0, x = k_447_cast_fp16)[name = tensor<string, []>("op_50381_cast_fp16")];
            tensor<int32, [4]> var_50385_begin_0 = const()[name = tensor<string, []>("op_50385_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_50385_end_0 = const()[name = tensor<string, []>("op_50385_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_50385_end_mask_0 = const()[name = tensor<string, []>("op_50385_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_50385_cast_fp16 = slice_by_index(begin = var_50385_begin_0, end = var_50385_end_0, end_mask = var_50385_end_mask_0, x = k_447_cast_fp16)[name = tensor<string, []>("op_50385_cast_fp16")];
            tensor<int32, [4]> var_50389_begin_0 = const()[name = tensor<string, []>("op_50389_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_50389_end_0 = const()[name = tensor<string, []>("op_50389_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_50389_end_mask_0 = const()[name = tensor<string, []>("op_50389_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_50389_cast_fp16 = slice_by_index(begin = var_50389_begin_0, end = var_50389_end_0, end_mask = var_50389_end_mask_0, x = k_447_cast_fp16)[name = tensor<string, []>("op_50389_cast_fp16")];
            tensor<int32, [4]> var_50393_begin_0 = const()[name = tensor<string, []>("op_50393_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_50393_end_0 = const()[name = tensor<string, []>("op_50393_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_50393_end_mask_0 = const()[name = tensor<string, []>("op_50393_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_50393_cast_fp16 = slice_by_index(begin = var_50393_begin_0, end = var_50393_end_0, end_mask = var_50393_end_mask_0, x = k_447_cast_fp16)[name = tensor<string, []>("op_50393_cast_fp16")];
            tensor<int32, [4]> var_50397_begin_0 = const()[name = tensor<string, []>("op_50397_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_50397_end_0 = const()[name = tensor<string, []>("op_50397_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_50397_end_mask_0 = const()[name = tensor<string, []>("op_50397_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_50397_cast_fp16 = slice_by_index(begin = var_50397_begin_0, end = var_50397_end_0, end_mask = var_50397_end_mask_0, x = k_447_cast_fp16)[name = tensor<string, []>("op_50397_cast_fp16")];
            tensor<int32, [4]> var_50401_begin_0 = const()[name = tensor<string, []>("op_50401_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_50401_end_0 = const()[name = tensor<string, []>("op_50401_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_50401_end_mask_0 = const()[name = tensor<string, []>("op_50401_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_50401_cast_fp16 = slice_by_index(begin = var_50401_begin_0, end = var_50401_end_0, end_mask = var_50401_end_mask_0, x = k_447_cast_fp16)[name = tensor<string, []>("op_50401_cast_fp16")];
            tensor<int32, [4]> var_50405_begin_0 = const()[name = tensor<string, []>("op_50405_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_50405_end_0 = const()[name = tensor<string, []>("op_50405_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_50405_end_mask_0 = const()[name = tensor<string, []>("op_50405_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_50405_cast_fp16 = slice_by_index(begin = var_50405_begin_0, end = var_50405_end_0, end_mask = var_50405_end_mask_0, x = k_447_cast_fp16)[name = tensor<string, []>("op_50405_cast_fp16")];
            tensor<int32, [4]> var_50409_begin_0 = const()[name = tensor<string, []>("op_50409_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_50409_end_0 = const()[name = tensor<string, []>("op_50409_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_50409_end_mask_0 = const()[name = tensor<string, []>("op_50409_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_50409_cast_fp16 = slice_by_index(begin = var_50409_begin_0, end = var_50409_end_0, end_mask = var_50409_end_mask_0, x = k_447_cast_fp16)[name = tensor<string, []>("op_50409_cast_fp16")];
            tensor<int32, [4]> var_50413_begin_0 = const()[name = tensor<string, []>("op_50413_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_50413_end_0 = const()[name = tensor<string, []>("op_50413_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_50413_end_mask_0 = const()[name = tensor<string, []>("op_50413_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_50413_cast_fp16 = slice_by_index(begin = var_50413_begin_0, end = var_50413_end_0, end_mask = var_50413_end_mask_0, x = k_447_cast_fp16)[name = tensor<string, []>("op_50413_cast_fp16")];
            tensor<int32, [4]> var_50417_begin_0 = const()[name = tensor<string, []>("op_50417_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_50417_end_0 = const()[name = tensor<string, []>("op_50417_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_50417_end_mask_0 = const()[name = tensor<string, []>("op_50417_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_50417_cast_fp16 = slice_by_index(begin = var_50417_begin_0, end = var_50417_end_0, end_mask = var_50417_end_mask_0, x = k_447_cast_fp16)[name = tensor<string, []>("op_50417_cast_fp16")];
            tensor<int32, [4]> var_50421_begin_0 = const()[name = tensor<string, []>("op_50421_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_50421_end_0 = const()[name = tensor<string, []>("op_50421_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_50421_end_mask_0 = const()[name = tensor<string, []>("op_50421_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_50421_cast_fp16 = slice_by_index(begin = var_50421_begin_0, end = var_50421_end_0, end_mask = var_50421_end_mask_0, x = k_447_cast_fp16)[name = tensor<string, []>("op_50421_cast_fp16")];
            tensor<int32, [4]> var_50425_begin_0 = const()[name = tensor<string, []>("op_50425_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_50425_end_0 = const()[name = tensor<string, []>("op_50425_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_50425_end_mask_0 = const()[name = tensor<string, []>("op_50425_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_50425_cast_fp16 = slice_by_index(begin = var_50425_begin_0, end = var_50425_end_0, end_mask = var_50425_end_mask_0, x = k_447_cast_fp16)[name = tensor<string, []>("op_50425_cast_fp16")];
            tensor<int32, [4]> var_50429_begin_0 = const()[name = tensor<string, []>("op_50429_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_50429_end_0 = const()[name = tensor<string, []>("op_50429_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_50429_end_mask_0 = const()[name = tensor<string, []>("op_50429_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_50429_cast_fp16 = slice_by_index(begin = var_50429_begin_0, end = var_50429_end_0, end_mask = var_50429_end_mask_0, x = k_447_cast_fp16)[name = tensor<string, []>("op_50429_cast_fp16")];
            tensor<int32, [4]> var_50431_begin_0 = const()[name = tensor<string, []>("op_50431_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_50431_end_0 = const()[name = tensor<string, []>("op_50431_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_50431_end_mask_0 = const()[name = tensor<string, []>("op_50431_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_50431_cast_fp16 = slice_by_index(begin = var_50431_begin_0, end = var_50431_end_0, end_mask = var_50431_end_mask_0, x = v_223_cast_fp16)[name = tensor<string, []>("op_50431_cast_fp16")];
            tensor<int32, [4]> var_50435_begin_0 = const()[name = tensor<string, []>("op_50435_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_50435_end_0 = const()[name = tensor<string, []>("op_50435_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_50435_end_mask_0 = const()[name = tensor<string, []>("op_50435_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_50435_cast_fp16 = slice_by_index(begin = var_50435_begin_0, end = var_50435_end_0, end_mask = var_50435_end_mask_0, x = v_223_cast_fp16)[name = tensor<string, []>("op_50435_cast_fp16")];
            tensor<int32, [4]> var_50439_begin_0 = const()[name = tensor<string, []>("op_50439_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_50439_end_0 = const()[name = tensor<string, []>("op_50439_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_50439_end_mask_0 = const()[name = tensor<string, []>("op_50439_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_50439_cast_fp16 = slice_by_index(begin = var_50439_begin_0, end = var_50439_end_0, end_mask = var_50439_end_mask_0, x = v_223_cast_fp16)[name = tensor<string, []>("op_50439_cast_fp16")];
            tensor<int32, [4]> var_50443_begin_0 = const()[name = tensor<string, []>("op_50443_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_50443_end_0 = const()[name = tensor<string, []>("op_50443_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_50443_end_mask_0 = const()[name = tensor<string, []>("op_50443_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_50443_cast_fp16 = slice_by_index(begin = var_50443_begin_0, end = var_50443_end_0, end_mask = var_50443_end_mask_0, x = v_223_cast_fp16)[name = tensor<string, []>("op_50443_cast_fp16")];
            tensor<int32, [4]> var_50447_begin_0 = const()[name = tensor<string, []>("op_50447_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_50447_end_0 = const()[name = tensor<string, []>("op_50447_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_50447_end_mask_0 = const()[name = tensor<string, []>("op_50447_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_50447_cast_fp16 = slice_by_index(begin = var_50447_begin_0, end = var_50447_end_0, end_mask = var_50447_end_mask_0, x = v_223_cast_fp16)[name = tensor<string, []>("op_50447_cast_fp16")];
            tensor<int32, [4]> var_50451_begin_0 = const()[name = tensor<string, []>("op_50451_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_50451_end_0 = const()[name = tensor<string, []>("op_50451_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_50451_end_mask_0 = const()[name = tensor<string, []>("op_50451_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_50451_cast_fp16 = slice_by_index(begin = var_50451_begin_0, end = var_50451_end_0, end_mask = var_50451_end_mask_0, x = v_223_cast_fp16)[name = tensor<string, []>("op_50451_cast_fp16")];
            tensor<int32, [4]> var_50455_begin_0 = const()[name = tensor<string, []>("op_50455_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_50455_end_0 = const()[name = tensor<string, []>("op_50455_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_50455_end_mask_0 = const()[name = tensor<string, []>("op_50455_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_50455_cast_fp16 = slice_by_index(begin = var_50455_begin_0, end = var_50455_end_0, end_mask = var_50455_end_mask_0, x = v_223_cast_fp16)[name = tensor<string, []>("op_50455_cast_fp16")];
            tensor<int32, [4]> var_50459_begin_0 = const()[name = tensor<string, []>("op_50459_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_50459_end_0 = const()[name = tensor<string, []>("op_50459_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_50459_end_mask_0 = const()[name = tensor<string, []>("op_50459_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_50459_cast_fp16 = slice_by_index(begin = var_50459_begin_0, end = var_50459_end_0, end_mask = var_50459_end_mask_0, x = v_223_cast_fp16)[name = tensor<string, []>("op_50459_cast_fp16")];
            tensor<int32, [4]> var_50463_begin_0 = const()[name = tensor<string, []>("op_50463_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_50463_end_0 = const()[name = tensor<string, []>("op_50463_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_50463_end_mask_0 = const()[name = tensor<string, []>("op_50463_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_50463_cast_fp16 = slice_by_index(begin = var_50463_begin_0, end = var_50463_end_0, end_mask = var_50463_end_mask_0, x = v_223_cast_fp16)[name = tensor<string, []>("op_50463_cast_fp16")];
            tensor<int32, [4]> var_50467_begin_0 = const()[name = tensor<string, []>("op_50467_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_50467_end_0 = const()[name = tensor<string, []>("op_50467_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_50467_end_mask_0 = const()[name = tensor<string, []>("op_50467_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_50467_cast_fp16 = slice_by_index(begin = var_50467_begin_0, end = var_50467_end_0, end_mask = var_50467_end_mask_0, x = v_223_cast_fp16)[name = tensor<string, []>("op_50467_cast_fp16")];
            tensor<int32, [4]> var_50471_begin_0 = const()[name = tensor<string, []>("op_50471_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_50471_end_0 = const()[name = tensor<string, []>("op_50471_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_50471_end_mask_0 = const()[name = tensor<string, []>("op_50471_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_50471_cast_fp16 = slice_by_index(begin = var_50471_begin_0, end = var_50471_end_0, end_mask = var_50471_end_mask_0, x = v_223_cast_fp16)[name = tensor<string, []>("op_50471_cast_fp16")];
            tensor<int32, [4]> var_50475_begin_0 = const()[name = tensor<string, []>("op_50475_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_50475_end_0 = const()[name = tensor<string, []>("op_50475_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_50475_end_mask_0 = const()[name = tensor<string, []>("op_50475_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_50475_cast_fp16 = slice_by_index(begin = var_50475_begin_0, end = var_50475_end_0, end_mask = var_50475_end_mask_0, x = v_223_cast_fp16)[name = tensor<string, []>("op_50475_cast_fp16")];
            tensor<int32, [4]> var_50479_begin_0 = const()[name = tensor<string, []>("op_50479_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_50479_end_0 = const()[name = tensor<string, []>("op_50479_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_50479_end_mask_0 = const()[name = tensor<string, []>("op_50479_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_50479_cast_fp16 = slice_by_index(begin = var_50479_begin_0, end = var_50479_end_0, end_mask = var_50479_end_mask_0, x = v_223_cast_fp16)[name = tensor<string, []>("op_50479_cast_fp16")];
            tensor<int32, [4]> var_50483_begin_0 = const()[name = tensor<string, []>("op_50483_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_50483_end_0 = const()[name = tensor<string, []>("op_50483_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_50483_end_mask_0 = const()[name = tensor<string, []>("op_50483_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_50483_cast_fp16 = slice_by_index(begin = var_50483_begin_0, end = var_50483_end_0, end_mask = var_50483_end_mask_0, x = v_223_cast_fp16)[name = tensor<string, []>("op_50483_cast_fp16")];
            tensor<int32, [4]> var_50487_begin_0 = const()[name = tensor<string, []>("op_50487_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_50487_end_0 = const()[name = tensor<string, []>("op_50487_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_50487_end_mask_0 = const()[name = tensor<string, []>("op_50487_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_50487_cast_fp16 = slice_by_index(begin = var_50487_begin_0, end = var_50487_end_0, end_mask = var_50487_end_mask_0, x = v_223_cast_fp16)[name = tensor<string, []>("op_50487_cast_fp16")];
            tensor<int32, [4]> var_50491_begin_0 = const()[name = tensor<string, []>("op_50491_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_50491_end_0 = const()[name = tensor<string, []>("op_50491_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_50491_end_mask_0 = const()[name = tensor<string, []>("op_50491_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_50491_cast_fp16 = slice_by_index(begin = var_50491_begin_0, end = var_50491_end_0, end_mask = var_50491_end_mask_0, x = v_223_cast_fp16)[name = tensor<string, []>("op_50491_cast_fp16")];
            tensor<int32, [4]> var_50495_begin_0 = const()[name = tensor<string, []>("op_50495_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_50495_end_0 = const()[name = tensor<string, []>("op_50495_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_50495_end_mask_0 = const()[name = tensor<string, []>("op_50495_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_50495_cast_fp16 = slice_by_index(begin = var_50495_begin_0, end = var_50495_end_0, end_mask = var_50495_end_mask_0, x = v_223_cast_fp16)[name = tensor<string, []>("op_50495_cast_fp16")];
            tensor<int32, [4]> var_50499_begin_0 = const()[name = tensor<string, []>("op_50499_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_50499_end_0 = const()[name = tensor<string, []>("op_50499_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_50499_end_mask_0 = const()[name = tensor<string, []>("op_50499_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_50499_cast_fp16 = slice_by_index(begin = var_50499_begin_0, end = var_50499_end_0, end_mask = var_50499_end_mask_0, x = v_223_cast_fp16)[name = tensor<string, []>("op_50499_cast_fp16")];
            tensor<int32, [4]> var_50503_begin_0 = const()[name = tensor<string, []>("op_50503_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_50503_end_0 = const()[name = tensor<string, []>("op_50503_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_50503_end_mask_0 = const()[name = tensor<string, []>("op_50503_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_50503_cast_fp16 = slice_by_index(begin = var_50503_begin_0, end = var_50503_end_0, end_mask = var_50503_end_mask_0, x = v_223_cast_fp16)[name = tensor<string, []>("op_50503_cast_fp16")];
            tensor<int32, [4]> var_50507_begin_0 = const()[name = tensor<string, []>("op_50507_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_50507_end_0 = const()[name = tensor<string, []>("op_50507_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_50507_end_mask_0 = const()[name = tensor<string, []>("op_50507_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_50507_cast_fp16 = slice_by_index(begin = var_50507_begin_0, end = var_50507_end_0, end_mask = var_50507_end_mask_0, x = v_223_cast_fp16)[name = tensor<string, []>("op_50507_cast_fp16")];
            tensor<string, []> var_50511_equation_0 = const()[name = tensor<string, []>("op_50511_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_50511_cast_fp16 = einsum(equation = var_50511_equation_0, values = (var_50353_cast_fp16, var_50270_cast_fp16))[name = tensor<string, []>("op_50511_cast_fp16")];
            tensor<fp16, []> var_50512_to_fp16 = const()[name = tensor<string, []>("op_50512_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4281_cast_fp16 = mul(x = var_50511_cast_fp16, y = var_50512_to_fp16)[name = tensor<string, []>("aw_4281_cast_fp16")];
            tensor<string, []> var_50515_equation_0 = const()[name = tensor<string, []>("op_50515_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_50515_cast_fp16 = einsum(equation = var_50515_equation_0, values = (var_50357_cast_fp16, var_50274_cast_fp16))[name = tensor<string, []>("op_50515_cast_fp16")];
            tensor<fp16, []> var_50516_to_fp16 = const()[name = tensor<string, []>("op_50516_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4283_cast_fp16 = mul(x = var_50515_cast_fp16, y = var_50516_to_fp16)[name = tensor<string, []>("aw_4283_cast_fp16")];
            tensor<string, []> var_50519_equation_0 = const()[name = tensor<string, []>("op_50519_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_50519_cast_fp16 = einsum(equation = var_50519_equation_0, values = (var_50361_cast_fp16, var_50278_cast_fp16))[name = tensor<string, []>("op_50519_cast_fp16")];
            tensor<fp16, []> var_50520_to_fp16 = const()[name = tensor<string, []>("op_50520_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4285_cast_fp16 = mul(x = var_50519_cast_fp16, y = var_50520_to_fp16)[name = tensor<string, []>("aw_4285_cast_fp16")];
            tensor<string, []> var_50523_equation_0 = const()[name = tensor<string, []>("op_50523_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_50523_cast_fp16 = einsum(equation = var_50523_equation_0, values = (var_50365_cast_fp16, var_50282_cast_fp16))[name = tensor<string, []>("op_50523_cast_fp16")];
            tensor<fp16, []> var_50524_to_fp16 = const()[name = tensor<string, []>("op_50524_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4287_cast_fp16 = mul(x = var_50523_cast_fp16, y = var_50524_to_fp16)[name = tensor<string, []>("aw_4287_cast_fp16")];
            tensor<string, []> var_50527_equation_0 = const()[name = tensor<string, []>("op_50527_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_50527_cast_fp16 = einsum(equation = var_50527_equation_0, values = (var_50369_cast_fp16, var_50286_cast_fp16))[name = tensor<string, []>("op_50527_cast_fp16")];
            tensor<fp16, []> var_50528_to_fp16 = const()[name = tensor<string, []>("op_50528_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4289_cast_fp16 = mul(x = var_50527_cast_fp16, y = var_50528_to_fp16)[name = tensor<string, []>("aw_4289_cast_fp16")];
            tensor<string, []> var_50531_equation_0 = const()[name = tensor<string, []>("op_50531_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_50531_cast_fp16 = einsum(equation = var_50531_equation_0, values = (var_50373_cast_fp16, var_50290_cast_fp16))[name = tensor<string, []>("op_50531_cast_fp16")];
            tensor<fp16, []> var_50532_to_fp16 = const()[name = tensor<string, []>("op_50532_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4291_cast_fp16 = mul(x = var_50531_cast_fp16, y = var_50532_to_fp16)[name = tensor<string, []>("aw_4291_cast_fp16")];
            tensor<string, []> var_50535_equation_0 = const()[name = tensor<string, []>("op_50535_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_50535_cast_fp16 = einsum(equation = var_50535_equation_0, values = (var_50377_cast_fp16, var_50294_cast_fp16))[name = tensor<string, []>("op_50535_cast_fp16")];
            tensor<fp16, []> var_50536_to_fp16 = const()[name = tensor<string, []>("op_50536_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4293_cast_fp16 = mul(x = var_50535_cast_fp16, y = var_50536_to_fp16)[name = tensor<string, []>("aw_4293_cast_fp16")];
            tensor<string, []> var_50539_equation_0 = const()[name = tensor<string, []>("op_50539_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_50539_cast_fp16 = einsum(equation = var_50539_equation_0, values = (var_50381_cast_fp16, var_50298_cast_fp16))[name = tensor<string, []>("op_50539_cast_fp16")];
            tensor<fp16, []> var_50540_to_fp16 = const()[name = tensor<string, []>("op_50540_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4295_cast_fp16 = mul(x = var_50539_cast_fp16, y = var_50540_to_fp16)[name = tensor<string, []>("aw_4295_cast_fp16")];
            tensor<string, []> var_50543_equation_0 = const()[name = tensor<string, []>("op_50543_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_50543_cast_fp16 = einsum(equation = var_50543_equation_0, values = (var_50385_cast_fp16, var_50302_cast_fp16))[name = tensor<string, []>("op_50543_cast_fp16")];
            tensor<fp16, []> var_50544_to_fp16 = const()[name = tensor<string, []>("op_50544_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4297_cast_fp16 = mul(x = var_50543_cast_fp16, y = var_50544_to_fp16)[name = tensor<string, []>("aw_4297_cast_fp16")];
            tensor<string, []> var_50547_equation_0 = const()[name = tensor<string, []>("op_50547_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_50547_cast_fp16 = einsum(equation = var_50547_equation_0, values = (var_50389_cast_fp16, var_50306_cast_fp16))[name = tensor<string, []>("op_50547_cast_fp16")];
            tensor<fp16, []> var_50548_to_fp16 = const()[name = tensor<string, []>("op_50548_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4299_cast_fp16 = mul(x = var_50547_cast_fp16, y = var_50548_to_fp16)[name = tensor<string, []>("aw_4299_cast_fp16")];
            tensor<string, []> var_50551_equation_0 = const()[name = tensor<string, []>("op_50551_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_50551_cast_fp16 = einsum(equation = var_50551_equation_0, values = (var_50393_cast_fp16, var_50310_cast_fp16))[name = tensor<string, []>("op_50551_cast_fp16")];
            tensor<fp16, []> var_50552_to_fp16 = const()[name = tensor<string, []>("op_50552_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4301_cast_fp16 = mul(x = var_50551_cast_fp16, y = var_50552_to_fp16)[name = tensor<string, []>("aw_4301_cast_fp16")];
            tensor<string, []> var_50555_equation_0 = const()[name = tensor<string, []>("op_50555_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_50555_cast_fp16 = einsum(equation = var_50555_equation_0, values = (var_50397_cast_fp16, var_50314_cast_fp16))[name = tensor<string, []>("op_50555_cast_fp16")];
            tensor<fp16, []> var_50556_to_fp16 = const()[name = tensor<string, []>("op_50556_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4303_cast_fp16 = mul(x = var_50555_cast_fp16, y = var_50556_to_fp16)[name = tensor<string, []>("aw_4303_cast_fp16")];
            tensor<string, []> var_50559_equation_0 = const()[name = tensor<string, []>("op_50559_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_50559_cast_fp16 = einsum(equation = var_50559_equation_0, values = (var_50401_cast_fp16, var_50318_cast_fp16))[name = tensor<string, []>("op_50559_cast_fp16")];
            tensor<fp16, []> var_50560_to_fp16 = const()[name = tensor<string, []>("op_50560_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4305_cast_fp16 = mul(x = var_50559_cast_fp16, y = var_50560_to_fp16)[name = tensor<string, []>("aw_4305_cast_fp16")];
            tensor<string, []> var_50563_equation_0 = const()[name = tensor<string, []>("op_50563_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_50563_cast_fp16 = einsum(equation = var_50563_equation_0, values = (var_50405_cast_fp16, var_50322_cast_fp16))[name = tensor<string, []>("op_50563_cast_fp16")];
            tensor<fp16, []> var_50564_to_fp16 = const()[name = tensor<string, []>("op_50564_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4307_cast_fp16 = mul(x = var_50563_cast_fp16, y = var_50564_to_fp16)[name = tensor<string, []>("aw_4307_cast_fp16")];
            tensor<string, []> var_50567_equation_0 = const()[name = tensor<string, []>("op_50567_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_50567_cast_fp16 = einsum(equation = var_50567_equation_0, values = (var_50409_cast_fp16, var_50326_cast_fp16))[name = tensor<string, []>("op_50567_cast_fp16")];
            tensor<fp16, []> var_50568_to_fp16 = const()[name = tensor<string, []>("op_50568_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4309_cast_fp16 = mul(x = var_50567_cast_fp16, y = var_50568_to_fp16)[name = tensor<string, []>("aw_4309_cast_fp16")];
            tensor<string, []> var_50571_equation_0 = const()[name = tensor<string, []>("op_50571_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_50571_cast_fp16 = einsum(equation = var_50571_equation_0, values = (var_50413_cast_fp16, var_50330_cast_fp16))[name = tensor<string, []>("op_50571_cast_fp16")];
            tensor<fp16, []> var_50572_to_fp16 = const()[name = tensor<string, []>("op_50572_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4311_cast_fp16 = mul(x = var_50571_cast_fp16, y = var_50572_to_fp16)[name = tensor<string, []>("aw_4311_cast_fp16")];
            tensor<string, []> var_50575_equation_0 = const()[name = tensor<string, []>("op_50575_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_50575_cast_fp16 = einsum(equation = var_50575_equation_0, values = (var_50417_cast_fp16, var_50334_cast_fp16))[name = tensor<string, []>("op_50575_cast_fp16")];
            tensor<fp16, []> var_50576_to_fp16 = const()[name = tensor<string, []>("op_50576_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4313_cast_fp16 = mul(x = var_50575_cast_fp16, y = var_50576_to_fp16)[name = tensor<string, []>("aw_4313_cast_fp16")];
            tensor<string, []> var_50579_equation_0 = const()[name = tensor<string, []>("op_50579_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_50579_cast_fp16 = einsum(equation = var_50579_equation_0, values = (var_50421_cast_fp16, var_50338_cast_fp16))[name = tensor<string, []>("op_50579_cast_fp16")];
            tensor<fp16, []> var_50580_to_fp16 = const()[name = tensor<string, []>("op_50580_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4315_cast_fp16 = mul(x = var_50579_cast_fp16, y = var_50580_to_fp16)[name = tensor<string, []>("aw_4315_cast_fp16")];
            tensor<string, []> var_50583_equation_0 = const()[name = tensor<string, []>("op_50583_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_50583_cast_fp16 = einsum(equation = var_50583_equation_0, values = (var_50425_cast_fp16, var_50342_cast_fp16))[name = tensor<string, []>("op_50583_cast_fp16")];
            tensor<fp16, []> var_50584_to_fp16 = const()[name = tensor<string, []>("op_50584_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4317_cast_fp16 = mul(x = var_50583_cast_fp16, y = var_50584_to_fp16)[name = tensor<string, []>("aw_4317_cast_fp16")];
            tensor<string, []> var_50587_equation_0 = const()[name = tensor<string, []>("op_50587_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_50587_cast_fp16 = einsum(equation = var_50587_equation_0, values = (var_50429_cast_fp16, var_50346_cast_fp16))[name = tensor<string, []>("op_50587_cast_fp16")];
            tensor<fp16, []> var_50588_to_fp16 = const()[name = tensor<string, []>("op_50588_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4319_cast_fp16 = mul(x = var_50587_cast_fp16, y = var_50588_to_fp16)[name = tensor<string, []>("aw_4319_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_50590_cast_fp16 = softmax(axis = var_30355, x = aw_4281_cast_fp16)[name = tensor<string, []>("op_50590_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_50591_cast_fp16 = softmax(axis = var_30355, x = aw_4283_cast_fp16)[name = tensor<string, []>("op_50591_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_50592_cast_fp16 = softmax(axis = var_30355, x = aw_4285_cast_fp16)[name = tensor<string, []>("op_50592_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_50593_cast_fp16 = softmax(axis = var_30355, x = aw_4287_cast_fp16)[name = tensor<string, []>("op_50593_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_50594_cast_fp16 = softmax(axis = var_30355, x = aw_4289_cast_fp16)[name = tensor<string, []>("op_50594_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_50595_cast_fp16 = softmax(axis = var_30355, x = aw_4291_cast_fp16)[name = tensor<string, []>("op_50595_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_50596_cast_fp16 = softmax(axis = var_30355, x = aw_4293_cast_fp16)[name = tensor<string, []>("op_50596_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_50597_cast_fp16 = softmax(axis = var_30355, x = aw_4295_cast_fp16)[name = tensor<string, []>("op_50597_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_50598_cast_fp16 = softmax(axis = var_30355, x = aw_4297_cast_fp16)[name = tensor<string, []>("op_50598_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_50599_cast_fp16 = softmax(axis = var_30355, x = aw_4299_cast_fp16)[name = tensor<string, []>("op_50599_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_50600_cast_fp16 = softmax(axis = var_30355, x = aw_4301_cast_fp16)[name = tensor<string, []>("op_50600_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_50601_cast_fp16 = softmax(axis = var_30355, x = aw_4303_cast_fp16)[name = tensor<string, []>("op_50601_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_50602_cast_fp16 = softmax(axis = var_30355, x = aw_4305_cast_fp16)[name = tensor<string, []>("op_50602_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_50603_cast_fp16 = softmax(axis = var_30355, x = aw_4307_cast_fp16)[name = tensor<string, []>("op_50603_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_50604_cast_fp16 = softmax(axis = var_30355, x = aw_4309_cast_fp16)[name = tensor<string, []>("op_50604_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_50605_cast_fp16 = softmax(axis = var_30355, x = aw_4311_cast_fp16)[name = tensor<string, []>("op_50605_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_50606_cast_fp16 = softmax(axis = var_30355, x = aw_4313_cast_fp16)[name = tensor<string, []>("op_50606_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_50607_cast_fp16 = softmax(axis = var_30355, x = aw_4315_cast_fp16)[name = tensor<string, []>("op_50607_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_50608_cast_fp16 = softmax(axis = var_30355, x = aw_4317_cast_fp16)[name = tensor<string, []>("op_50608_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_50609_cast_fp16 = softmax(axis = var_30355, x = aw_4319_cast_fp16)[name = tensor<string, []>("op_50609_cast_fp16")];
            tensor<string, []> var_50611_equation_0 = const()[name = tensor<string, []>("op_50611_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50611_cast_fp16 = einsum(equation = var_50611_equation_0, values = (var_50431_cast_fp16, var_50590_cast_fp16))[name = tensor<string, []>("op_50611_cast_fp16")];
            tensor<string, []> var_50613_equation_0 = const()[name = tensor<string, []>("op_50613_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50613_cast_fp16 = einsum(equation = var_50613_equation_0, values = (var_50435_cast_fp16, var_50591_cast_fp16))[name = tensor<string, []>("op_50613_cast_fp16")];
            tensor<string, []> var_50615_equation_0 = const()[name = tensor<string, []>("op_50615_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50615_cast_fp16 = einsum(equation = var_50615_equation_0, values = (var_50439_cast_fp16, var_50592_cast_fp16))[name = tensor<string, []>("op_50615_cast_fp16")];
            tensor<string, []> var_50617_equation_0 = const()[name = tensor<string, []>("op_50617_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50617_cast_fp16 = einsum(equation = var_50617_equation_0, values = (var_50443_cast_fp16, var_50593_cast_fp16))[name = tensor<string, []>("op_50617_cast_fp16")];
            tensor<string, []> var_50619_equation_0 = const()[name = tensor<string, []>("op_50619_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50619_cast_fp16 = einsum(equation = var_50619_equation_0, values = (var_50447_cast_fp16, var_50594_cast_fp16))[name = tensor<string, []>("op_50619_cast_fp16")];
            tensor<string, []> var_50621_equation_0 = const()[name = tensor<string, []>("op_50621_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50621_cast_fp16 = einsum(equation = var_50621_equation_0, values = (var_50451_cast_fp16, var_50595_cast_fp16))[name = tensor<string, []>("op_50621_cast_fp16")];
            tensor<string, []> var_50623_equation_0 = const()[name = tensor<string, []>("op_50623_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50623_cast_fp16 = einsum(equation = var_50623_equation_0, values = (var_50455_cast_fp16, var_50596_cast_fp16))[name = tensor<string, []>("op_50623_cast_fp16")];
            tensor<string, []> var_50625_equation_0 = const()[name = tensor<string, []>("op_50625_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50625_cast_fp16 = einsum(equation = var_50625_equation_0, values = (var_50459_cast_fp16, var_50597_cast_fp16))[name = tensor<string, []>("op_50625_cast_fp16")];
            tensor<string, []> var_50627_equation_0 = const()[name = tensor<string, []>("op_50627_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50627_cast_fp16 = einsum(equation = var_50627_equation_0, values = (var_50463_cast_fp16, var_50598_cast_fp16))[name = tensor<string, []>("op_50627_cast_fp16")];
            tensor<string, []> var_50629_equation_0 = const()[name = tensor<string, []>("op_50629_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50629_cast_fp16 = einsum(equation = var_50629_equation_0, values = (var_50467_cast_fp16, var_50599_cast_fp16))[name = tensor<string, []>("op_50629_cast_fp16")];
            tensor<string, []> var_50631_equation_0 = const()[name = tensor<string, []>("op_50631_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50631_cast_fp16 = einsum(equation = var_50631_equation_0, values = (var_50471_cast_fp16, var_50600_cast_fp16))[name = tensor<string, []>("op_50631_cast_fp16")];
            tensor<string, []> var_50633_equation_0 = const()[name = tensor<string, []>("op_50633_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50633_cast_fp16 = einsum(equation = var_50633_equation_0, values = (var_50475_cast_fp16, var_50601_cast_fp16))[name = tensor<string, []>("op_50633_cast_fp16")];
            tensor<string, []> var_50635_equation_0 = const()[name = tensor<string, []>("op_50635_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50635_cast_fp16 = einsum(equation = var_50635_equation_0, values = (var_50479_cast_fp16, var_50602_cast_fp16))[name = tensor<string, []>("op_50635_cast_fp16")];
            tensor<string, []> var_50637_equation_0 = const()[name = tensor<string, []>("op_50637_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50637_cast_fp16 = einsum(equation = var_50637_equation_0, values = (var_50483_cast_fp16, var_50603_cast_fp16))[name = tensor<string, []>("op_50637_cast_fp16")];
            tensor<string, []> var_50639_equation_0 = const()[name = tensor<string, []>("op_50639_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50639_cast_fp16 = einsum(equation = var_50639_equation_0, values = (var_50487_cast_fp16, var_50604_cast_fp16))[name = tensor<string, []>("op_50639_cast_fp16")];
            tensor<string, []> var_50641_equation_0 = const()[name = tensor<string, []>("op_50641_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50641_cast_fp16 = einsum(equation = var_50641_equation_0, values = (var_50491_cast_fp16, var_50605_cast_fp16))[name = tensor<string, []>("op_50641_cast_fp16")];
            tensor<string, []> var_50643_equation_0 = const()[name = tensor<string, []>("op_50643_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50643_cast_fp16 = einsum(equation = var_50643_equation_0, values = (var_50495_cast_fp16, var_50606_cast_fp16))[name = tensor<string, []>("op_50643_cast_fp16")];
            tensor<string, []> var_50645_equation_0 = const()[name = tensor<string, []>("op_50645_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50645_cast_fp16 = einsum(equation = var_50645_equation_0, values = (var_50499_cast_fp16, var_50607_cast_fp16))[name = tensor<string, []>("op_50645_cast_fp16")];
            tensor<string, []> var_50647_equation_0 = const()[name = tensor<string, []>("op_50647_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50647_cast_fp16 = einsum(equation = var_50647_equation_0, values = (var_50503_cast_fp16, var_50608_cast_fp16))[name = tensor<string, []>("op_50647_cast_fp16")];
            tensor<string, []> var_50649_equation_0 = const()[name = tensor<string, []>("op_50649_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50649_cast_fp16 = einsum(equation = var_50649_equation_0, values = (var_50507_cast_fp16, var_50609_cast_fp16))[name = tensor<string, []>("op_50649_cast_fp16")];
            tensor<bool, []> input_645_interleave_0 = const()[name = tensor<string, []>("input_645_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_645_cast_fp16 = concat(axis = var_30355, interleave = input_645_interleave_0, values = (var_50611_cast_fp16, var_50613_cast_fp16, var_50615_cast_fp16, var_50617_cast_fp16, var_50619_cast_fp16, var_50621_cast_fp16, var_50623_cast_fp16, var_50625_cast_fp16, var_50627_cast_fp16, var_50629_cast_fp16, var_50631_cast_fp16, var_50633_cast_fp16, var_50635_cast_fp16, var_50637_cast_fp16, var_50639_cast_fp16, var_50641_cast_fp16, var_50643_cast_fp16, var_50645_cast_fp16, var_50647_cast_fp16, var_50649_cast_fp16))[name = tensor<string, []>("input_645_cast_fp16")];
            tensor<string, []> var_50659_pad_type_0 = const()[name = tensor<string, []>("op_50659_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_50659_strides_0 = const()[name = tensor<string, []>("op_50659_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_50659_pad_0 = const()[name = tensor<string, []>("op_50659_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_50659_dilations_0 = const()[name = tensor<string, []>("op_50659_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_50659_groups_0 = const()[name = tensor<string, []>("op_50659_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_1_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(634431872))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(635660736))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_1_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_1_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_1_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(635660928)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_50659_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_1_attn2_to_out_0_bias_to_fp16, dilations = var_50659_dilations_0, groups = var_50659_groups_0, pad = var_50659_pad_0, pad_type = var_50659_pad_type_0, strides = var_50659_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_1_attn2_to_out_0_weight_to_fp16_palettized, x = input_645_cast_fp16)[name = tensor<string, []>("op_50659_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_335_cast_fp16 = add(x = var_50659_cast_fp16, y = inputs_333_cast_fp16)[name = tensor<string, []>("inputs_335_cast_fp16")];
            tensor<int32, [1]> input_647_axes_0 = const()[name = tensor<string, []>("input_647_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> input_647_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_647_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(635663552)))];
            tensor<fp16, [1280]> input_647_beta_0_to_fp16 = const()[name = tensor<string, []>("input_647_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(635666176)))];
            tensor<fp16, []> var_50669_to_fp16 = const()[name = tensor<string, []>("op_50669_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> input_647_cast_fp16 = layer_norm(axes = input_647_axes_0, beta = input_647_beta_0_to_fp16, epsilon = var_50669_to_fp16, gamma = input_647_gamma_0_to_fp16, x = inputs_335_cast_fp16)[name = tensor<string, []>("input_647_cast_fp16")];
            tensor<string, []> var_50689_pad_type_0 = const()[name = tensor<string, []>("op_50689_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_50689_strides_0 = const()[name = tensor<string, []>("op_50689_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_50689_pad_0 = const()[name = tensor<string, []>("op_50689_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_50689_dilations_0 = const()[name = tensor<string, []>("op_50689_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_50689_groups_0 = const()[name = tensor<string, []>("op_50689_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_1_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [9830400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(635668800))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(645499264))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_1_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([10240, 1280, 1, 1])];
            tensor<fp16, [10240]> up_blocks_0_attentions_2_transformer_blocks_1_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_1_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(645499456)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_50689_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_1_ff_net_0_proj_bias_to_fp16, dilations = var_50689_dilations_0, groups = var_50689_groups_0, pad = var_50689_pad_0, pad_type = var_50689_pad_type_0, strides = var_50689_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_1_ff_net_0_proj_weight_to_fp16_palettized, x = input_647_cast_fp16)[name = tensor<string, []>("op_50689_cast_fp16")];
            tensor<int32, [2]> var_50690_split_sizes_0 = const()[name = tensor<string, []>("op_50690_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_50690_axis_0 = const()[name = tensor<string, []>("op_50690_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_50690_cast_fp16_0, tensor<fp16, [2, 5120, 1, 1024]> var_50690_cast_fp16_1 = split(axis = var_50690_axis_0, split_sizes = var_50690_split_sizes_0, x = var_50689_cast_fp16)[name = tensor<string, []>("op_50690_cast_fp16")];
            tensor<string, []> var_50692_mode_0 = const()[name = tensor<string, []>("op_50692_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_50692_cast_fp16 = gelu(mode = var_50692_mode_0, x = var_50690_cast_fp16_1)[name = tensor<string, []>("op_50692_cast_fp16")];
            tensor<fp16, [2, 5120, 1, 1024]> input_649_cast_fp16 = mul(x = var_50690_cast_fp16_0, y = var_50692_cast_fp16)[name = tensor<string, []>("input_649_cast_fp16")];
            tensor<string, []> var_50700_pad_type_0 = const()[name = tensor<string, []>("op_50700_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_50700_strides_0 = const()[name = tensor<string, []>("op_50700_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_50700_pad_0 = const()[name = tensor<string, []>("op_50700_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_50700_dilations_0 = const()[name = tensor<string, []>("op_50700_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_50700_groups_0 = const()[name = tensor<string, []>("op_50700_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_1_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [4915200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(645520000))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(650435264))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_1_ff_net_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 5120, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_1_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_1_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(650435456)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_50700_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_1_ff_net_2_bias_to_fp16, dilations = var_50700_dilations_0, groups = var_50700_groups_0, pad = var_50700_pad_0, pad_type = var_50700_pad_type_0, strides = var_50700_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_1_ff_net_2_weight_to_fp16_palettized, x = input_649_cast_fp16)[name = tensor<string, []>("op_50700_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_337_cast_fp16 = add(x = var_50700_cast_fp16, y = inputs_335_cast_fp16)[name = tensor<string, []>("inputs_337_cast_fp16")];
            tensor<int32, [1]> hidden_states_447_axes_0 = const()[name = tensor<string, []>("hidden_states_447_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_447_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_447_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(650438080)))];
            tensor<fp16, [1280]> hidden_states_447_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_447_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(650440704)))];
            tensor<fp16, []> var_50716_to_fp16 = const()[name = tensor<string, []>("op_50716_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_447_cast_fp16 = layer_norm(axes = hidden_states_447_axes_0, beta = hidden_states_447_beta_0_to_fp16, epsilon = var_50716_to_fp16, gamma = hidden_states_447_gamma_0_to_fp16, x = inputs_337_cast_fp16)[name = tensor<string, []>("hidden_states_447_cast_fp16")];
            tensor<string, []> q_225_pad_type_0 = const()[name = tensor<string, []>("q_225_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_225_strides_0 = const()[name = tensor<string, []>("q_225_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_225_pad_0 = const()[name = tensor<string, []>("q_225_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_225_dilations_0 = const()[name = tensor<string, []>("q_225_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_225_groups_0 = const()[name = tensor<string, []>("q_225_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_2_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(650443328))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(651672192))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_2_attn1_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_225_cast_fp16 = conv(dilations = q_225_dilations_0, groups = q_225_groups_0, pad = q_225_pad_0, pad_type = q_225_pad_type_0, strides = q_225_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_2_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_447_cast_fp16)[name = tensor<string, []>("q_225_cast_fp16")];
            tensor<string, []> k_449_pad_type_0 = const()[name = tensor<string, []>("k_449_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_449_strides_0 = const()[name = tensor<string, []>("k_449_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_449_pad_0 = const()[name = tensor<string, []>("k_449_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_449_dilations_0 = const()[name = tensor<string, []>("k_449_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_449_groups_0 = const()[name = tensor<string, []>("k_449_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_2_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(651672384))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(652901248))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_2_attn1_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> k_449_cast_fp16 = conv(dilations = k_449_dilations_0, groups = k_449_groups_0, pad = k_449_pad_0, pad_type = k_449_pad_type_0, strides = k_449_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_2_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_447_cast_fp16)[name = tensor<string, []>("k_449_cast_fp16")];
            tensor<string, []> v_225_pad_type_0 = const()[name = tensor<string, []>("v_225_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_225_strides_0 = const()[name = tensor<string, []>("v_225_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_225_pad_0 = const()[name = tensor<string, []>("v_225_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_225_dilations_0 = const()[name = tensor<string, []>("v_225_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_225_groups_0 = const()[name = tensor<string, []>("v_225_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_2_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(652901440))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(654130304))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_2_attn1_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> v_225_cast_fp16 = conv(dilations = v_225_dilations_0, groups = v_225_groups_0, pad = v_225_pad_0, pad_type = v_225_pad_type_0, strides = v_225_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_2_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_447_cast_fp16)[name = tensor<string, []>("v_225_cast_fp16")];
            tensor<int32, [4]> var_50749_begin_0 = const()[name = tensor<string, []>("op_50749_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_50749_end_0 = const()[name = tensor<string, []>("op_50749_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_50749_end_mask_0 = const()[name = tensor<string, []>("op_50749_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50749_cast_fp16 = slice_by_index(begin = var_50749_begin_0, end = var_50749_end_0, end_mask = var_50749_end_mask_0, x = q_225_cast_fp16)[name = tensor<string, []>("op_50749_cast_fp16")];
            tensor<int32, [4]> var_50753_begin_0 = const()[name = tensor<string, []>("op_50753_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_50753_end_0 = const()[name = tensor<string, []>("op_50753_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_50753_end_mask_0 = const()[name = tensor<string, []>("op_50753_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50753_cast_fp16 = slice_by_index(begin = var_50753_begin_0, end = var_50753_end_0, end_mask = var_50753_end_mask_0, x = q_225_cast_fp16)[name = tensor<string, []>("op_50753_cast_fp16")];
            tensor<int32, [4]> var_50757_begin_0 = const()[name = tensor<string, []>("op_50757_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_50757_end_0 = const()[name = tensor<string, []>("op_50757_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_50757_end_mask_0 = const()[name = tensor<string, []>("op_50757_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50757_cast_fp16 = slice_by_index(begin = var_50757_begin_0, end = var_50757_end_0, end_mask = var_50757_end_mask_0, x = q_225_cast_fp16)[name = tensor<string, []>("op_50757_cast_fp16")];
            tensor<int32, [4]> var_50761_begin_0 = const()[name = tensor<string, []>("op_50761_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_50761_end_0 = const()[name = tensor<string, []>("op_50761_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_50761_end_mask_0 = const()[name = tensor<string, []>("op_50761_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50761_cast_fp16 = slice_by_index(begin = var_50761_begin_0, end = var_50761_end_0, end_mask = var_50761_end_mask_0, x = q_225_cast_fp16)[name = tensor<string, []>("op_50761_cast_fp16")];
            tensor<int32, [4]> var_50765_begin_0 = const()[name = tensor<string, []>("op_50765_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_50765_end_0 = const()[name = tensor<string, []>("op_50765_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_50765_end_mask_0 = const()[name = tensor<string, []>("op_50765_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50765_cast_fp16 = slice_by_index(begin = var_50765_begin_0, end = var_50765_end_0, end_mask = var_50765_end_mask_0, x = q_225_cast_fp16)[name = tensor<string, []>("op_50765_cast_fp16")];
            tensor<int32, [4]> var_50769_begin_0 = const()[name = tensor<string, []>("op_50769_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_50769_end_0 = const()[name = tensor<string, []>("op_50769_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_50769_end_mask_0 = const()[name = tensor<string, []>("op_50769_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50769_cast_fp16 = slice_by_index(begin = var_50769_begin_0, end = var_50769_end_0, end_mask = var_50769_end_mask_0, x = q_225_cast_fp16)[name = tensor<string, []>("op_50769_cast_fp16")];
            tensor<int32, [4]> var_50773_begin_0 = const()[name = tensor<string, []>("op_50773_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_50773_end_0 = const()[name = tensor<string, []>("op_50773_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_50773_end_mask_0 = const()[name = tensor<string, []>("op_50773_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50773_cast_fp16 = slice_by_index(begin = var_50773_begin_0, end = var_50773_end_0, end_mask = var_50773_end_mask_0, x = q_225_cast_fp16)[name = tensor<string, []>("op_50773_cast_fp16")];
            tensor<int32, [4]> var_50777_begin_0 = const()[name = tensor<string, []>("op_50777_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_50777_end_0 = const()[name = tensor<string, []>("op_50777_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_50777_end_mask_0 = const()[name = tensor<string, []>("op_50777_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50777_cast_fp16 = slice_by_index(begin = var_50777_begin_0, end = var_50777_end_0, end_mask = var_50777_end_mask_0, x = q_225_cast_fp16)[name = tensor<string, []>("op_50777_cast_fp16")];
            tensor<int32, [4]> var_50781_begin_0 = const()[name = tensor<string, []>("op_50781_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_50781_end_0 = const()[name = tensor<string, []>("op_50781_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_50781_end_mask_0 = const()[name = tensor<string, []>("op_50781_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50781_cast_fp16 = slice_by_index(begin = var_50781_begin_0, end = var_50781_end_0, end_mask = var_50781_end_mask_0, x = q_225_cast_fp16)[name = tensor<string, []>("op_50781_cast_fp16")];
            tensor<int32, [4]> var_50785_begin_0 = const()[name = tensor<string, []>("op_50785_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_50785_end_0 = const()[name = tensor<string, []>("op_50785_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_50785_end_mask_0 = const()[name = tensor<string, []>("op_50785_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50785_cast_fp16 = slice_by_index(begin = var_50785_begin_0, end = var_50785_end_0, end_mask = var_50785_end_mask_0, x = q_225_cast_fp16)[name = tensor<string, []>("op_50785_cast_fp16")];
            tensor<int32, [4]> var_50789_begin_0 = const()[name = tensor<string, []>("op_50789_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_50789_end_0 = const()[name = tensor<string, []>("op_50789_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_50789_end_mask_0 = const()[name = tensor<string, []>("op_50789_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50789_cast_fp16 = slice_by_index(begin = var_50789_begin_0, end = var_50789_end_0, end_mask = var_50789_end_mask_0, x = q_225_cast_fp16)[name = tensor<string, []>("op_50789_cast_fp16")];
            tensor<int32, [4]> var_50793_begin_0 = const()[name = tensor<string, []>("op_50793_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_50793_end_0 = const()[name = tensor<string, []>("op_50793_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_50793_end_mask_0 = const()[name = tensor<string, []>("op_50793_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50793_cast_fp16 = slice_by_index(begin = var_50793_begin_0, end = var_50793_end_0, end_mask = var_50793_end_mask_0, x = q_225_cast_fp16)[name = tensor<string, []>("op_50793_cast_fp16")];
            tensor<int32, [4]> var_50797_begin_0 = const()[name = tensor<string, []>("op_50797_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_50797_end_0 = const()[name = tensor<string, []>("op_50797_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_50797_end_mask_0 = const()[name = tensor<string, []>("op_50797_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50797_cast_fp16 = slice_by_index(begin = var_50797_begin_0, end = var_50797_end_0, end_mask = var_50797_end_mask_0, x = q_225_cast_fp16)[name = tensor<string, []>("op_50797_cast_fp16")];
            tensor<int32, [4]> var_50801_begin_0 = const()[name = tensor<string, []>("op_50801_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_50801_end_0 = const()[name = tensor<string, []>("op_50801_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_50801_end_mask_0 = const()[name = tensor<string, []>("op_50801_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50801_cast_fp16 = slice_by_index(begin = var_50801_begin_0, end = var_50801_end_0, end_mask = var_50801_end_mask_0, x = q_225_cast_fp16)[name = tensor<string, []>("op_50801_cast_fp16")];
            tensor<int32, [4]> var_50805_begin_0 = const()[name = tensor<string, []>("op_50805_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_50805_end_0 = const()[name = tensor<string, []>("op_50805_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_50805_end_mask_0 = const()[name = tensor<string, []>("op_50805_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50805_cast_fp16 = slice_by_index(begin = var_50805_begin_0, end = var_50805_end_0, end_mask = var_50805_end_mask_0, x = q_225_cast_fp16)[name = tensor<string, []>("op_50805_cast_fp16")];
            tensor<int32, [4]> var_50809_begin_0 = const()[name = tensor<string, []>("op_50809_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_50809_end_0 = const()[name = tensor<string, []>("op_50809_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_50809_end_mask_0 = const()[name = tensor<string, []>("op_50809_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50809_cast_fp16 = slice_by_index(begin = var_50809_begin_0, end = var_50809_end_0, end_mask = var_50809_end_mask_0, x = q_225_cast_fp16)[name = tensor<string, []>("op_50809_cast_fp16")];
            tensor<int32, [4]> var_50813_begin_0 = const()[name = tensor<string, []>("op_50813_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_50813_end_0 = const()[name = tensor<string, []>("op_50813_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_50813_end_mask_0 = const()[name = tensor<string, []>("op_50813_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50813_cast_fp16 = slice_by_index(begin = var_50813_begin_0, end = var_50813_end_0, end_mask = var_50813_end_mask_0, x = q_225_cast_fp16)[name = tensor<string, []>("op_50813_cast_fp16")];
            tensor<int32, [4]> var_50817_begin_0 = const()[name = tensor<string, []>("op_50817_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_50817_end_0 = const()[name = tensor<string, []>("op_50817_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_50817_end_mask_0 = const()[name = tensor<string, []>("op_50817_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50817_cast_fp16 = slice_by_index(begin = var_50817_begin_0, end = var_50817_end_0, end_mask = var_50817_end_mask_0, x = q_225_cast_fp16)[name = tensor<string, []>("op_50817_cast_fp16")];
            tensor<int32, [4]> var_50821_begin_0 = const()[name = tensor<string, []>("op_50821_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_50821_end_0 = const()[name = tensor<string, []>("op_50821_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_50821_end_mask_0 = const()[name = tensor<string, []>("op_50821_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50821_cast_fp16 = slice_by_index(begin = var_50821_begin_0, end = var_50821_end_0, end_mask = var_50821_end_mask_0, x = q_225_cast_fp16)[name = tensor<string, []>("op_50821_cast_fp16")];
            tensor<int32, [4]> var_50825_begin_0 = const()[name = tensor<string, []>("op_50825_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_50825_end_0 = const()[name = tensor<string, []>("op_50825_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_50825_end_mask_0 = const()[name = tensor<string, []>("op_50825_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50825_cast_fp16 = slice_by_index(begin = var_50825_begin_0, end = var_50825_end_0, end_mask = var_50825_end_mask_0, x = q_225_cast_fp16)[name = tensor<string, []>("op_50825_cast_fp16")];
            tensor<int32, [4]> k_451_perm_0 = const()[name = tensor<string, []>("k_451_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_50832_begin_0 = const()[name = tensor<string, []>("op_50832_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_50832_end_0 = const()[name = tensor<string, []>("op_50832_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_50832_end_mask_0 = const()[name = tensor<string, []>("op_50832_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> k_451_cast_fp16 = transpose(perm = k_451_perm_0, x = k_449_cast_fp16)[name = tensor<string, []>("transpose_27")];
            tensor<fp16, [2, 1024, 1, 64]> var_50832_cast_fp16 = slice_by_index(begin = var_50832_begin_0, end = var_50832_end_0, end_mask = var_50832_end_mask_0, x = k_451_cast_fp16)[name = tensor<string, []>("op_50832_cast_fp16")];
            tensor<int32, [4]> var_50836_begin_0 = const()[name = tensor<string, []>("op_50836_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_50836_end_0 = const()[name = tensor<string, []>("op_50836_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_50836_end_mask_0 = const()[name = tensor<string, []>("op_50836_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_50836_cast_fp16 = slice_by_index(begin = var_50836_begin_0, end = var_50836_end_0, end_mask = var_50836_end_mask_0, x = k_451_cast_fp16)[name = tensor<string, []>("op_50836_cast_fp16")];
            tensor<int32, [4]> var_50840_begin_0 = const()[name = tensor<string, []>("op_50840_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_50840_end_0 = const()[name = tensor<string, []>("op_50840_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_50840_end_mask_0 = const()[name = tensor<string, []>("op_50840_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_50840_cast_fp16 = slice_by_index(begin = var_50840_begin_0, end = var_50840_end_0, end_mask = var_50840_end_mask_0, x = k_451_cast_fp16)[name = tensor<string, []>("op_50840_cast_fp16")];
            tensor<int32, [4]> var_50844_begin_0 = const()[name = tensor<string, []>("op_50844_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_50844_end_0 = const()[name = tensor<string, []>("op_50844_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_50844_end_mask_0 = const()[name = tensor<string, []>("op_50844_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_50844_cast_fp16 = slice_by_index(begin = var_50844_begin_0, end = var_50844_end_0, end_mask = var_50844_end_mask_0, x = k_451_cast_fp16)[name = tensor<string, []>("op_50844_cast_fp16")];
            tensor<int32, [4]> var_50848_begin_0 = const()[name = tensor<string, []>("op_50848_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_50848_end_0 = const()[name = tensor<string, []>("op_50848_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_50848_end_mask_0 = const()[name = tensor<string, []>("op_50848_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_50848_cast_fp16 = slice_by_index(begin = var_50848_begin_0, end = var_50848_end_0, end_mask = var_50848_end_mask_0, x = k_451_cast_fp16)[name = tensor<string, []>("op_50848_cast_fp16")];
            tensor<int32, [4]> var_50852_begin_0 = const()[name = tensor<string, []>("op_50852_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_50852_end_0 = const()[name = tensor<string, []>("op_50852_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_50852_end_mask_0 = const()[name = tensor<string, []>("op_50852_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_50852_cast_fp16 = slice_by_index(begin = var_50852_begin_0, end = var_50852_end_0, end_mask = var_50852_end_mask_0, x = k_451_cast_fp16)[name = tensor<string, []>("op_50852_cast_fp16")];
            tensor<int32, [4]> var_50856_begin_0 = const()[name = tensor<string, []>("op_50856_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_50856_end_0 = const()[name = tensor<string, []>("op_50856_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_50856_end_mask_0 = const()[name = tensor<string, []>("op_50856_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_50856_cast_fp16 = slice_by_index(begin = var_50856_begin_0, end = var_50856_end_0, end_mask = var_50856_end_mask_0, x = k_451_cast_fp16)[name = tensor<string, []>("op_50856_cast_fp16")];
            tensor<int32, [4]> var_50860_begin_0 = const()[name = tensor<string, []>("op_50860_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_50860_end_0 = const()[name = tensor<string, []>("op_50860_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_50860_end_mask_0 = const()[name = tensor<string, []>("op_50860_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_50860_cast_fp16 = slice_by_index(begin = var_50860_begin_0, end = var_50860_end_0, end_mask = var_50860_end_mask_0, x = k_451_cast_fp16)[name = tensor<string, []>("op_50860_cast_fp16")];
            tensor<int32, [4]> var_50864_begin_0 = const()[name = tensor<string, []>("op_50864_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_50864_end_0 = const()[name = tensor<string, []>("op_50864_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_50864_end_mask_0 = const()[name = tensor<string, []>("op_50864_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_50864_cast_fp16 = slice_by_index(begin = var_50864_begin_0, end = var_50864_end_0, end_mask = var_50864_end_mask_0, x = k_451_cast_fp16)[name = tensor<string, []>("op_50864_cast_fp16")];
            tensor<int32, [4]> var_50868_begin_0 = const()[name = tensor<string, []>("op_50868_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_50868_end_0 = const()[name = tensor<string, []>("op_50868_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_50868_end_mask_0 = const()[name = tensor<string, []>("op_50868_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_50868_cast_fp16 = slice_by_index(begin = var_50868_begin_0, end = var_50868_end_0, end_mask = var_50868_end_mask_0, x = k_451_cast_fp16)[name = tensor<string, []>("op_50868_cast_fp16")];
            tensor<int32, [4]> var_50872_begin_0 = const()[name = tensor<string, []>("op_50872_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_50872_end_0 = const()[name = tensor<string, []>("op_50872_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_50872_end_mask_0 = const()[name = tensor<string, []>("op_50872_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_50872_cast_fp16 = slice_by_index(begin = var_50872_begin_0, end = var_50872_end_0, end_mask = var_50872_end_mask_0, x = k_451_cast_fp16)[name = tensor<string, []>("op_50872_cast_fp16")];
            tensor<int32, [4]> var_50876_begin_0 = const()[name = tensor<string, []>("op_50876_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_50876_end_0 = const()[name = tensor<string, []>("op_50876_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_50876_end_mask_0 = const()[name = tensor<string, []>("op_50876_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_50876_cast_fp16 = slice_by_index(begin = var_50876_begin_0, end = var_50876_end_0, end_mask = var_50876_end_mask_0, x = k_451_cast_fp16)[name = tensor<string, []>("op_50876_cast_fp16")];
            tensor<int32, [4]> var_50880_begin_0 = const()[name = tensor<string, []>("op_50880_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_50880_end_0 = const()[name = tensor<string, []>("op_50880_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_50880_end_mask_0 = const()[name = tensor<string, []>("op_50880_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_50880_cast_fp16 = slice_by_index(begin = var_50880_begin_0, end = var_50880_end_0, end_mask = var_50880_end_mask_0, x = k_451_cast_fp16)[name = tensor<string, []>("op_50880_cast_fp16")];
            tensor<int32, [4]> var_50884_begin_0 = const()[name = tensor<string, []>("op_50884_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_50884_end_0 = const()[name = tensor<string, []>("op_50884_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_50884_end_mask_0 = const()[name = tensor<string, []>("op_50884_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_50884_cast_fp16 = slice_by_index(begin = var_50884_begin_0, end = var_50884_end_0, end_mask = var_50884_end_mask_0, x = k_451_cast_fp16)[name = tensor<string, []>("op_50884_cast_fp16")];
            tensor<int32, [4]> var_50888_begin_0 = const()[name = tensor<string, []>("op_50888_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_50888_end_0 = const()[name = tensor<string, []>("op_50888_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_50888_end_mask_0 = const()[name = tensor<string, []>("op_50888_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_50888_cast_fp16 = slice_by_index(begin = var_50888_begin_0, end = var_50888_end_0, end_mask = var_50888_end_mask_0, x = k_451_cast_fp16)[name = tensor<string, []>("op_50888_cast_fp16")];
            tensor<int32, [4]> var_50892_begin_0 = const()[name = tensor<string, []>("op_50892_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_50892_end_0 = const()[name = tensor<string, []>("op_50892_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_50892_end_mask_0 = const()[name = tensor<string, []>("op_50892_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_50892_cast_fp16 = slice_by_index(begin = var_50892_begin_0, end = var_50892_end_0, end_mask = var_50892_end_mask_0, x = k_451_cast_fp16)[name = tensor<string, []>("op_50892_cast_fp16")];
            tensor<int32, [4]> var_50896_begin_0 = const()[name = tensor<string, []>("op_50896_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_50896_end_0 = const()[name = tensor<string, []>("op_50896_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_50896_end_mask_0 = const()[name = tensor<string, []>("op_50896_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_50896_cast_fp16 = slice_by_index(begin = var_50896_begin_0, end = var_50896_end_0, end_mask = var_50896_end_mask_0, x = k_451_cast_fp16)[name = tensor<string, []>("op_50896_cast_fp16")];
            tensor<int32, [4]> var_50900_begin_0 = const()[name = tensor<string, []>("op_50900_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_50900_end_0 = const()[name = tensor<string, []>("op_50900_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_50900_end_mask_0 = const()[name = tensor<string, []>("op_50900_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_50900_cast_fp16 = slice_by_index(begin = var_50900_begin_0, end = var_50900_end_0, end_mask = var_50900_end_mask_0, x = k_451_cast_fp16)[name = tensor<string, []>("op_50900_cast_fp16")];
            tensor<int32, [4]> var_50904_begin_0 = const()[name = tensor<string, []>("op_50904_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_50904_end_0 = const()[name = tensor<string, []>("op_50904_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_50904_end_mask_0 = const()[name = tensor<string, []>("op_50904_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_50904_cast_fp16 = slice_by_index(begin = var_50904_begin_0, end = var_50904_end_0, end_mask = var_50904_end_mask_0, x = k_451_cast_fp16)[name = tensor<string, []>("op_50904_cast_fp16")];
            tensor<int32, [4]> var_50908_begin_0 = const()[name = tensor<string, []>("op_50908_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_50908_end_0 = const()[name = tensor<string, []>("op_50908_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_50908_end_mask_0 = const()[name = tensor<string, []>("op_50908_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_50908_cast_fp16 = slice_by_index(begin = var_50908_begin_0, end = var_50908_end_0, end_mask = var_50908_end_mask_0, x = k_451_cast_fp16)[name = tensor<string, []>("op_50908_cast_fp16")];
            tensor<int32, [4]> var_50910_begin_0 = const()[name = tensor<string, []>("op_50910_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_50910_end_0 = const()[name = tensor<string, []>("op_50910_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_50910_end_mask_0 = const()[name = tensor<string, []>("op_50910_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50910_cast_fp16 = slice_by_index(begin = var_50910_begin_0, end = var_50910_end_0, end_mask = var_50910_end_mask_0, x = v_225_cast_fp16)[name = tensor<string, []>("op_50910_cast_fp16")];
            tensor<int32, [4]> var_50914_begin_0 = const()[name = tensor<string, []>("op_50914_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_50914_end_0 = const()[name = tensor<string, []>("op_50914_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_50914_end_mask_0 = const()[name = tensor<string, []>("op_50914_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50914_cast_fp16 = slice_by_index(begin = var_50914_begin_0, end = var_50914_end_0, end_mask = var_50914_end_mask_0, x = v_225_cast_fp16)[name = tensor<string, []>("op_50914_cast_fp16")];
            tensor<int32, [4]> var_50918_begin_0 = const()[name = tensor<string, []>("op_50918_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_50918_end_0 = const()[name = tensor<string, []>("op_50918_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_50918_end_mask_0 = const()[name = tensor<string, []>("op_50918_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50918_cast_fp16 = slice_by_index(begin = var_50918_begin_0, end = var_50918_end_0, end_mask = var_50918_end_mask_0, x = v_225_cast_fp16)[name = tensor<string, []>("op_50918_cast_fp16")];
            tensor<int32, [4]> var_50922_begin_0 = const()[name = tensor<string, []>("op_50922_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_50922_end_0 = const()[name = tensor<string, []>("op_50922_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_50922_end_mask_0 = const()[name = tensor<string, []>("op_50922_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50922_cast_fp16 = slice_by_index(begin = var_50922_begin_0, end = var_50922_end_0, end_mask = var_50922_end_mask_0, x = v_225_cast_fp16)[name = tensor<string, []>("op_50922_cast_fp16")];
            tensor<int32, [4]> var_50926_begin_0 = const()[name = tensor<string, []>("op_50926_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_50926_end_0 = const()[name = tensor<string, []>("op_50926_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_50926_end_mask_0 = const()[name = tensor<string, []>("op_50926_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50926_cast_fp16 = slice_by_index(begin = var_50926_begin_0, end = var_50926_end_0, end_mask = var_50926_end_mask_0, x = v_225_cast_fp16)[name = tensor<string, []>("op_50926_cast_fp16")];
            tensor<int32, [4]> var_50930_begin_0 = const()[name = tensor<string, []>("op_50930_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_50930_end_0 = const()[name = tensor<string, []>("op_50930_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_50930_end_mask_0 = const()[name = tensor<string, []>("op_50930_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50930_cast_fp16 = slice_by_index(begin = var_50930_begin_0, end = var_50930_end_0, end_mask = var_50930_end_mask_0, x = v_225_cast_fp16)[name = tensor<string, []>("op_50930_cast_fp16")];
            tensor<int32, [4]> var_50934_begin_0 = const()[name = tensor<string, []>("op_50934_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_50934_end_0 = const()[name = tensor<string, []>("op_50934_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_50934_end_mask_0 = const()[name = tensor<string, []>("op_50934_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50934_cast_fp16 = slice_by_index(begin = var_50934_begin_0, end = var_50934_end_0, end_mask = var_50934_end_mask_0, x = v_225_cast_fp16)[name = tensor<string, []>("op_50934_cast_fp16")];
            tensor<int32, [4]> var_50938_begin_0 = const()[name = tensor<string, []>("op_50938_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_50938_end_0 = const()[name = tensor<string, []>("op_50938_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_50938_end_mask_0 = const()[name = tensor<string, []>("op_50938_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50938_cast_fp16 = slice_by_index(begin = var_50938_begin_0, end = var_50938_end_0, end_mask = var_50938_end_mask_0, x = v_225_cast_fp16)[name = tensor<string, []>("op_50938_cast_fp16")];
            tensor<int32, [4]> var_50942_begin_0 = const()[name = tensor<string, []>("op_50942_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_50942_end_0 = const()[name = tensor<string, []>("op_50942_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_50942_end_mask_0 = const()[name = tensor<string, []>("op_50942_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50942_cast_fp16 = slice_by_index(begin = var_50942_begin_0, end = var_50942_end_0, end_mask = var_50942_end_mask_0, x = v_225_cast_fp16)[name = tensor<string, []>("op_50942_cast_fp16")];
            tensor<int32, [4]> var_50946_begin_0 = const()[name = tensor<string, []>("op_50946_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_50946_end_0 = const()[name = tensor<string, []>("op_50946_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_50946_end_mask_0 = const()[name = tensor<string, []>("op_50946_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50946_cast_fp16 = slice_by_index(begin = var_50946_begin_0, end = var_50946_end_0, end_mask = var_50946_end_mask_0, x = v_225_cast_fp16)[name = tensor<string, []>("op_50946_cast_fp16")];
            tensor<int32, [4]> var_50950_begin_0 = const()[name = tensor<string, []>("op_50950_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_50950_end_0 = const()[name = tensor<string, []>("op_50950_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_50950_end_mask_0 = const()[name = tensor<string, []>("op_50950_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50950_cast_fp16 = slice_by_index(begin = var_50950_begin_0, end = var_50950_end_0, end_mask = var_50950_end_mask_0, x = v_225_cast_fp16)[name = tensor<string, []>("op_50950_cast_fp16")];
            tensor<int32, [4]> var_50954_begin_0 = const()[name = tensor<string, []>("op_50954_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_50954_end_0 = const()[name = tensor<string, []>("op_50954_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_50954_end_mask_0 = const()[name = tensor<string, []>("op_50954_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50954_cast_fp16 = slice_by_index(begin = var_50954_begin_0, end = var_50954_end_0, end_mask = var_50954_end_mask_0, x = v_225_cast_fp16)[name = tensor<string, []>("op_50954_cast_fp16")];
            tensor<int32, [4]> var_50958_begin_0 = const()[name = tensor<string, []>("op_50958_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_50958_end_0 = const()[name = tensor<string, []>("op_50958_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_50958_end_mask_0 = const()[name = tensor<string, []>("op_50958_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50958_cast_fp16 = slice_by_index(begin = var_50958_begin_0, end = var_50958_end_0, end_mask = var_50958_end_mask_0, x = v_225_cast_fp16)[name = tensor<string, []>("op_50958_cast_fp16")];
            tensor<int32, [4]> var_50962_begin_0 = const()[name = tensor<string, []>("op_50962_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_50962_end_0 = const()[name = tensor<string, []>("op_50962_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_50962_end_mask_0 = const()[name = tensor<string, []>("op_50962_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50962_cast_fp16 = slice_by_index(begin = var_50962_begin_0, end = var_50962_end_0, end_mask = var_50962_end_mask_0, x = v_225_cast_fp16)[name = tensor<string, []>("op_50962_cast_fp16")];
            tensor<int32, [4]> var_50966_begin_0 = const()[name = tensor<string, []>("op_50966_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_50966_end_0 = const()[name = tensor<string, []>("op_50966_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_50966_end_mask_0 = const()[name = tensor<string, []>("op_50966_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50966_cast_fp16 = slice_by_index(begin = var_50966_begin_0, end = var_50966_end_0, end_mask = var_50966_end_mask_0, x = v_225_cast_fp16)[name = tensor<string, []>("op_50966_cast_fp16")];
            tensor<int32, [4]> var_50970_begin_0 = const()[name = tensor<string, []>("op_50970_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_50970_end_0 = const()[name = tensor<string, []>("op_50970_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_50970_end_mask_0 = const()[name = tensor<string, []>("op_50970_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50970_cast_fp16 = slice_by_index(begin = var_50970_begin_0, end = var_50970_end_0, end_mask = var_50970_end_mask_0, x = v_225_cast_fp16)[name = tensor<string, []>("op_50970_cast_fp16")];
            tensor<int32, [4]> var_50974_begin_0 = const()[name = tensor<string, []>("op_50974_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_50974_end_0 = const()[name = tensor<string, []>("op_50974_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_50974_end_mask_0 = const()[name = tensor<string, []>("op_50974_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50974_cast_fp16 = slice_by_index(begin = var_50974_begin_0, end = var_50974_end_0, end_mask = var_50974_end_mask_0, x = v_225_cast_fp16)[name = tensor<string, []>("op_50974_cast_fp16")];
            tensor<int32, [4]> var_50978_begin_0 = const()[name = tensor<string, []>("op_50978_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_50978_end_0 = const()[name = tensor<string, []>("op_50978_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_50978_end_mask_0 = const()[name = tensor<string, []>("op_50978_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50978_cast_fp16 = slice_by_index(begin = var_50978_begin_0, end = var_50978_end_0, end_mask = var_50978_end_mask_0, x = v_225_cast_fp16)[name = tensor<string, []>("op_50978_cast_fp16")];
            tensor<int32, [4]> var_50982_begin_0 = const()[name = tensor<string, []>("op_50982_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_50982_end_0 = const()[name = tensor<string, []>("op_50982_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_50982_end_mask_0 = const()[name = tensor<string, []>("op_50982_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50982_cast_fp16 = slice_by_index(begin = var_50982_begin_0, end = var_50982_end_0, end_mask = var_50982_end_mask_0, x = v_225_cast_fp16)[name = tensor<string, []>("op_50982_cast_fp16")];
            tensor<int32, [4]> var_50986_begin_0 = const()[name = tensor<string, []>("op_50986_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_50986_end_0 = const()[name = tensor<string, []>("op_50986_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_50986_end_mask_0 = const()[name = tensor<string, []>("op_50986_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50986_cast_fp16 = slice_by_index(begin = var_50986_begin_0, end = var_50986_end_0, end_mask = var_50986_end_mask_0, x = v_225_cast_fp16)[name = tensor<string, []>("op_50986_cast_fp16")];
            tensor<string, []> var_50990_equation_0 = const()[name = tensor<string, []>("op_50990_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50990_cast_fp16 = einsum(equation = var_50990_equation_0, values = (var_50832_cast_fp16, var_50749_cast_fp16))[name = tensor<string, []>("op_50990_cast_fp16")];
            tensor<fp16, []> var_50991_to_fp16 = const()[name = tensor<string, []>("op_50991_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4321_cast_fp16 = mul(x = var_50990_cast_fp16, y = var_50991_to_fp16)[name = tensor<string, []>("aw_4321_cast_fp16")];
            tensor<string, []> var_50994_equation_0 = const()[name = tensor<string, []>("op_50994_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50994_cast_fp16 = einsum(equation = var_50994_equation_0, values = (var_50836_cast_fp16, var_50753_cast_fp16))[name = tensor<string, []>("op_50994_cast_fp16")];
            tensor<fp16, []> var_50995_to_fp16 = const()[name = tensor<string, []>("op_50995_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4323_cast_fp16 = mul(x = var_50994_cast_fp16, y = var_50995_to_fp16)[name = tensor<string, []>("aw_4323_cast_fp16")];
            tensor<string, []> var_50998_equation_0 = const()[name = tensor<string, []>("op_50998_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50998_cast_fp16 = einsum(equation = var_50998_equation_0, values = (var_50840_cast_fp16, var_50757_cast_fp16))[name = tensor<string, []>("op_50998_cast_fp16")];
            tensor<fp16, []> var_50999_to_fp16 = const()[name = tensor<string, []>("op_50999_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4325_cast_fp16 = mul(x = var_50998_cast_fp16, y = var_50999_to_fp16)[name = tensor<string, []>("aw_4325_cast_fp16")];
            tensor<string, []> var_51002_equation_0 = const()[name = tensor<string, []>("op_51002_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51002_cast_fp16 = einsum(equation = var_51002_equation_0, values = (var_50844_cast_fp16, var_50761_cast_fp16))[name = tensor<string, []>("op_51002_cast_fp16")];
            tensor<fp16, []> var_51003_to_fp16 = const()[name = tensor<string, []>("op_51003_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4327_cast_fp16 = mul(x = var_51002_cast_fp16, y = var_51003_to_fp16)[name = tensor<string, []>("aw_4327_cast_fp16")];
            tensor<string, []> var_51006_equation_0 = const()[name = tensor<string, []>("op_51006_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51006_cast_fp16 = einsum(equation = var_51006_equation_0, values = (var_50848_cast_fp16, var_50765_cast_fp16))[name = tensor<string, []>("op_51006_cast_fp16")];
            tensor<fp16, []> var_51007_to_fp16 = const()[name = tensor<string, []>("op_51007_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4329_cast_fp16 = mul(x = var_51006_cast_fp16, y = var_51007_to_fp16)[name = tensor<string, []>("aw_4329_cast_fp16")];
            tensor<string, []> var_51010_equation_0 = const()[name = tensor<string, []>("op_51010_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51010_cast_fp16 = einsum(equation = var_51010_equation_0, values = (var_50852_cast_fp16, var_50769_cast_fp16))[name = tensor<string, []>("op_51010_cast_fp16")];
            tensor<fp16, []> var_51011_to_fp16 = const()[name = tensor<string, []>("op_51011_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4331_cast_fp16 = mul(x = var_51010_cast_fp16, y = var_51011_to_fp16)[name = tensor<string, []>("aw_4331_cast_fp16")];
            tensor<string, []> var_51014_equation_0 = const()[name = tensor<string, []>("op_51014_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51014_cast_fp16 = einsum(equation = var_51014_equation_0, values = (var_50856_cast_fp16, var_50773_cast_fp16))[name = tensor<string, []>("op_51014_cast_fp16")];
            tensor<fp16, []> var_51015_to_fp16 = const()[name = tensor<string, []>("op_51015_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4333_cast_fp16 = mul(x = var_51014_cast_fp16, y = var_51015_to_fp16)[name = tensor<string, []>("aw_4333_cast_fp16")];
            tensor<string, []> var_51018_equation_0 = const()[name = tensor<string, []>("op_51018_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51018_cast_fp16 = einsum(equation = var_51018_equation_0, values = (var_50860_cast_fp16, var_50777_cast_fp16))[name = tensor<string, []>("op_51018_cast_fp16")];
            tensor<fp16, []> var_51019_to_fp16 = const()[name = tensor<string, []>("op_51019_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4335_cast_fp16 = mul(x = var_51018_cast_fp16, y = var_51019_to_fp16)[name = tensor<string, []>("aw_4335_cast_fp16")];
            tensor<string, []> var_51022_equation_0 = const()[name = tensor<string, []>("op_51022_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51022_cast_fp16 = einsum(equation = var_51022_equation_0, values = (var_50864_cast_fp16, var_50781_cast_fp16))[name = tensor<string, []>("op_51022_cast_fp16")];
            tensor<fp16, []> var_51023_to_fp16 = const()[name = tensor<string, []>("op_51023_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4337_cast_fp16 = mul(x = var_51022_cast_fp16, y = var_51023_to_fp16)[name = tensor<string, []>("aw_4337_cast_fp16")];
            tensor<string, []> var_51026_equation_0 = const()[name = tensor<string, []>("op_51026_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51026_cast_fp16 = einsum(equation = var_51026_equation_0, values = (var_50868_cast_fp16, var_50785_cast_fp16))[name = tensor<string, []>("op_51026_cast_fp16")];
            tensor<fp16, []> var_51027_to_fp16 = const()[name = tensor<string, []>("op_51027_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4339_cast_fp16 = mul(x = var_51026_cast_fp16, y = var_51027_to_fp16)[name = tensor<string, []>("aw_4339_cast_fp16")];
            tensor<string, []> var_51030_equation_0 = const()[name = tensor<string, []>("op_51030_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51030_cast_fp16 = einsum(equation = var_51030_equation_0, values = (var_50872_cast_fp16, var_50789_cast_fp16))[name = tensor<string, []>("op_51030_cast_fp16")];
            tensor<fp16, []> var_51031_to_fp16 = const()[name = tensor<string, []>("op_51031_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4341_cast_fp16 = mul(x = var_51030_cast_fp16, y = var_51031_to_fp16)[name = tensor<string, []>("aw_4341_cast_fp16")];
            tensor<string, []> var_51034_equation_0 = const()[name = tensor<string, []>("op_51034_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51034_cast_fp16 = einsum(equation = var_51034_equation_0, values = (var_50876_cast_fp16, var_50793_cast_fp16))[name = tensor<string, []>("op_51034_cast_fp16")];
            tensor<fp16, []> var_51035_to_fp16 = const()[name = tensor<string, []>("op_51035_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4343_cast_fp16 = mul(x = var_51034_cast_fp16, y = var_51035_to_fp16)[name = tensor<string, []>("aw_4343_cast_fp16")];
            tensor<string, []> var_51038_equation_0 = const()[name = tensor<string, []>("op_51038_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51038_cast_fp16 = einsum(equation = var_51038_equation_0, values = (var_50880_cast_fp16, var_50797_cast_fp16))[name = tensor<string, []>("op_51038_cast_fp16")];
            tensor<fp16, []> var_51039_to_fp16 = const()[name = tensor<string, []>("op_51039_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4345_cast_fp16 = mul(x = var_51038_cast_fp16, y = var_51039_to_fp16)[name = tensor<string, []>("aw_4345_cast_fp16")];
            tensor<string, []> var_51042_equation_0 = const()[name = tensor<string, []>("op_51042_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51042_cast_fp16 = einsum(equation = var_51042_equation_0, values = (var_50884_cast_fp16, var_50801_cast_fp16))[name = tensor<string, []>("op_51042_cast_fp16")];
            tensor<fp16, []> var_51043_to_fp16 = const()[name = tensor<string, []>("op_51043_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4347_cast_fp16 = mul(x = var_51042_cast_fp16, y = var_51043_to_fp16)[name = tensor<string, []>("aw_4347_cast_fp16")];
            tensor<string, []> var_51046_equation_0 = const()[name = tensor<string, []>("op_51046_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51046_cast_fp16 = einsum(equation = var_51046_equation_0, values = (var_50888_cast_fp16, var_50805_cast_fp16))[name = tensor<string, []>("op_51046_cast_fp16")];
            tensor<fp16, []> var_51047_to_fp16 = const()[name = tensor<string, []>("op_51047_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4349_cast_fp16 = mul(x = var_51046_cast_fp16, y = var_51047_to_fp16)[name = tensor<string, []>("aw_4349_cast_fp16")];
            tensor<string, []> var_51050_equation_0 = const()[name = tensor<string, []>("op_51050_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51050_cast_fp16 = einsum(equation = var_51050_equation_0, values = (var_50892_cast_fp16, var_50809_cast_fp16))[name = tensor<string, []>("op_51050_cast_fp16")];
            tensor<fp16, []> var_51051_to_fp16 = const()[name = tensor<string, []>("op_51051_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4351_cast_fp16 = mul(x = var_51050_cast_fp16, y = var_51051_to_fp16)[name = tensor<string, []>("aw_4351_cast_fp16")];
            tensor<string, []> var_51054_equation_0 = const()[name = tensor<string, []>("op_51054_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51054_cast_fp16 = einsum(equation = var_51054_equation_0, values = (var_50896_cast_fp16, var_50813_cast_fp16))[name = tensor<string, []>("op_51054_cast_fp16")];
            tensor<fp16, []> var_51055_to_fp16 = const()[name = tensor<string, []>("op_51055_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4353_cast_fp16 = mul(x = var_51054_cast_fp16, y = var_51055_to_fp16)[name = tensor<string, []>("aw_4353_cast_fp16")];
            tensor<string, []> var_51058_equation_0 = const()[name = tensor<string, []>("op_51058_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51058_cast_fp16 = einsum(equation = var_51058_equation_0, values = (var_50900_cast_fp16, var_50817_cast_fp16))[name = tensor<string, []>("op_51058_cast_fp16")];
            tensor<fp16, []> var_51059_to_fp16 = const()[name = tensor<string, []>("op_51059_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4355_cast_fp16 = mul(x = var_51058_cast_fp16, y = var_51059_to_fp16)[name = tensor<string, []>("aw_4355_cast_fp16")];
            tensor<string, []> var_51062_equation_0 = const()[name = tensor<string, []>("op_51062_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51062_cast_fp16 = einsum(equation = var_51062_equation_0, values = (var_50904_cast_fp16, var_50821_cast_fp16))[name = tensor<string, []>("op_51062_cast_fp16")];
            tensor<fp16, []> var_51063_to_fp16 = const()[name = tensor<string, []>("op_51063_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4357_cast_fp16 = mul(x = var_51062_cast_fp16, y = var_51063_to_fp16)[name = tensor<string, []>("aw_4357_cast_fp16")];
            tensor<string, []> var_51066_equation_0 = const()[name = tensor<string, []>("op_51066_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51066_cast_fp16 = einsum(equation = var_51066_equation_0, values = (var_50908_cast_fp16, var_50825_cast_fp16))[name = tensor<string, []>("op_51066_cast_fp16")];
            tensor<fp16, []> var_51067_to_fp16 = const()[name = tensor<string, []>("op_51067_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4359_cast_fp16 = mul(x = var_51066_cast_fp16, y = var_51067_to_fp16)[name = tensor<string, []>("aw_4359_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51069_cast_fp16 = softmax(axis = var_30355, x = aw_4321_cast_fp16)[name = tensor<string, []>("op_51069_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51070_cast_fp16 = softmax(axis = var_30355, x = aw_4323_cast_fp16)[name = tensor<string, []>("op_51070_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51071_cast_fp16 = softmax(axis = var_30355, x = aw_4325_cast_fp16)[name = tensor<string, []>("op_51071_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51072_cast_fp16 = softmax(axis = var_30355, x = aw_4327_cast_fp16)[name = tensor<string, []>("op_51072_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51073_cast_fp16 = softmax(axis = var_30355, x = aw_4329_cast_fp16)[name = tensor<string, []>("op_51073_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51074_cast_fp16 = softmax(axis = var_30355, x = aw_4331_cast_fp16)[name = tensor<string, []>("op_51074_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51075_cast_fp16 = softmax(axis = var_30355, x = aw_4333_cast_fp16)[name = tensor<string, []>("op_51075_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51076_cast_fp16 = softmax(axis = var_30355, x = aw_4335_cast_fp16)[name = tensor<string, []>("op_51076_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51077_cast_fp16 = softmax(axis = var_30355, x = aw_4337_cast_fp16)[name = tensor<string, []>("op_51077_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51078_cast_fp16 = softmax(axis = var_30355, x = aw_4339_cast_fp16)[name = tensor<string, []>("op_51078_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51079_cast_fp16 = softmax(axis = var_30355, x = aw_4341_cast_fp16)[name = tensor<string, []>("op_51079_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51080_cast_fp16 = softmax(axis = var_30355, x = aw_4343_cast_fp16)[name = tensor<string, []>("op_51080_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51081_cast_fp16 = softmax(axis = var_30355, x = aw_4345_cast_fp16)[name = tensor<string, []>("op_51081_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51082_cast_fp16 = softmax(axis = var_30355, x = aw_4347_cast_fp16)[name = tensor<string, []>("op_51082_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51083_cast_fp16 = softmax(axis = var_30355, x = aw_4349_cast_fp16)[name = tensor<string, []>("op_51083_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51084_cast_fp16 = softmax(axis = var_30355, x = aw_4351_cast_fp16)[name = tensor<string, []>("op_51084_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51085_cast_fp16 = softmax(axis = var_30355, x = aw_4353_cast_fp16)[name = tensor<string, []>("op_51085_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51086_cast_fp16 = softmax(axis = var_30355, x = aw_4355_cast_fp16)[name = tensor<string, []>("op_51086_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51087_cast_fp16 = softmax(axis = var_30355, x = aw_4357_cast_fp16)[name = tensor<string, []>("op_51087_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51088_cast_fp16 = softmax(axis = var_30355, x = aw_4359_cast_fp16)[name = tensor<string, []>("op_51088_cast_fp16")];
            tensor<string, []> var_51090_equation_0 = const()[name = tensor<string, []>("op_51090_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51090_cast_fp16 = einsum(equation = var_51090_equation_0, values = (var_50910_cast_fp16, var_51069_cast_fp16))[name = tensor<string, []>("op_51090_cast_fp16")];
            tensor<string, []> var_51092_equation_0 = const()[name = tensor<string, []>("op_51092_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51092_cast_fp16 = einsum(equation = var_51092_equation_0, values = (var_50914_cast_fp16, var_51070_cast_fp16))[name = tensor<string, []>("op_51092_cast_fp16")];
            tensor<string, []> var_51094_equation_0 = const()[name = tensor<string, []>("op_51094_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51094_cast_fp16 = einsum(equation = var_51094_equation_0, values = (var_50918_cast_fp16, var_51071_cast_fp16))[name = tensor<string, []>("op_51094_cast_fp16")];
            tensor<string, []> var_51096_equation_0 = const()[name = tensor<string, []>("op_51096_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51096_cast_fp16 = einsum(equation = var_51096_equation_0, values = (var_50922_cast_fp16, var_51072_cast_fp16))[name = tensor<string, []>("op_51096_cast_fp16")];
            tensor<string, []> var_51098_equation_0 = const()[name = tensor<string, []>("op_51098_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51098_cast_fp16 = einsum(equation = var_51098_equation_0, values = (var_50926_cast_fp16, var_51073_cast_fp16))[name = tensor<string, []>("op_51098_cast_fp16")];
            tensor<string, []> var_51100_equation_0 = const()[name = tensor<string, []>("op_51100_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51100_cast_fp16 = einsum(equation = var_51100_equation_0, values = (var_50930_cast_fp16, var_51074_cast_fp16))[name = tensor<string, []>("op_51100_cast_fp16")];
            tensor<string, []> var_51102_equation_0 = const()[name = tensor<string, []>("op_51102_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51102_cast_fp16 = einsum(equation = var_51102_equation_0, values = (var_50934_cast_fp16, var_51075_cast_fp16))[name = tensor<string, []>("op_51102_cast_fp16")];
            tensor<string, []> var_51104_equation_0 = const()[name = tensor<string, []>("op_51104_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51104_cast_fp16 = einsum(equation = var_51104_equation_0, values = (var_50938_cast_fp16, var_51076_cast_fp16))[name = tensor<string, []>("op_51104_cast_fp16")];
            tensor<string, []> var_51106_equation_0 = const()[name = tensor<string, []>("op_51106_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51106_cast_fp16 = einsum(equation = var_51106_equation_0, values = (var_50942_cast_fp16, var_51077_cast_fp16))[name = tensor<string, []>("op_51106_cast_fp16")];
            tensor<string, []> var_51108_equation_0 = const()[name = tensor<string, []>("op_51108_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51108_cast_fp16 = einsum(equation = var_51108_equation_0, values = (var_50946_cast_fp16, var_51078_cast_fp16))[name = tensor<string, []>("op_51108_cast_fp16")];
            tensor<string, []> var_51110_equation_0 = const()[name = tensor<string, []>("op_51110_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51110_cast_fp16 = einsum(equation = var_51110_equation_0, values = (var_50950_cast_fp16, var_51079_cast_fp16))[name = tensor<string, []>("op_51110_cast_fp16")];
            tensor<string, []> var_51112_equation_0 = const()[name = tensor<string, []>("op_51112_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51112_cast_fp16 = einsum(equation = var_51112_equation_0, values = (var_50954_cast_fp16, var_51080_cast_fp16))[name = tensor<string, []>("op_51112_cast_fp16")];
            tensor<string, []> var_51114_equation_0 = const()[name = tensor<string, []>("op_51114_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51114_cast_fp16 = einsum(equation = var_51114_equation_0, values = (var_50958_cast_fp16, var_51081_cast_fp16))[name = tensor<string, []>("op_51114_cast_fp16")];
            tensor<string, []> var_51116_equation_0 = const()[name = tensor<string, []>("op_51116_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51116_cast_fp16 = einsum(equation = var_51116_equation_0, values = (var_50962_cast_fp16, var_51082_cast_fp16))[name = tensor<string, []>("op_51116_cast_fp16")];
            tensor<string, []> var_51118_equation_0 = const()[name = tensor<string, []>("op_51118_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51118_cast_fp16 = einsum(equation = var_51118_equation_0, values = (var_50966_cast_fp16, var_51083_cast_fp16))[name = tensor<string, []>("op_51118_cast_fp16")];
            tensor<string, []> var_51120_equation_0 = const()[name = tensor<string, []>("op_51120_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51120_cast_fp16 = einsum(equation = var_51120_equation_0, values = (var_50970_cast_fp16, var_51084_cast_fp16))[name = tensor<string, []>("op_51120_cast_fp16")];
            tensor<string, []> var_51122_equation_0 = const()[name = tensor<string, []>("op_51122_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51122_cast_fp16 = einsum(equation = var_51122_equation_0, values = (var_50974_cast_fp16, var_51085_cast_fp16))[name = tensor<string, []>("op_51122_cast_fp16")];
            tensor<string, []> var_51124_equation_0 = const()[name = tensor<string, []>("op_51124_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51124_cast_fp16 = einsum(equation = var_51124_equation_0, values = (var_50978_cast_fp16, var_51086_cast_fp16))[name = tensor<string, []>("op_51124_cast_fp16")];
            tensor<string, []> var_51126_equation_0 = const()[name = tensor<string, []>("op_51126_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51126_cast_fp16 = einsum(equation = var_51126_equation_0, values = (var_50982_cast_fp16, var_51087_cast_fp16))[name = tensor<string, []>("op_51126_cast_fp16")];
            tensor<string, []> var_51128_equation_0 = const()[name = tensor<string, []>("op_51128_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51128_cast_fp16 = einsum(equation = var_51128_equation_0, values = (var_50986_cast_fp16, var_51088_cast_fp16))[name = tensor<string, []>("op_51128_cast_fp16")];
            tensor<bool, []> input_651_interleave_0 = const()[name = tensor<string, []>("input_651_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_651_cast_fp16 = concat(axis = var_30355, interleave = input_651_interleave_0, values = (var_51090_cast_fp16, var_51092_cast_fp16, var_51094_cast_fp16, var_51096_cast_fp16, var_51098_cast_fp16, var_51100_cast_fp16, var_51102_cast_fp16, var_51104_cast_fp16, var_51106_cast_fp16, var_51108_cast_fp16, var_51110_cast_fp16, var_51112_cast_fp16, var_51114_cast_fp16, var_51116_cast_fp16, var_51118_cast_fp16, var_51120_cast_fp16, var_51122_cast_fp16, var_51124_cast_fp16, var_51126_cast_fp16, var_51128_cast_fp16))[name = tensor<string, []>("input_651_cast_fp16")];
            tensor<string, []> var_51138_pad_type_0 = const()[name = tensor<string, []>("op_51138_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_51138_strides_0 = const()[name = tensor<string, []>("op_51138_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_51138_pad_0 = const()[name = tensor<string, []>("op_51138_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_51138_dilations_0 = const()[name = tensor<string, []>("op_51138_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_51138_groups_0 = const()[name = tensor<string, []>("op_51138_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_2_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(654130496))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(655359360))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_2_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_2_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_2_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(655359552)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_51138_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_2_attn1_to_out_0_bias_to_fp16, dilations = var_51138_dilations_0, groups = var_51138_groups_0, pad = var_51138_pad_0, pad_type = var_51138_pad_type_0, strides = var_51138_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_2_attn1_to_out_0_weight_to_fp16_palettized, x = input_651_cast_fp16)[name = tensor<string, []>("op_51138_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_339_cast_fp16 = add(x = var_51138_cast_fp16, y = inputs_337_cast_fp16)[name = tensor<string, []>("inputs_339_cast_fp16")];
            tensor<int32, [1]> hidden_states_449_axes_0 = const()[name = tensor<string, []>("hidden_states_449_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_449_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_449_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(655362176)))];
            tensor<fp16, [1280]> hidden_states_449_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_449_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(655364800)))];
            tensor<fp16, []> var_51148_to_fp16 = const()[name = tensor<string, []>("op_51148_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_449_cast_fp16 = layer_norm(axes = hidden_states_449_axes_0, beta = hidden_states_449_beta_0_to_fp16, epsilon = var_51148_to_fp16, gamma = hidden_states_449_gamma_0_to_fp16, x = inputs_339_cast_fp16)[name = tensor<string, []>("hidden_states_449_cast_fp16")];
            tensor<string, []> q_227_pad_type_0 = const()[name = tensor<string, []>("q_227_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_227_strides_0 = const()[name = tensor<string, []>("q_227_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_227_pad_0 = const()[name = tensor<string, []>("q_227_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_227_dilations_0 = const()[name = tensor<string, []>("q_227_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_227_groups_0 = const()[name = tensor<string, []>("q_227_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_2_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(655367424))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(656596288))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_2_attn2_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_227_cast_fp16 = conv(dilations = q_227_dilations_0, groups = q_227_groups_0, pad = q_227_pad_0, pad_type = q_227_pad_type_0, strides = q_227_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_2_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_449_cast_fp16)[name = tensor<string, []>("q_227_cast_fp16")];
            tensor<string, []> k_453_pad_type_0 = const()[name = tensor<string, []>("k_453_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_453_strides_0 = const()[name = tensor<string, []>("k_453_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_453_pad_0 = const()[name = tensor<string, []>("k_453_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_453_dilations_0 = const()[name = tensor<string, []>("k_453_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_453_groups_0 = const()[name = tensor<string, []>("k_453_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_2_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(656596480))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(658562624))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_2_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> k_453_cast_fp16 = conv(dilations = k_453_dilations_0, groups = k_453_groups_0, pad = k_453_pad_0, pad_type = k_453_pad_type_0, strides = k_453_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_2_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("k_453_cast_fp16")];
            tensor<string, []> v_227_pad_type_0 = const()[name = tensor<string, []>("v_227_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_227_strides_0 = const()[name = tensor<string, []>("v_227_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_227_pad_0 = const()[name = tensor<string, []>("v_227_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_227_dilations_0 = const()[name = tensor<string, []>("v_227_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_227_groups_0 = const()[name = tensor<string, []>("v_227_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_2_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(658562816))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(660528960))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_2_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> v_227_cast_fp16 = conv(dilations = v_227_dilations_0, groups = v_227_groups_0, pad = v_227_pad_0, pad_type = v_227_pad_type_0, strides = v_227_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_2_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("v_227_cast_fp16")];
            tensor<int32, [4]> var_51181_begin_0 = const()[name = tensor<string, []>("op_51181_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_51181_end_0 = const()[name = tensor<string, []>("op_51181_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_51181_end_mask_0 = const()[name = tensor<string, []>("op_51181_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51181_cast_fp16 = slice_by_index(begin = var_51181_begin_0, end = var_51181_end_0, end_mask = var_51181_end_mask_0, x = q_227_cast_fp16)[name = tensor<string, []>("op_51181_cast_fp16")];
            tensor<int32, [4]> var_51185_begin_0 = const()[name = tensor<string, []>("op_51185_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_51185_end_0 = const()[name = tensor<string, []>("op_51185_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_51185_end_mask_0 = const()[name = tensor<string, []>("op_51185_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51185_cast_fp16 = slice_by_index(begin = var_51185_begin_0, end = var_51185_end_0, end_mask = var_51185_end_mask_0, x = q_227_cast_fp16)[name = tensor<string, []>("op_51185_cast_fp16")];
            tensor<int32, [4]> var_51189_begin_0 = const()[name = tensor<string, []>("op_51189_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_51189_end_0 = const()[name = tensor<string, []>("op_51189_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_51189_end_mask_0 = const()[name = tensor<string, []>("op_51189_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51189_cast_fp16 = slice_by_index(begin = var_51189_begin_0, end = var_51189_end_0, end_mask = var_51189_end_mask_0, x = q_227_cast_fp16)[name = tensor<string, []>("op_51189_cast_fp16")];
            tensor<int32, [4]> var_51193_begin_0 = const()[name = tensor<string, []>("op_51193_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_51193_end_0 = const()[name = tensor<string, []>("op_51193_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_51193_end_mask_0 = const()[name = tensor<string, []>("op_51193_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51193_cast_fp16 = slice_by_index(begin = var_51193_begin_0, end = var_51193_end_0, end_mask = var_51193_end_mask_0, x = q_227_cast_fp16)[name = tensor<string, []>("op_51193_cast_fp16")];
            tensor<int32, [4]> var_51197_begin_0 = const()[name = tensor<string, []>("op_51197_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_51197_end_0 = const()[name = tensor<string, []>("op_51197_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_51197_end_mask_0 = const()[name = tensor<string, []>("op_51197_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51197_cast_fp16 = slice_by_index(begin = var_51197_begin_0, end = var_51197_end_0, end_mask = var_51197_end_mask_0, x = q_227_cast_fp16)[name = tensor<string, []>("op_51197_cast_fp16")];
            tensor<int32, [4]> var_51201_begin_0 = const()[name = tensor<string, []>("op_51201_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_51201_end_0 = const()[name = tensor<string, []>("op_51201_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_51201_end_mask_0 = const()[name = tensor<string, []>("op_51201_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51201_cast_fp16 = slice_by_index(begin = var_51201_begin_0, end = var_51201_end_0, end_mask = var_51201_end_mask_0, x = q_227_cast_fp16)[name = tensor<string, []>("op_51201_cast_fp16")];
            tensor<int32, [4]> var_51205_begin_0 = const()[name = tensor<string, []>("op_51205_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_51205_end_0 = const()[name = tensor<string, []>("op_51205_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_51205_end_mask_0 = const()[name = tensor<string, []>("op_51205_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51205_cast_fp16 = slice_by_index(begin = var_51205_begin_0, end = var_51205_end_0, end_mask = var_51205_end_mask_0, x = q_227_cast_fp16)[name = tensor<string, []>("op_51205_cast_fp16")];
            tensor<int32, [4]> var_51209_begin_0 = const()[name = tensor<string, []>("op_51209_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_51209_end_0 = const()[name = tensor<string, []>("op_51209_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_51209_end_mask_0 = const()[name = tensor<string, []>("op_51209_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51209_cast_fp16 = slice_by_index(begin = var_51209_begin_0, end = var_51209_end_0, end_mask = var_51209_end_mask_0, x = q_227_cast_fp16)[name = tensor<string, []>("op_51209_cast_fp16")];
            tensor<int32, [4]> var_51213_begin_0 = const()[name = tensor<string, []>("op_51213_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_51213_end_0 = const()[name = tensor<string, []>("op_51213_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_51213_end_mask_0 = const()[name = tensor<string, []>("op_51213_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51213_cast_fp16 = slice_by_index(begin = var_51213_begin_0, end = var_51213_end_0, end_mask = var_51213_end_mask_0, x = q_227_cast_fp16)[name = tensor<string, []>("op_51213_cast_fp16")];
            tensor<int32, [4]> var_51217_begin_0 = const()[name = tensor<string, []>("op_51217_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_51217_end_0 = const()[name = tensor<string, []>("op_51217_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_51217_end_mask_0 = const()[name = tensor<string, []>("op_51217_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51217_cast_fp16 = slice_by_index(begin = var_51217_begin_0, end = var_51217_end_0, end_mask = var_51217_end_mask_0, x = q_227_cast_fp16)[name = tensor<string, []>("op_51217_cast_fp16")];
            tensor<int32, [4]> var_51221_begin_0 = const()[name = tensor<string, []>("op_51221_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_51221_end_0 = const()[name = tensor<string, []>("op_51221_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_51221_end_mask_0 = const()[name = tensor<string, []>("op_51221_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51221_cast_fp16 = slice_by_index(begin = var_51221_begin_0, end = var_51221_end_0, end_mask = var_51221_end_mask_0, x = q_227_cast_fp16)[name = tensor<string, []>("op_51221_cast_fp16")];
            tensor<int32, [4]> var_51225_begin_0 = const()[name = tensor<string, []>("op_51225_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_51225_end_0 = const()[name = tensor<string, []>("op_51225_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_51225_end_mask_0 = const()[name = tensor<string, []>("op_51225_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51225_cast_fp16 = slice_by_index(begin = var_51225_begin_0, end = var_51225_end_0, end_mask = var_51225_end_mask_0, x = q_227_cast_fp16)[name = tensor<string, []>("op_51225_cast_fp16")];
            tensor<int32, [4]> var_51229_begin_0 = const()[name = tensor<string, []>("op_51229_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_51229_end_0 = const()[name = tensor<string, []>("op_51229_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_51229_end_mask_0 = const()[name = tensor<string, []>("op_51229_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51229_cast_fp16 = slice_by_index(begin = var_51229_begin_0, end = var_51229_end_0, end_mask = var_51229_end_mask_0, x = q_227_cast_fp16)[name = tensor<string, []>("op_51229_cast_fp16")];
            tensor<int32, [4]> var_51233_begin_0 = const()[name = tensor<string, []>("op_51233_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_51233_end_0 = const()[name = tensor<string, []>("op_51233_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_51233_end_mask_0 = const()[name = tensor<string, []>("op_51233_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51233_cast_fp16 = slice_by_index(begin = var_51233_begin_0, end = var_51233_end_0, end_mask = var_51233_end_mask_0, x = q_227_cast_fp16)[name = tensor<string, []>("op_51233_cast_fp16")];
            tensor<int32, [4]> var_51237_begin_0 = const()[name = tensor<string, []>("op_51237_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_51237_end_0 = const()[name = tensor<string, []>("op_51237_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_51237_end_mask_0 = const()[name = tensor<string, []>("op_51237_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51237_cast_fp16 = slice_by_index(begin = var_51237_begin_0, end = var_51237_end_0, end_mask = var_51237_end_mask_0, x = q_227_cast_fp16)[name = tensor<string, []>("op_51237_cast_fp16")];
            tensor<int32, [4]> var_51241_begin_0 = const()[name = tensor<string, []>("op_51241_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_51241_end_0 = const()[name = tensor<string, []>("op_51241_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_51241_end_mask_0 = const()[name = tensor<string, []>("op_51241_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51241_cast_fp16 = slice_by_index(begin = var_51241_begin_0, end = var_51241_end_0, end_mask = var_51241_end_mask_0, x = q_227_cast_fp16)[name = tensor<string, []>("op_51241_cast_fp16")];
            tensor<int32, [4]> var_51245_begin_0 = const()[name = tensor<string, []>("op_51245_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_51245_end_0 = const()[name = tensor<string, []>("op_51245_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_51245_end_mask_0 = const()[name = tensor<string, []>("op_51245_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51245_cast_fp16 = slice_by_index(begin = var_51245_begin_0, end = var_51245_end_0, end_mask = var_51245_end_mask_0, x = q_227_cast_fp16)[name = tensor<string, []>("op_51245_cast_fp16")];
            tensor<int32, [4]> var_51249_begin_0 = const()[name = tensor<string, []>("op_51249_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_51249_end_0 = const()[name = tensor<string, []>("op_51249_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_51249_end_mask_0 = const()[name = tensor<string, []>("op_51249_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51249_cast_fp16 = slice_by_index(begin = var_51249_begin_0, end = var_51249_end_0, end_mask = var_51249_end_mask_0, x = q_227_cast_fp16)[name = tensor<string, []>("op_51249_cast_fp16")];
            tensor<int32, [4]> var_51253_begin_0 = const()[name = tensor<string, []>("op_51253_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_51253_end_0 = const()[name = tensor<string, []>("op_51253_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_51253_end_mask_0 = const()[name = tensor<string, []>("op_51253_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51253_cast_fp16 = slice_by_index(begin = var_51253_begin_0, end = var_51253_end_0, end_mask = var_51253_end_mask_0, x = q_227_cast_fp16)[name = tensor<string, []>("op_51253_cast_fp16")];
            tensor<int32, [4]> var_51257_begin_0 = const()[name = tensor<string, []>("op_51257_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_51257_end_0 = const()[name = tensor<string, []>("op_51257_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_51257_end_mask_0 = const()[name = tensor<string, []>("op_51257_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51257_cast_fp16 = slice_by_index(begin = var_51257_begin_0, end = var_51257_end_0, end_mask = var_51257_end_mask_0, x = q_227_cast_fp16)[name = tensor<string, []>("op_51257_cast_fp16")];
            tensor<int32, [4]> k_455_perm_0 = const()[name = tensor<string, []>("k_455_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_51264_begin_0 = const()[name = tensor<string, []>("op_51264_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_51264_end_0 = const()[name = tensor<string, []>("op_51264_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_51264_end_mask_0 = const()[name = tensor<string, []>("op_51264_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> k_455_cast_fp16 = transpose(perm = k_455_perm_0, x = k_453_cast_fp16)[name = tensor<string, []>("transpose_26")];
            tensor<fp16, [2, 77, 1, 64]> var_51264_cast_fp16 = slice_by_index(begin = var_51264_begin_0, end = var_51264_end_0, end_mask = var_51264_end_mask_0, x = k_455_cast_fp16)[name = tensor<string, []>("op_51264_cast_fp16")];
            tensor<int32, [4]> var_51268_begin_0 = const()[name = tensor<string, []>("op_51268_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_51268_end_0 = const()[name = tensor<string, []>("op_51268_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_51268_end_mask_0 = const()[name = tensor<string, []>("op_51268_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_51268_cast_fp16 = slice_by_index(begin = var_51268_begin_0, end = var_51268_end_0, end_mask = var_51268_end_mask_0, x = k_455_cast_fp16)[name = tensor<string, []>("op_51268_cast_fp16")];
            tensor<int32, [4]> var_51272_begin_0 = const()[name = tensor<string, []>("op_51272_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_51272_end_0 = const()[name = tensor<string, []>("op_51272_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_51272_end_mask_0 = const()[name = tensor<string, []>("op_51272_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_51272_cast_fp16 = slice_by_index(begin = var_51272_begin_0, end = var_51272_end_0, end_mask = var_51272_end_mask_0, x = k_455_cast_fp16)[name = tensor<string, []>("op_51272_cast_fp16")];
            tensor<int32, [4]> var_51276_begin_0 = const()[name = tensor<string, []>("op_51276_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_51276_end_0 = const()[name = tensor<string, []>("op_51276_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_51276_end_mask_0 = const()[name = tensor<string, []>("op_51276_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_51276_cast_fp16 = slice_by_index(begin = var_51276_begin_0, end = var_51276_end_0, end_mask = var_51276_end_mask_0, x = k_455_cast_fp16)[name = tensor<string, []>("op_51276_cast_fp16")];
            tensor<int32, [4]> var_51280_begin_0 = const()[name = tensor<string, []>("op_51280_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_51280_end_0 = const()[name = tensor<string, []>("op_51280_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_51280_end_mask_0 = const()[name = tensor<string, []>("op_51280_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_51280_cast_fp16 = slice_by_index(begin = var_51280_begin_0, end = var_51280_end_0, end_mask = var_51280_end_mask_0, x = k_455_cast_fp16)[name = tensor<string, []>("op_51280_cast_fp16")];
            tensor<int32, [4]> var_51284_begin_0 = const()[name = tensor<string, []>("op_51284_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_51284_end_0 = const()[name = tensor<string, []>("op_51284_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_51284_end_mask_0 = const()[name = tensor<string, []>("op_51284_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_51284_cast_fp16 = slice_by_index(begin = var_51284_begin_0, end = var_51284_end_0, end_mask = var_51284_end_mask_0, x = k_455_cast_fp16)[name = tensor<string, []>("op_51284_cast_fp16")];
            tensor<int32, [4]> var_51288_begin_0 = const()[name = tensor<string, []>("op_51288_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_51288_end_0 = const()[name = tensor<string, []>("op_51288_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_51288_end_mask_0 = const()[name = tensor<string, []>("op_51288_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_51288_cast_fp16 = slice_by_index(begin = var_51288_begin_0, end = var_51288_end_0, end_mask = var_51288_end_mask_0, x = k_455_cast_fp16)[name = tensor<string, []>("op_51288_cast_fp16")];
            tensor<int32, [4]> var_51292_begin_0 = const()[name = tensor<string, []>("op_51292_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_51292_end_0 = const()[name = tensor<string, []>("op_51292_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_51292_end_mask_0 = const()[name = tensor<string, []>("op_51292_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_51292_cast_fp16 = slice_by_index(begin = var_51292_begin_0, end = var_51292_end_0, end_mask = var_51292_end_mask_0, x = k_455_cast_fp16)[name = tensor<string, []>("op_51292_cast_fp16")];
            tensor<int32, [4]> var_51296_begin_0 = const()[name = tensor<string, []>("op_51296_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_51296_end_0 = const()[name = tensor<string, []>("op_51296_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_51296_end_mask_0 = const()[name = tensor<string, []>("op_51296_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_51296_cast_fp16 = slice_by_index(begin = var_51296_begin_0, end = var_51296_end_0, end_mask = var_51296_end_mask_0, x = k_455_cast_fp16)[name = tensor<string, []>("op_51296_cast_fp16")];
            tensor<int32, [4]> var_51300_begin_0 = const()[name = tensor<string, []>("op_51300_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_51300_end_0 = const()[name = tensor<string, []>("op_51300_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_51300_end_mask_0 = const()[name = tensor<string, []>("op_51300_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_51300_cast_fp16 = slice_by_index(begin = var_51300_begin_0, end = var_51300_end_0, end_mask = var_51300_end_mask_0, x = k_455_cast_fp16)[name = tensor<string, []>("op_51300_cast_fp16")];
            tensor<int32, [4]> var_51304_begin_0 = const()[name = tensor<string, []>("op_51304_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_51304_end_0 = const()[name = tensor<string, []>("op_51304_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_51304_end_mask_0 = const()[name = tensor<string, []>("op_51304_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_51304_cast_fp16 = slice_by_index(begin = var_51304_begin_0, end = var_51304_end_0, end_mask = var_51304_end_mask_0, x = k_455_cast_fp16)[name = tensor<string, []>("op_51304_cast_fp16")];
            tensor<int32, [4]> var_51308_begin_0 = const()[name = tensor<string, []>("op_51308_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_51308_end_0 = const()[name = tensor<string, []>("op_51308_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_51308_end_mask_0 = const()[name = tensor<string, []>("op_51308_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_51308_cast_fp16 = slice_by_index(begin = var_51308_begin_0, end = var_51308_end_0, end_mask = var_51308_end_mask_0, x = k_455_cast_fp16)[name = tensor<string, []>("op_51308_cast_fp16")];
            tensor<int32, [4]> var_51312_begin_0 = const()[name = tensor<string, []>("op_51312_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_51312_end_0 = const()[name = tensor<string, []>("op_51312_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_51312_end_mask_0 = const()[name = tensor<string, []>("op_51312_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_51312_cast_fp16 = slice_by_index(begin = var_51312_begin_0, end = var_51312_end_0, end_mask = var_51312_end_mask_0, x = k_455_cast_fp16)[name = tensor<string, []>("op_51312_cast_fp16")];
            tensor<int32, [4]> var_51316_begin_0 = const()[name = tensor<string, []>("op_51316_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_51316_end_0 = const()[name = tensor<string, []>("op_51316_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_51316_end_mask_0 = const()[name = tensor<string, []>("op_51316_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_51316_cast_fp16 = slice_by_index(begin = var_51316_begin_0, end = var_51316_end_0, end_mask = var_51316_end_mask_0, x = k_455_cast_fp16)[name = tensor<string, []>("op_51316_cast_fp16")];
            tensor<int32, [4]> var_51320_begin_0 = const()[name = tensor<string, []>("op_51320_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_51320_end_0 = const()[name = tensor<string, []>("op_51320_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_51320_end_mask_0 = const()[name = tensor<string, []>("op_51320_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_51320_cast_fp16 = slice_by_index(begin = var_51320_begin_0, end = var_51320_end_0, end_mask = var_51320_end_mask_0, x = k_455_cast_fp16)[name = tensor<string, []>("op_51320_cast_fp16")];
            tensor<int32, [4]> var_51324_begin_0 = const()[name = tensor<string, []>("op_51324_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_51324_end_0 = const()[name = tensor<string, []>("op_51324_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_51324_end_mask_0 = const()[name = tensor<string, []>("op_51324_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_51324_cast_fp16 = slice_by_index(begin = var_51324_begin_0, end = var_51324_end_0, end_mask = var_51324_end_mask_0, x = k_455_cast_fp16)[name = tensor<string, []>("op_51324_cast_fp16")];
            tensor<int32, [4]> var_51328_begin_0 = const()[name = tensor<string, []>("op_51328_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_51328_end_0 = const()[name = tensor<string, []>("op_51328_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_51328_end_mask_0 = const()[name = tensor<string, []>("op_51328_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_51328_cast_fp16 = slice_by_index(begin = var_51328_begin_0, end = var_51328_end_0, end_mask = var_51328_end_mask_0, x = k_455_cast_fp16)[name = tensor<string, []>("op_51328_cast_fp16")];
            tensor<int32, [4]> var_51332_begin_0 = const()[name = tensor<string, []>("op_51332_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_51332_end_0 = const()[name = tensor<string, []>("op_51332_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_51332_end_mask_0 = const()[name = tensor<string, []>("op_51332_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_51332_cast_fp16 = slice_by_index(begin = var_51332_begin_0, end = var_51332_end_0, end_mask = var_51332_end_mask_0, x = k_455_cast_fp16)[name = tensor<string, []>("op_51332_cast_fp16")];
            tensor<int32, [4]> var_51336_begin_0 = const()[name = tensor<string, []>("op_51336_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_51336_end_0 = const()[name = tensor<string, []>("op_51336_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_51336_end_mask_0 = const()[name = tensor<string, []>("op_51336_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_51336_cast_fp16 = slice_by_index(begin = var_51336_begin_0, end = var_51336_end_0, end_mask = var_51336_end_mask_0, x = k_455_cast_fp16)[name = tensor<string, []>("op_51336_cast_fp16")];
            tensor<int32, [4]> var_51340_begin_0 = const()[name = tensor<string, []>("op_51340_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_51340_end_0 = const()[name = tensor<string, []>("op_51340_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_51340_end_mask_0 = const()[name = tensor<string, []>("op_51340_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_51340_cast_fp16 = slice_by_index(begin = var_51340_begin_0, end = var_51340_end_0, end_mask = var_51340_end_mask_0, x = k_455_cast_fp16)[name = tensor<string, []>("op_51340_cast_fp16")];
            tensor<int32, [4]> var_51342_begin_0 = const()[name = tensor<string, []>("op_51342_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_51342_end_0 = const()[name = tensor<string, []>("op_51342_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_51342_end_mask_0 = const()[name = tensor<string, []>("op_51342_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_51342_cast_fp16 = slice_by_index(begin = var_51342_begin_0, end = var_51342_end_0, end_mask = var_51342_end_mask_0, x = v_227_cast_fp16)[name = tensor<string, []>("op_51342_cast_fp16")];
            tensor<int32, [4]> var_51346_begin_0 = const()[name = tensor<string, []>("op_51346_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_51346_end_0 = const()[name = tensor<string, []>("op_51346_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_51346_end_mask_0 = const()[name = tensor<string, []>("op_51346_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_51346_cast_fp16 = slice_by_index(begin = var_51346_begin_0, end = var_51346_end_0, end_mask = var_51346_end_mask_0, x = v_227_cast_fp16)[name = tensor<string, []>("op_51346_cast_fp16")];
            tensor<int32, [4]> var_51350_begin_0 = const()[name = tensor<string, []>("op_51350_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_51350_end_0 = const()[name = tensor<string, []>("op_51350_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_51350_end_mask_0 = const()[name = tensor<string, []>("op_51350_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_51350_cast_fp16 = slice_by_index(begin = var_51350_begin_0, end = var_51350_end_0, end_mask = var_51350_end_mask_0, x = v_227_cast_fp16)[name = tensor<string, []>("op_51350_cast_fp16")];
            tensor<int32, [4]> var_51354_begin_0 = const()[name = tensor<string, []>("op_51354_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_51354_end_0 = const()[name = tensor<string, []>("op_51354_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_51354_end_mask_0 = const()[name = tensor<string, []>("op_51354_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_51354_cast_fp16 = slice_by_index(begin = var_51354_begin_0, end = var_51354_end_0, end_mask = var_51354_end_mask_0, x = v_227_cast_fp16)[name = tensor<string, []>("op_51354_cast_fp16")];
            tensor<int32, [4]> var_51358_begin_0 = const()[name = tensor<string, []>("op_51358_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_51358_end_0 = const()[name = tensor<string, []>("op_51358_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_51358_end_mask_0 = const()[name = tensor<string, []>("op_51358_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_51358_cast_fp16 = slice_by_index(begin = var_51358_begin_0, end = var_51358_end_0, end_mask = var_51358_end_mask_0, x = v_227_cast_fp16)[name = tensor<string, []>("op_51358_cast_fp16")];
            tensor<int32, [4]> var_51362_begin_0 = const()[name = tensor<string, []>("op_51362_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_51362_end_0 = const()[name = tensor<string, []>("op_51362_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_51362_end_mask_0 = const()[name = tensor<string, []>("op_51362_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_51362_cast_fp16 = slice_by_index(begin = var_51362_begin_0, end = var_51362_end_0, end_mask = var_51362_end_mask_0, x = v_227_cast_fp16)[name = tensor<string, []>("op_51362_cast_fp16")];
            tensor<int32, [4]> var_51366_begin_0 = const()[name = tensor<string, []>("op_51366_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_51366_end_0 = const()[name = tensor<string, []>("op_51366_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_51366_end_mask_0 = const()[name = tensor<string, []>("op_51366_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_51366_cast_fp16 = slice_by_index(begin = var_51366_begin_0, end = var_51366_end_0, end_mask = var_51366_end_mask_0, x = v_227_cast_fp16)[name = tensor<string, []>("op_51366_cast_fp16")];
            tensor<int32, [4]> var_51370_begin_0 = const()[name = tensor<string, []>("op_51370_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_51370_end_0 = const()[name = tensor<string, []>("op_51370_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_51370_end_mask_0 = const()[name = tensor<string, []>("op_51370_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_51370_cast_fp16 = slice_by_index(begin = var_51370_begin_0, end = var_51370_end_0, end_mask = var_51370_end_mask_0, x = v_227_cast_fp16)[name = tensor<string, []>("op_51370_cast_fp16")];
            tensor<int32, [4]> var_51374_begin_0 = const()[name = tensor<string, []>("op_51374_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_51374_end_0 = const()[name = tensor<string, []>("op_51374_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_51374_end_mask_0 = const()[name = tensor<string, []>("op_51374_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_51374_cast_fp16 = slice_by_index(begin = var_51374_begin_0, end = var_51374_end_0, end_mask = var_51374_end_mask_0, x = v_227_cast_fp16)[name = tensor<string, []>("op_51374_cast_fp16")];
            tensor<int32, [4]> var_51378_begin_0 = const()[name = tensor<string, []>("op_51378_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_51378_end_0 = const()[name = tensor<string, []>("op_51378_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_51378_end_mask_0 = const()[name = tensor<string, []>("op_51378_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_51378_cast_fp16 = slice_by_index(begin = var_51378_begin_0, end = var_51378_end_0, end_mask = var_51378_end_mask_0, x = v_227_cast_fp16)[name = tensor<string, []>("op_51378_cast_fp16")];
            tensor<int32, [4]> var_51382_begin_0 = const()[name = tensor<string, []>("op_51382_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_51382_end_0 = const()[name = tensor<string, []>("op_51382_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_51382_end_mask_0 = const()[name = tensor<string, []>("op_51382_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_51382_cast_fp16 = slice_by_index(begin = var_51382_begin_0, end = var_51382_end_0, end_mask = var_51382_end_mask_0, x = v_227_cast_fp16)[name = tensor<string, []>("op_51382_cast_fp16")];
            tensor<int32, [4]> var_51386_begin_0 = const()[name = tensor<string, []>("op_51386_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_51386_end_0 = const()[name = tensor<string, []>("op_51386_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_51386_end_mask_0 = const()[name = tensor<string, []>("op_51386_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_51386_cast_fp16 = slice_by_index(begin = var_51386_begin_0, end = var_51386_end_0, end_mask = var_51386_end_mask_0, x = v_227_cast_fp16)[name = tensor<string, []>("op_51386_cast_fp16")];
            tensor<int32, [4]> var_51390_begin_0 = const()[name = tensor<string, []>("op_51390_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_51390_end_0 = const()[name = tensor<string, []>("op_51390_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_51390_end_mask_0 = const()[name = tensor<string, []>("op_51390_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_51390_cast_fp16 = slice_by_index(begin = var_51390_begin_0, end = var_51390_end_0, end_mask = var_51390_end_mask_0, x = v_227_cast_fp16)[name = tensor<string, []>("op_51390_cast_fp16")];
            tensor<int32, [4]> var_51394_begin_0 = const()[name = tensor<string, []>("op_51394_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_51394_end_0 = const()[name = tensor<string, []>("op_51394_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_51394_end_mask_0 = const()[name = tensor<string, []>("op_51394_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_51394_cast_fp16 = slice_by_index(begin = var_51394_begin_0, end = var_51394_end_0, end_mask = var_51394_end_mask_0, x = v_227_cast_fp16)[name = tensor<string, []>("op_51394_cast_fp16")];
            tensor<int32, [4]> var_51398_begin_0 = const()[name = tensor<string, []>("op_51398_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_51398_end_0 = const()[name = tensor<string, []>("op_51398_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_51398_end_mask_0 = const()[name = tensor<string, []>("op_51398_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_51398_cast_fp16 = slice_by_index(begin = var_51398_begin_0, end = var_51398_end_0, end_mask = var_51398_end_mask_0, x = v_227_cast_fp16)[name = tensor<string, []>("op_51398_cast_fp16")];
            tensor<int32, [4]> var_51402_begin_0 = const()[name = tensor<string, []>("op_51402_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_51402_end_0 = const()[name = tensor<string, []>("op_51402_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_51402_end_mask_0 = const()[name = tensor<string, []>("op_51402_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_51402_cast_fp16 = slice_by_index(begin = var_51402_begin_0, end = var_51402_end_0, end_mask = var_51402_end_mask_0, x = v_227_cast_fp16)[name = tensor<string, []>("op_51402_cast_fp16")];
            tensor<int32, [4]> var_51406_begin_0 = const()[name = tensor<string, []>("op_51406_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_51406_end_0 = const()[name = tensor<string, []>("op_51406_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_51406_end_mask_0 = const()[name = tensor<string, []>("op_51406_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_51406_cast_fp16 = slice_by_index(begin = var_51406_begin_0, end = var_51406_end_0, end_mask = var_51406_end_mask_0, x = v_227_cast_fp16)[name = tensor<string, []>("op_51406_cast_fp16")];
            tensor<int32, [4]> var_51410_begin_0 = const()[name = tensor<string, []>("op_51410_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_51410_end_0 = const()[name = tensor<string, []>("op_51410_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_51410_end_mask_0 = const()[name = tensor<string, []>("op_51410_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_51410_cast_fp16 = slice_by_index(begin = var_51410_begin_0, end = var_51410_end_0, end_mask = var_51410_end_mask_0, x = v_227_cast_fp16)[name = tensor<string, []>("op_51410_cast_fp16")];
            tensor<int32, [4]> var_51414_begin_0 = const()[name = tensor<string, []>("op_51414_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_51414_end_0 = const()[name = tensor<string, []>("op_51414_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_51414_end_mask_0 = const()[name = tensor<string, []>("op_51414_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_51414_cast_fp16 = slice_by_index(begin = var_51414_begin_0, end = var_51414_end_0, end_mask = var_51414_end_mask_0, x = v_227_cast_fp16)[name = tensor<string, []>("op_51414_cast_fp16")];
            tensor<int32, [4]> var_51418_begin_0 = const()[name = tensor<string, []>("op_51418_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_51418_end_0 = const()[name = tensor<string, []>("op_51418_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_51418_end_mask_0 = const()[name = tensor<string, []>("op_51418_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_51418_cast_fp16 = slice_by_index(begin = var_51418_begin_0, end = var_51418_end_0, end_mask = var_51418_end_mask_0, x = v_227_cast_fp16)[name = tensor<string, []>("op_51418_cast_fp16")];
            tensor<string, []> var_51422_equation_0 = const()[name = tensor<string, []>("op_51422_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_51422_cast_fp16 = einsum(equation = var_51422_equation_0, values = (var_51264_cast_fp16, var_51181_cast_fp16))[name = tensor<string, []>("op_51422_cast_fp16")];
            tensor<fp16, []> var_51423_to_fp16 = const()[name = tensor<string, []>("op_51423_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4361_cast_fp16 = mul(x = var_51422_cast_fp16, y = var_51423_to_fp16)[name = tensor<string, []>("aw_4361_cast_fp16")];
            tensor<string, []> var_51426_equation_0 = const()[name = tensor<string, []>("op_51426_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_51426_cast_fp16 = einsum(equation = var_51426_equation_0, values = (var_51268_cast_fp16, var_51185_cast_fp16))[name = tensor<string, []>("op_51426_cast_fp16")];
            tensor<fp16, []> var_51427_to_fp16 = const()[name = tensor<string, []>("op_51427_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4363_cast_fp16 = mul(x = var_51426_cast_fp16, y = var_51427_to_fp16)[name = tensor<string, []>("aw_4363_cast_fp16")];
            tensor<string, []> var_51430_equation_0 = const()[name = tensor<string, []>("op_51430_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_51430_cast_fp16 = einsum(equation = var_51430_equation_0, values = (var_51272_cast_fp16, var_51189_cast_fp16))[name = tensor<string, []>("op_51430_cast_fp16")];
            tensor<fp16, []> var_51431_to_fp16 = const()[name = tensor<string, []>("op_51431_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4365_cast_fp16 = mul(x = var_51430_cast_fp16, y = var_51431_to_fp16)[name = tensor<string, []>("aw_4365_cast_fp16")];
            tensor<string, []> var_51434_equation_0 = const()[name = tensor<string, []>("op_51434_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_51434_cast_fp16 = einsum(equation = var_51434_equation_0, values = (var_51276_cast_fp16, var_51193_cast_fp16))[name = tensor<string, []>("op_51434_cast_fp16")];
            tensor<fp16, []> var_51435_to_fp16 = const()[name = tensor<string, []>("op_51435_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4367_cast_fp16 = mul(x = var_51434_cast_fp16, y = var_51435_to_fp16)[name = tensor<string, []>("aw_4367_cast_fp16")];
            tensor<string, []> var_51438_equation_0 = const()[name = tensor<string, []>("op_51438_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_51438_cast_fp16 = einsum(equation = var_51438_equation_0, values = (var_51280_cast_fp16, var_51197_cast_fp16))[name = tensor<string, []>("op_51438_cast_fp16")];
            tensor<fp16, []> var_51439_to_fp16 = const()[name = tensor<string, []>("op_51439_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4369_cast_fp16 = mul(x = var_51438_cast_fp16, y = var_51439_to_fp16)[name = tensor<string, []>("aw_4369_cast_fp16")];
            tensor<string, []> var_51442_equation_0 = const()[name = tensor<string, []>("op_51442_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_51442_cast_fp16 = einsum(equation = var_51442_equation_0, values = (var_51284_cast_fp16, var_51201_cast_fp16))[name = tensor<string, []>("op_51442_cast_fp16")];
            tensor<fp16, []> var_51443_to_fp16 = const()[name = tensor<string, []>("op_51443_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4371_cast_fp16 = mul(x = var_51442_cast_fp16, y = var_51443_to_fp16)[name = tensor<string, []>("aw_4371_cast_fp16")];
            tensor<string, []> var_51446_equation_0 = const()[name = tensor<string, []>("op_51446_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_51446_cast_fp16 = einsum(equation = var_51446_equation_0, values = (var_51288_cast_fp16, var_51205_cast_fp16))[name = tensor<string, []>("op_51446_cast_fp16")];
            tensor<fp16, []> var_51447_to_fp16 = const()[name = tensor<string, []>("op_51447_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4373_cast_fp16 = mul(x = var_51446_cast_fp16, y = var_51447_to_fp16)[name = tensor<string, []>("aw_4373_cast_fp16")];
            tensor<string, []> var_51450_equation_0 = const()[name = tensor<string, []>("op_51450_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_51450_cast_fp16 = einsum(equation = var_51450_equation_0, values = (var_51292_cast_fp16, var_51209_cast_fp16))[name = tensor<string, []>("op_51450_cast_fp16")];
            tensor<fp16, []> var_51451_to_fp16 = const()[name = tensor<string, []>("op_51451_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4375_cast_fp16 = mul(x = var_51450_cast_fp16, y = var_51451_to_fp16)[name = tensor<string, []>("aw_4375_cast_fp16")];
            tensor<string, []> var_51454_equation_0 = const()[name = tensor<string, []>("op_51454_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_51454_cast_fp16 = einsum(equation = var_51454_equation_0, values = (var_51296_cast_fp16, var_51213_cast_fp16))[name = tensor<string, []>("op_51454_cast_fp16")];
            tensor<fp16, []> var_51455_to_fp16 = const()[name = tensor<string, []>("op_51455_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4377_cast_fp16 = mul(x = var_51454_cast_fp16, y = var_51455_to_fp16)[name = tensor<string, []>("aw_4377_cast_fp16")];
            tensor<string, []> var_51458_equation_0 = const()[name = tensor<string, []>("op_51458_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_51458_cast_fp16 = einsum(equation = var_51458_equation_0, values = (var_51300_cast_fp16, var_51217_cast_fp16))[name = tensor<string, []>("op_51458_cast_fp16")];
            tensor<fp16, []> var_51459_to_fp16 = const()[name = tensor<string, []>("op_51459_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4379_cast_fp16 = mul(x = var_51458_cast_fp16, y = var_51459_to_fp16)[name = tensor<string, []>("aw_4379_cast_fp16")];
            tensor<string, []> var_51462_equation_0 = const()[name = tensor<string, []>("op_51462_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_51462_cast_fp16 = einsum(equation = var_51462_equation_0, values = (var_51304_cast_fp16, var_51221_cast_fp16))[name = tensor<string, []>("op_51462_cast_fp16")];
            tensor<fp16, []> var_51463_to_fp16 = const()[name = tensor<string, []>("op_51463_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4381_cast_fp16 = mul(x = var_51462_cast_fp16, y = var_51463_to_fp16)[name = tensor<string, []>("aw_4381_cast_fp16")];
            tensor<string, []> var_51466_equation_0 = const()[name = tensor<string, []>("op_51466_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_51466_cast_fp16 = einsum(equation = var_51466_equation_0, values = (var_51308_cast_fp16, var_51225_cast_fp16))[name = tensor<string, []>("op_51466_cast_fp16")];
            tensor<fp16, []> var_51467_to_fp16 = const()[name = tensor<string, []>("op_51467_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4383_cast_fp16 = mul(x = var_51466_cast_fp16, y = var_51467_to_fp16)[name = tensor<string, []>("aw_4383_cast_fp16")];
            tensor<string, []> var_51470_equation_0 = const()[name = tensor<string, []>("op_51470_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_51470_cast_fp16 = einsum(equation = var_51470_equation_0, values = (var_51312_cast_fp16, var_51229_cast_fp16))[name = tensor<string, []>("op_51470_cast_fp16")];
            tensor<fp16, []> var_51471_to_fp16 = const()[name = tensor<string, []>("op_51471_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4385_cast_fp16 = mul(x = var_51470_cast_fp16, y = var_51471_to_fp16)[name = tensor<string, []>("aw_4385_cast_fp16")];
            tensor<string, []> var_51474_equation_0 = const()[name = tensor<string, []>("op_51474_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_51474_cast_fp16 = einsum(equation = var_51474_equation_0, values = (var_51316_cast_fp16, var_51233_cast_fp16))[name = tensor<string, []>("op_51474_cast_fp16")];
            tensor<fp16, []> var_51475_to_fp16 = const()[name = tensor<string, []>("op_51475_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4387_cast_fp16 = mul(x = var_51474_cast_fp16, y = var_51475_to_fp16)[name = tensor<string, []>("aw_4387_cast_fp16")];
            tensor<string, []> var_51478_equation_0 = const()[name = tensor<string, []>("op_51478_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_51478_cast_fp16 = einsum(equation = var_51478_equation_0, values = (var_51320_cast_fp16, var_51237_cast_fp16))[name = tensor<string, []>("op_51478_cast_fp16")];
            tensor<fp16, []> var_51479_to_fp16 = const()[name = tensor<string, []>("op_51479_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4389_cast_fp16 = mul(x = var_51478_cast_fp16, y = var_51479_to_fp16)[name = tensor<string, []>("aw_4389_cast_fp16")];
            tensor<string, []> var_51482_equation_0 = const()[name = tensor<string, []>("op_51482_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_51482_cast_fp16 = einsum(equation = var_51482_equation_0, values = (var_51324_cast_fp16, var_51241_cast_fp16))[name = tensor<string, []>("op_51482_cast_fp16")];
            tensor<fp16, []> var_51483_to_fp16 = const()[name = tensor<string, []>("op_51483_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4391_cast_fp16 = mul(x = var_51482_cast_fp16, y = var_51483_to_fp16)[name = tensor<string, []>("aw_4391_cast_fp16")];
            tensor<string, []> var_51486_equation_0 = const()[name = tensor<string, []>("op_51486_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_51486_cast_fp16 = einsum(equation = var_51486_equation_0, values = (var_51328_cast_fp16, var_51245_cast_fp16))[name = tensor<string, []>("op_51486_cast_fp16")];
            tensor<fp16, []> var_51487_to_fp16 = const()[name = tensor<string, []>("op_51487_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4393_cast_fp16 = mul(x = var_51486_cast_fp16, y = var_51487_to_fp16)[name = tensor<string, []>("aw_4393_cast_fp16")];
            tensor<string, []> var_51490_equation_0 = const()[name = tensor<string, []>("op_51490_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_51490_cast_fp16 = einsum(equation = var_51490_equation_0, values = (var_51332_cast_fp16, var_51249_cast_fp16))[name = tensor<string, []>("op_51490_cast_fp16")];
            tensor<fp16, []> var_51491_to_fp16 = const()[name = tensor<string, []>("op_51491_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4395_cast_fp16 = mul(x = var_51490_cast_fp16, y = var_51491_to_fp16)[name = tensor<string, []>("aw_4395_cast_fp16")];
            tensor<string, []> var_51494_equation_0 = const()[name = tensor<string, []>("op_51494_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_51494_cast_fp16 = einsum(equation = var_51494_equation_0, values = (var_51336_cast_fp16, var_51253_cast_fp16))[name = tensor<string, []>("op_51494_cast_fp16")];
            tensor<fp16, []> var_51495_to_fp16 = const()[name = tensor<string, []>("op_51495_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4397_cast_fp16 = mul(x = var_51494_cast_fp16, y = var_51495_to_fp16)[name = tensor<string, []>("aw_4397_cast_fp16")];
            tensor<string, []> var_51498_equation_0 = const()[name = tensor<string, []>("op_51498_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_51498_cast_fp16 = einsum(equation = var_51498_equation_0, values = (var_51340_cast_fp16, var_51257_cast_fp16))[name = tensor<string, []>("op_51498_cast_fp16")];
            tensor<fp16, []> var_51499_to_fp16 = const()[name = tensor<string, []>("op_51499_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4399_cast_fp16 = mul(x = var_51498_cast_fp16, y = var_51499_to_fp16)[name = tensor<string, []>("aw_4399_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_51501_cast_fp16 = softmax(axis = var_30355, x = aw_4361_cast_fp16)[name = tensor<string, []>("op_51501_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_51502_cast_fp16 = softmax(axis = var_30355, x = aw_4363_cast_fp16)[name = tensor<string, []>("op_51502_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_51503_cast_fp16 = softmax(axis = var_30355, x = aw_4365_cast_fp16)[name = tensor<string, []>("op_51503_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_51504_cast_fp16 = softmax(axis = var_30355, x = aw_4367_cast_fp16)[name = tensor<string, []>("op_51504_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_51505_cast_fp16 = softmax(axis = var_30355, x = aw_4369_cast_fp16)[name = tensor<string, []>("op_51505_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_51506_cast_fp16 = softmax(axis = var_30355, x = aw_4371_cast_fp16)[name = tensor<string, []>("op_51506_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_51507_cast_fp16 = softmax(axis = var_30355, x = aw_4373_cast_fp16)[name = tensor<string, []>("op_51507_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_51508_cast_fp16 = softmax(axis = var_30355, x = aw_4375_cast_fp16)[name = tensor<string, []>("op_51508_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_51509_cast_fp16 = softmax(axis = var_30355, x = aw_4377_cast_fp16)[name = tensor<string, []>("op_51509_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_51510_cast_fp16 = softmax(axis = var_30355, x = aw_4379_cast_fp16)[name = tensor<string, []>("op_51510_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_51511_cast_fp16 = softmax(axis = var_30355, x = aw_4381_cast_fp16)[name = tensor<string, []>("op_51511_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_51512_cast_fp16 = softmax(axis = var_30355, x = aw_4383_cast_fp16)[name = tensor<string, []>("op_51512_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_51513_cast_fp16 = softmax(axis = var_30355, x = aw_4385_cast_fp16)[name = tensor<string, []>("op_51513_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_51514_cast_fp16 = softmax(axis = var_30355, x = aw_4387_cast_fp16)[name = tensor<string, []>("op_51514_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_51515_cast_fp16 = softmax(axis = var_30355, x = aw_4389_cast_fp16)[name = tensor<string, []>("op_51515_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_51516_cast_fp16 = softmax(axis = var_30355, x = aw_4391_cast_fp16)[name = tensor<string, []>("op_51516_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_51517_cast_fp16 = softmax(axis = var_30355, x = aw_4393_cast_fp16)[name = tensor<string, []>("op_51517_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_51518_cast_fp16 = softmax(axis = var_30355, x = aw_4395_cast_fp16)[name = tensor<string, []>("op_51518_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_51519_cast_fp16 = softmax(axis = var_30355, x = aw_4397_cast_fp16)[name = tensor<string, []>("op_51519_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_51520_cast_fp16 = softmax(axis = var_30355, x = aw_4399_cast_fp16)[name = tensor<string, []>("op_51520_cast_fp16")];
            tensor<string, []> var_51522_equation_0 = const()[name = tensor<string, []>("op_51522_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51522_cast_fp16 = einsum(equation = var_51522_equation_0, values = (var_51342_cast_fp16, var_51501_cast_fp16))[name = tensor<string, []>("op_51522_cast_fp16")];
            tensor<string, []> var_51524_equation_0 = const()[name = tensor<string, []>("op_51524_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51524_cast_fp16 = einsum(equation = var_51524_equation_0, values = (var_51346_cast_fp16, var_51502_cast_fp16))[name = tensor<string, []>("op_51524_cast_fp16")];
            tensor<string, []> var_51526_equation_0 = const()[name = tensor<string, []>("op_51526_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51526_cast_fp16 = einsum(equation = var_51526_equation_0, values = (var_51350_cast_fp16, var_51503_cast_fp16))[name = tensor<string, []>("op_51526_cast_fp16")];
            tensor<string, []> var_51528_equation_0 = const()[name = tensor<string, []>("op_51528_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51528_cast_fp16 = einsum(equation = var_51528_equation_0, values = (var_51354_cast_fp16, var_51504_cast_fp16))[name = tensor<string, []>("op_51528_cast_fp16")];
            tensor<string, []> var_51530_equation_0 = const()[name = tensor<string, []>("op_51530_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51530_cast_fp16 = einsum(equation = var_51530_equation_0, values = (var_51358_cast_fp16, var_51505_cast_fp16))[name = tensor<string, []>("op_51530_cast_fp16")];
            tensor<string, []> var_51532_equation_0 = const()[name = tensor<string, []>("op_51532_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51532_cast_fp16 = einsum(equation = var_51532_equation_0, values = (var_51362_cast_fp16, var_51506_cast_fp16))[name = tensor<string, []>("op_51532_cast_fp16")];
            tensor<string, []> var_51534_equation_0 = const()[name = tensor<string, []>("op_51534_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51534_cast_fp16 = einsum(equation = var_51534_equation_0, values = (var_51366_cast_fp16, var_51507_cast_fp16))[name = tensor<string, []>("op_51534_cast_fp16")];
            tensor<string, []> var_51536_equation_0 = const()[name = tensor<string, []>("op_51536_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51536_cast_fp16 = einsum(equation = var_51536_equation_0, values = (var_51370_cast_fp16, var_51508_cast_fp16))[name = tensor<string, []>("op_51536_cast_fp16")];
            tensor<string, []> var_51538_equation_0 = const()[name = tensor<string, []>("op_51538_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51538_cast_fp16 = einsum(equation = var_51538_equation_0, values = (var_51374_cast_fp16, var_51509_cast_fp16))[name = tensor<string, []>("op_51538_cast_fp16")];
            tensor<string, []> var_51540_equation_0 = const()[name = tensor<string, []>("op_51540_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51540_cast_fp16 = einsum(equation = var_51540_equation_0, values = (var_51378_cast_fp16, var_51510_cast_fp16))[name = tensor<string, []>("op_51540_cast_fp16")];
            tensor<string, []> var_51542_equation_0 = const()[name = tensor<string, []>("op_51542_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51542_cast_fp16 = einsum(equation = var_51542_equation_0, values = (var_51382_cast_fp16, var_51511_cast_fp16))[name = tensor<string, []>("op_51542_cast_fp16")];
            tensor<string, []> var_51544_equation_0 = const()[name = tensor<string, []>("op_51544_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51544_cast_fp16 = einsum(equation = var_51544_equation_0, values = (var_51386_cast_fp16, var_51512_cast_fp16))[name = tensor<string, []>("op_51544_cast_fp16")];
            tensor<string, []> var_51546_equation_0 = const()[name = tensor<string, []>("op_51546_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51546_cast_fp16 = einsum(equation = var_51546_equation_0, values = (var_51390_cast_fp16, var_51513_cast_fp16))[name = tensor<string, []>("op_51546_cast_fp16")];
            tensor<string, []> var_51548_equation_0 = const()[name = tensor<string, []>("op_51548_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51548_cast_fp16 = einsum(equation = var_51548_equation_0, values = (var_51394_cast_fp16, var_51514_cast_fp16))[name = tensor<string, []>("op_51548_cast_fp16")];
            tensor<string, []> var_51550_equation_0 = const()[name = tensor<string, []>("op_51550_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51550_cast_fp16 = einsum(equation = var_51550_equation_0, values = (var_51398_cast_fp16, var_51515_cast_fp16))[name = tensor<string, []>("op_51550_cast_fp16")];
            tensor<string, []> var_51552_equation_0 = const()[name = tensor<string, []>("op_51552_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51552_cast_fp16 = einsum(equation = var_51552_equation_0, values = (var_51402_cast_fp16, var_51516_cast_fp16))[name = tensor<string, []>("op_51552_cast_fp16")];
            tensor<string, []> var_51554_equation_0 = const()[name = tensor<string, []>("op_51554_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51554_cast_fp16 = einsum(equation = var_51554_equation_0, values = (var_51406_cast_fp16, var_51517_cast_fp16))[name = tensor<string, []>("op_51554_cast_fp16")];
            tensor<string, []> var_51556_equation_0 = const()[name = tensor<string, []>("op_51556_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51556_cast_fp16 = einsum(equation = var_51556_equation_0, values = (var_51410_cast_fp16, var_51518_cast_fp16))[name = tensor<string, []>("op_51556_cast_fp16")];
            tensor<string, []> var_51558_equation_0 = const()[name = tensor<string, []>("op_51558_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51558_cast_fp16 = einsum(equation = var_51558_equation_0, values = (var_51414_cast_fp16, var_51519_cast_fp16))[name = tensor<string, []>("op_51558_cast_fp16")];
            tensor<string, []> var_51560_equation_0 = const()[name = tensor<string, []>("op_51560_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51560_cast_fp16 = einsum(equation = var_51560_equation_0, values = (var_51418_cast_fp16, var_51520_cast_fp16))[name = tensor<string, []>("op_51560_cast_fp16")];
            tensor<bool, []> input_653_interleave_0 = const()[name = tensor<string, []>("input_653_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_653_cast_fp16 = concat(axis = var_30355, interleave = input_653_interleave_0, values = (var_51522_cast_fp16, var_51524_cast_fp16, var_51526_cast_fp16, var_51528_cast_fp16, var_51530_cast_fp16, var_51532_cast_fp16, var_51534_cast_fp16, var_51536_cast_fp16, var_51538_cast_fp16, var_51540_cast_fp16, var_51542_cast_fp16, var_51544_cast_fp16, var_51546_cast_fp16, var_51548_cast_fp16, var_51550_cast_fp16, var_51552_cast_fp16, var_51554_cast_fp16, var_51556_cast_fp16, var_51558_cast_fp16, var_51560_cast_fp16))[name = tensor<string, []>("input_653_cast_fp16")];
            tensor<string, []> var_51570_pad_type_0 = const()[name = tensor<string, []>("op_51570_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_51570_strides_0 = const()[name = tensor<string, []>("op_51570_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_51570_pad_0 = const()[name = tensor<string, []>("op_51570_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_51570_dilations_0 = const()[name = tensor<string, []>("op_51570_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_51570_groups_0 = const()[name = tensor<string, []>("op_51570_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_2_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(660529152))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(661758016))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_2_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_2_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_2_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(661758208)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_51570_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_2_attn2_to_out_0_bias_to_fp16, dilations = var_51570_dilations_0, groups = var_51570_groups_0, pad = var_51570_pad_0, pad_type = var_51570_pad_type_0, strides = var_51570_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_2_attn2_to_out_0_weight_to_fp16_palettized, x = input_653_cast_fp16)[name = tensor<string, []>("op_51570_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_341_cast_fp16 = add(x = var_51570_cast_fp16, y = inputs_339_cast_fp16)[name = tensor<string, []>("inputs_341_cast_fp16")];
            tensor<int32, [1]> input_655_axes_0 = const()[name = tensor<string, []>("input_655_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> input_655_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_655_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(661760832)))];
            tensor<fp16, [1280]> input_655_beta_0_to_fp16 = const()[name = tensor<string, []>("input_655_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(661763456)))];
            tensor<fp16, []> var_51580_to_fp16 = const()[name = tensor<string, []>("op_51580_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> input_655_cast_fp16 = layer_norm(axes = input_655_axes_0, beta = input_655_beta_0_to_fp16, epsilon = var_51580_to_fp16, gamma = input_655_gamma_0_to_fp16, x = inputs_341_cast_fp16)[name = tensor<string, []>("input_655_cast_fp16")];
            tensor<string, []> var_51600_pad_type_0 = const()[name = tensor<string, []>("op_51600_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_51600_strides_0 = const()[name = tensor<string, []>("op_51600_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_51600_pad_0 = const()[name = tensor<string, []>("op_51600_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_51600_dilations_0 = const()[name = tensor<string, []>("op_51600_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_51600_groups_0 = const()[name = tensor<string, []>("op_51600_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_2_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [9830400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(661766080))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(671596544))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_2_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([10240, 1280, 1, 1])];
            tensor<fp16, [10240]> up_blocks_0_attentions_2_transformer_blocks_2_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_2_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(671596736)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_51600_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_2_ff_net_0_proj_bias_to_fp16, dilations = var_51600_dilations_0, groups = var_51600_groups_0, pad = var_51600_pad_0, pad_type = var_51600_pad_type_0, strides = var_51600_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_2_ff_net_0_proj_weight_to_fp16_palettized, x = input_655_cast_fp16)[name = tensor<string, []>("op_51600_cast_fp16")];
            tensor<int32, [2]> var_51601_split_sizes_0 = const()[name = tensor<string, []>("op_51601_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_51601_axis_0 = const()[name = tensor<string, []>("op_51601_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_51601_cast_fp16_0, tensor<fp16, [2, 5120, 1, 1024]> var_51601_cast_fp16_1 = split(axis = var_51601_axis_0, split_sizes = var_51601_split_sizes_0, x = var_51600_cast_fp16)[name = tensor<string, []>("op_51601_cast_fp16")];
            tensor<string, []> var_51603_mode_0 = const()[name = tensor<string, []>("op_51603_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_51603_cast_fp16 = gelu(mode = var_51603_mode_0, x = var_51601_cast_fp16_1)[name = tensor<string, []>("op_51603_cast_fp16")];
            tensor<fp16, [2, 5120, 1, 1024]> input_657_cast_fp16 = mul(x = var_51601_cast_fp16_0, y = var_51603_cast_fp16)[name = tensor<string, []>("input_657_cast_fp16")];
            tensor<string, []> var_51611_pad_type_0 = const()[name = tensor<string, []>("op_51611_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_51611_strides_0 = const()[name = tensor<string, []>("op_51611_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_51611_pad_0 = const()[name = tensor<string, []>("op_51611_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_51611_dilations_0 = const()[name = tensor<string, []>("op_51611_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_51611_groups_0 = const()[name = tensor<string, []>("op_51611_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_2_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [4915200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(671617280))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(676532544))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_2_ff_net_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 5120, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_2_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_2_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(676532736)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_51611_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_2_ff_net_2_bias_to_fp16, dilations = var_51611_dilations_0, groups = var_51611_groups_0, pad = var_51611_pad_0, pad_type = var_51611_pad_type_0, strides = var_51611_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_2_ff_net_2_weight_to_fp16_palettized, x = input_657_cast_fp16)[name = tensor<string, []>("op_51611_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_343_cast_fp16 = add(x = var_51611_cast_fp16, y = inputs_341_cast_fp16)[name = tensor<string, []>("inputs_343_cast_fp16")];
            tensor<int32, [1]> hidden_states_453_axes_0 = const()[name = tensor<string, []>("hidden_states_453_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_453_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_453_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(676535360)))];
            tensor<fp16, [1280]> hidden_states_453_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_453_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(676537984)))];
            tensor<fp16, []> var_51627_to_fp16 = const()[name = tensor<string, []>("op_51627_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_453_cast_fp16 = layer_norm(axes = hidden_states_453_axes_0, beta = hidden_states_453_beta_0_to_fp16, epsilon = var_51627_to_fp16, gamma = hidden_states_453_gamma_0_to_fp16, x = inputs_343_cast_fp16)[name = tensor<string, []>("hidden_states_453_cast_fp16")];
            tensor<string, []> q_229_pad_type_0 = const()[name = tensor<string, []>("q_229_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_229_strides_0 = const()[name = tensor<string, []>("q_229_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_229_pad_0 = const()[name = tensor<string, []>("q_229_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_229_dilations_0 = const()[name = tensor<string, []>("q_229_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_229_groups_0 = const()[name = tensor<string, []>("q_229_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_3_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(676540608))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(677769472))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_3_attn1_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_229_cast_fp16 = conv(dilations = q_229_dilations_0, groups = q_229_groups_0, pad = q_229_pad_0, pad_type = q_229_pad_type_0, strides = q_229_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_3_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_453_cast_fp16)[name = tensor<string, []>("q_229_cast_fp16")];
            tensor<string, []> k_457_pad_type_0 = const()[name = tensor<string, []>("k_457_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_457_strides_0 = const()[name = tensor<string, []>("k_457_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_457_pad_0 = const()[name = tensor<string, []>("k_457_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_457_dilations_0 = const()[name = tensor<string, []>("k_457_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_457_groups_0 = const()[name = tensor<string, []>("k_457_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_3_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(677769664))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(678998528))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_3_attn1_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> k_457_cast_fp16 = conv(dilations = k_457_dilations_0, groups = k_457_groups_0, pad = k_457_pad_0, pad_type = k_457_pad_type_0, strides = k_457_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_3_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_453_cast_fp16)[name = tensor<string, []>("k_457_cast_fp16")];
            tensor<string, []> v_229_pad_type_0 = const()[name = tensor<string, []>("v_229_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_229_strides_0 = const()[name = tensor<string, []>("v_229_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_229_pad_0 = const()[name = tensor<string, []>("v_229_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_229_dilations_0 = const()[name = tensor<string, []>("v_229_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_229_groups_0 = const()[name = tensor<string, []>("v_229_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_3_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(678998720))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(680227584))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_3_attn1_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> v_229_cast_fp16 = conv(dilations = v_229_dilations_0, groups = v_229_groups_0, pad = v_229_pad_0, pad_type = v_229_pad_type_0, strides = v_229_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_3_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_453_cast_fp16)[name = tensor<string, []>("v_229_cast_fp16")];
            tensor<int32, [4]> var_51660_begin_0 = const()[name = tensor<string, []>("op_51660_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_51660_end_0 = const()[name = tensor<string, []>("op_51660_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_51660_end_mask_0 = const()[name = tensor<string, []>("op_51660_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51660_cast_fp16 = slice_by_index(begin = var_51660_begin_0, end = var_51660_end_0, end_mask = var_51660_end_mask_0, x = q_229_cast_fp16)[name = tensor<string, []>("op_51660_cast_fp16")];
            tensor<int32, [4]> var_51664_begin_0 = const()[name = tensor<string, []>("op_51664_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_51664_end_0 = const()[name = tensor<string, []>("op_51664_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_51664_end_mask_0 = const()[name = tensor<string, []>("op_51664_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51664_cast_fp16 = slice_by_index(begin = var_51664_begin_0, end = var_51664_end_0, end_mask = var_51664_end_mask_0, x = q_229_cast_fp16)[name = tensor<string, []>("op_51664_cast_fp16")];
            tensor<int32, [4]> var_51668_begin_0 = const()[name = tensor<string, []>("op_51668_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_51668_end_0 = const()[name = tensor<string, []>("op_51668_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_51668_end_mask_0 = const()[name = tensor<string, []>("op_51668_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51668_cast_fp16 = slice_by_index(begin = var_51668_begin_0, end = var_51668_end_0, end_mask = var_51668_end_mask_0, x = q_229_cast_fp16)[name = tensor<string, []>("op_51668_cast_fp16")];
            tensor<int32, [4]> var_51672_begin_0 = const()[name = tensor<string, []>("op_51672_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_51672_end_0 = const()[name = tensor<string, []>("op_51672_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_51672_end_mask_0 = const()[name = tensor<string, []>("op_51672_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51672_cast_fp16 = slice_by_index(begin = var_51672_begin_0, end = var_51672_end_0, end_mask = var_51672_end_mask_0, x = q_229_cast_fp16)[name = tensor<string, []>("op_51672_cast_fp16")];
            tensor<int32, [4]> var_51676_begin_0 = const()[name = tensor<string, []>("op_51676_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_51676_end_0 = const()[name = tensor<string, []>("op_51676_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_51676_end_mask_0 = const()[name = tensor<string, []>("op_51676_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51676_cast_fp16 = slice_by_index(begin = var_51676_begin_0, end = var_51676_end_0, end_mask = var_51676_end_mask_0, x = q_229_cast_fp16)[name = tensor<string, []>("op_51676_cast_fp16")];
            tensor<int32, [4]> var_51680_begin_0 = const()[name = tensor<string, []>("op_51680_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_51680_end_0 = const()[name = tensor<string, []>("op_51680_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_51680_end_mask_0 = const()[name = tensor<string, []>("op_51680_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51680_cast_fp16 = slice_by_index(begin = var_51680_begin_0, end = var_51680_end_0, end_mask = var_51680_end_mask_0, x = q_229_cast_fp16)[name = tensor<string, []>("op_51680_cast_fp16")];
            tensor<int32, [4]> var_51684_begin_0 = const()[name = tensor<string, []>("op_51684_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_51684_end_0 = const()[name = tensor<string, []>("op_51684_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_51684_end_mask_0 = const()[name = tensor<string, []>("op_51684_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51684_cast_fp16 = slice_by_index(begin = var_51684_begin_0, end = var_51684_end_0, end_mask = var_51684_end_mask_0, x = q_229_cast_fp16)[name = tensor<string, []>("op_51684_cast_fp16")];
            tensor<int32, [4]> var_51688_begin_0 = const()[name = tensor<string, []>("op_51688_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_51688_end_0 = const()[name = tensor<string, []>("op_51688_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_51688_end_mask_0 = const()[name = tensor<string, []>("op_51688_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51688_cast_fp16 = slice_by_index(begin = var_51688_begin_0, end = var_51688_end_0, end_mask = var_51688_end_mask_0, x = q_229_cast_fp16)[name = tensor<string, []>("op_51688_cast_fp16")];
            tensor<int32, [4]> var_51692_begin_0 = const()[name = tensor<string, []>("op_51692_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_51692_end_0 = const()[name = tensor<string, []>("op_51692_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_51692_end_mask_0 = const()[name = tensor<string, []>("op_51692_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51692_cast_fp16 = slice_by_index(begin = var_51692_begin_0, end = var_51692_end_0, end_mask = var_51692_end_mask_0, x = q_229_cast_fp16)[name = tensor<string, []>("op_51692_cast_fp16")];
            tensor<int32, [4]> var_51696_begin_0 = const()[name = tensor<string, []>("op_51696_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_51696_end_0 = const()[name = tensor<string, []>("op_51696_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_51696_end_mask_0 = const()[name = tensor<string, []>("op_51696_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51696_cast_fp16 = slice_by_index(begin = var_51696_begin_0, end = var_51696_end_0, end_mask = var_51696_end_mask_0, x = q_229_cast_fp16)[name = tensor<string, []>("op_51696_cast_fp16")];
            tensor<int32, [4]> var_51700_begin_0 = const()[name = tensor<string, []>("op_51700_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_51700_end_0 = const()[name = tensor<string, []>("op_51700_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_51700_end_mask_0 = const()[name = tensor<string, []>("op_51700_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51700_cast_fp16 = slice_by_index(begin = var_51700_begin_0, end = var_51700_end_0, end_mask = var_51700_end_mask_0, x = q_229_cast_fp16)[name = tensor<string, []>("op_51700_cast_fp16")];
            tensor<int32, [4]> var_51704_begin_0 = const()[name = tensor<string, []>("op_51704_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_51704_end_0 = const()[name = tensor<string, []>("op_51704_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_51704_end_mask_0 = const()[name = tensor<string, []>("op_51704_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51704_cast_fp16 = slice_by_index(begin = var_51704_begin_0, end = var_51704_end_0, end_mask = var_51704_end_mask_0, x = q_229_cast_fp16)[name = tensor<string, []>("op_51704_cast_fp16")];
            tensor<int32, [4]> var_51708_begin_0 = const()[name = tensor<string, []>("op_51708_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_51708_end_0 = const()[name = tensor<string, []>("op_51708_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_51708_end_mask_0 = const()[name = tensor<string, []>("op_51708_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51708_cast_fp16 = slice_by_index(begin = var_51708_begin_0, end = var_51708_end_0, end_mask = var_51708_end_mask_0, x = q_229_cast_fp16)[name = tensor<string, []>("op_51708_cast_fp16")];
            tensor<int32, [4]> var_51712_begin_0 = const()[name = tensor<string, []>("op_51712_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_51712_end_0 = const()[name = tensor<string, []>("op_51712_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_51712_end_mask_0 = const()[name = tensor<string, []>("op_51712_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51712_cast_fp16 = slice_by_index(begin = var_51712_begin_0, end = var_51712_end_0, end_mask = var_51712_end_mask_0, x = q_229_cast_fp16)[name = tensor<string, []>("op_51712_cast_fp16")];
            tensor<int32, [4]> var_51716_begin_0 = const()[name = tensor<string, []>("op_51716_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_51716_end_0 = const()[name = tensor<string, []>("op_51716_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_51716_end_mask_0 = const()[name = tensor<string, []>("op_51716_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51716_cast_fp16 = slice_by_index(begin = var_51716_begin_0, end = var_51716_end_0, end_mask = var_51716_end_mask_0, x = q_229_cast_fp16)[name = tensor<string, []>("op_51716_cast_fp16")];
            tensor<int32, [4]> var_51720_begin_0 = const()[name = tensor<string, []>("op_51720_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_51720_end_0 = const()[name = tensor<string, []>("op_51720_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_51720_end_mask_0 = const()[name = tensor<string, []>("op_51720_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51720_cast_fp16 = slice_by_index(begin = var_51720_begin_0, end = var_51720_end_0, end_mask = var_51720_end_mask_0, x = q_229_cast_fp16)[name = tensor<string, []>("op_51720_cast_fp16")];
            tensor<int32, [4]> var_51724_begin_0 = const()[name = tensor<string, []>("op_51724_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_51724_end_0 = const()[name = tensor<string, []>("op_51724_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_51724_end_mask_0 = const()[name = tensor<string, []>("op_51724_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51724_cast_fp16 = slice_by_index(begin = var_51724_begin_0, end = var_51724_end_0, end_mask = var_51724_end_mask_0, x = q_229_cast_fp16)[name = tensor<string, []>("op_51724_cast_fp16")];
            tensor<int32, [4]> var_51728_begin_0 = const()[name = tensor<string, []>("op_51728_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_51728_end_0 = const()[name = tensor<string, []>("op_51728_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_51728_end_mask_0 = const()[name = tensor<string, []>("op_51728_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51728_cast_fp16 = slice_by_index(begin = var_51728_begin_0, end = var_51728_end_0, end_mask = var_51728_end_mask_0, x = q_229_cast_fp16)[name = tensor<string, []>("op_51728_cast_fp16")];
            tensor<int32, [4]> var_51732_begin_0 = const()[name = tensor<string, []>("op_51732_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_51732_end_0 = const()[name = tensor<string, []>("op_51732_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_51732_end_mask_0 = const()[name = tensor<string, []>("op_51732_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51732_cast_fp16 = slice_by_index(begin = var_51732_begin_0, end = var_51732_end_0, end_mask = var_51732_end_mask_0, x = q_229_cast_fp16)[name = tensor<string, []>("op_51732_cast_fp16")];
            tensor<int32, [4]> var_51736_begin_0 = const()[name = tensor<string, []>("op_51736_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_51736_end_0 = const()[name = tensor<string, []>("op_51736_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_51736_end_mask_0 = const()[name = tensor<string, []>("op_51736_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51736_cast_fp16 = slice_by_index(begin = var_51736_begin_0, end = var_51736_end_0, end_mask = var_51736_end_mask_0, x = q_229_cast_fp16)[name = tensor<string, []>("op_51736_cast_fp16")];
            tensor<int32, [4]> k_459_perm_0 = const()[name = tensor<string, []>("k_459_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_51743_begin_0 = const()[name = tensor<string, []>("op_51743_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_51743_end_0 = const()[name = tensor<string, []>("op_51743_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_51743_end_mask_0 = const()[name = tensor<string, []>("op_51743_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> k_459_cast_fp16 = transpose(perm = k_459_perm_0, x = k_457_cast_fp16)[name = tensor<string, []>("transpose_25")];
            tensor<fp16, [2, 1024, 1, 64]> var_51743_cast_fp16 = slice_by_index(begin = var_51743_begin_0, end = var_51743_end_0, end_mask = var_51743_end_mask_0, x = k_459_cast_fp16)[name = tensor<string, []>("op_51743_cast_fp16")];
            tensor<int32, [4]> var_51747_begin_0 = const()[name = tensor<string, []>("op_51747_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_51747_end_0 = const()[name = tensor<string, []>("op_51747_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_51747_end_mask_0 = const()[name = tensor<string, []>("op_51747_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_51747_cast_fp16 = slice_by_index(begin = var_51747_begin_0, end = var_51747_end_0, end_mask = var_51747_end_mask_0, x = k_459_cast_fp16)[name = tensor<string, []>("op_51747_cast_fp16")];
            tensor<int32, [4]> var_51751_begin_0 = const()[name = tensor<string, []>("op_51751_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_51751_end_0 = const()[name = tensor<string, []>("op_51751_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_51751_end_mask_0 = const()[name = tensor<string, []>("op_51751_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_51751_cast_fp16 = slice_by_index(begin = var_51751_begin_0, end = var_51751_end_0, end_mask = var_51751_end_mask_0, x = k_459_cast_fp16)[name = tensor<string, []>("op_51751_cast_fp16")];
            tensor<int32, [4]> var_51755_begin_0 = const()[name = tensor<string, []>("op_51755_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_51755_end_0 = const()[name = tensor<string, []>("op_51755_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_51755_end_mask_0 = const()[name = tensor<string, []>("op_51755_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_51755_cast_fp16 = slice_by_index(begin = var_51755_begin_0, end = var_51755_end_0, end_mask = var_51755_end_mask_0, x = k_459_cast_fp16)[name = tensor<string, []>("op_51755_cast_fp16")];
            tensor<int32, [4]> var_51759_begin_0 = const()[name = tensor<string, []>("op_51759_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_51759_end_0 = const()[name = tensor<string, []>("op_51759_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_51759_end_mask_0 = const()[name = tensor<string, []>("op_51759_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_51759_cast_fp16 = slice_by_index(begin = var_51759_begin_0, end = var_51759_end_0, end_mask = var_51759_end_mask_0, x = k_459_cast_fp16)[name = tensor<string, []>("op_51759_cast_fp16")];
            tensor<int32, [4]> var_51763_begin_0 = const()[name = tensor<string, []>("op_51763_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_51763_end_0 = const()[name = tensor<string, []>("op_51763_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_51763_end_mask_0 = const()[name = tensor<string, []>("op_51763_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_51763_cast_fp16 = slice_by_index(begin = var_51763_begin_0, end = var_51763_end_0, end_mask = var_51763_end_mask_0, x = k_459_cast_fp16)[name = tensor<string, []>("op_51763_cast_fp16")];
            tensor<int32, [4]> var_51767_begin_0 = const()[name = tensor<string, []>("op_51767_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_51767_end_0 = const()[name = tensor<string, []>("op_51767_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_51767_end_mask_0 = const()[name = tensor<string, []>("op_51767_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_51767_cast_fp16 = slice_by_index(begin = var_51767_begin_0, end = var_51767_end_0, end_mask = var_51767_end_mask_0, x = k_459_cast_fp16)[name = tensor<string, []>("op_51767_cast_fp16")];
            tensor<int32, [4]> var_51771_begin_0 = const()[name = tensor<string, []>("op_51771_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_51771_end_0 = const()[name = tensor<string, []>("op_51771_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_51771_end_mask_0 = const()[name = tensor<string, []>("op_51771_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_51771_cast_fp16 = slice_by_index(begin = var_51771_begin_0, end = var_51771_end_0, end_mask = var_51771_end_mask_0, x = k_459_cast_fp16)[name = tensor<string, []>("op_51771_cast_fp16")];
            tensor<int32, [4]> var_51775_begin_0 = const()[name = tensor<string, []>("op_51775_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_51775_end_0 = const()[name = tensor<string, []>("op_51775_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_51775_end_mask_0 = const()[name = tensor<string, []>("op_51775_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_51775_cast_fp16 = slice_by_index(begin = var_51775_begin_0, end = var_51775_end_0, end_mask = var_51775_end_mask_0, x = k_459_cast_fp16)[name = tensor<string, []>("op_51775_cast_fp16")];
            tensor<int32, [4]> var_51779_begin_0 = const()[name = tensor<string, []>("op_51779_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_51779_end_0 = const()[name = tensor<string, []>("op_51779_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_51779_end_mask_0 = const()[name = tensor<string, []>("op_51779_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_51779_cast_fp16 = slice_by_index(begin = var_51779_begin_0, end = var_51779_end_0, end_mask = var_51779_end_mask_0, x = k_459_cast_fp16)[name = tensor<string, []>("op_51779_cast_fp16")];
            tensor<int32, [4]> var_51783_begin_0 = const()[name = tensor<string, []>("op_51783_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_51783_end_0 = const()[name = tensor<string, []>("op_51783_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_51783_end_mask_0 = const()[name = tensor<string, []>("op_51783_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_51783_cast_fp16 = slice_by_index(begin = var_51783_begin_0, end = var_51783_end_0, end_mask = var_51783_end_mask_0, x = k_459_cast_fp16)[name = tensor<string, []>("op_51783_cast_fp16")];
            tensor<int32, [4]> var_51787_begin_0 = const()[name = tensor<string, []>("op_51787_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_51787_end_0 = const()[name = tensor<string, []>("op_51787_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_51787_end_mask_0 = const()[name = tensor<string, []>("op_51787_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_51787_cast_fp16 = slice_by_index(begin = var_51787_begin_0, end = var_51787_end_0, end_mask = var_51787_end_mask_0, x = k_459_cast_fp16)[name = tensor<string, []>("op_51787_cast_fp16")];
            tensor<int32, [4]> var_51791_begin_0 = const()[name = tensor<string, []>("op_51791_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_51791_end_0 = const()[name = tensor<string, []>("op_51791_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_51791_end_mask_0 = const()[name = tensor<string, []>("op_51791_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_51791_cast_fp16 = slice_by_index(begin = var_51791_begin_0, end = var_51791_end_0, end_mask = var_51791_end_mask_0, x = k_459_cast_fp16)[name = tensor<string, []>("op_51791_cast_fp16")];
            tensor<int32, [4]> var_51795_begin_0 = const()[name = tensor<string, []>("op_51795_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_51795_end_0 = const()[name = tensor<string, []>("op_51795_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_51795_end_mask_0 = const()[name = tensor<string, []>("op_51795_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_51795_cast_fp16 = slice_by_index(begin = var_51795_begin_0, end = var_51795_end_0, end_mask = var_51795_end_mask_0, x = k_459_cast_fp16)[name = tensor<string, []>("op_51795_cast_fp16")];
            tensor<int32, [4]> var_51799_begin_0 = const()[name = tensor<string, []>("op_51799_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_51799_end_0 = const()[name = tensor<string, []>("op_51799_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_51799_end_mask_0 = const()[name = tensor<string, []>("op_51799_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_51799_cast_fp16 = slice_by_index(begin = var_51799_begin_0, end = var_51799_end_0, end_mask = var_51799_end_mask_0, x = k_459_cast_fp16)[name = tensor<string, []>("op_51799_cast_fp16")];
            tensor<int32, [4]> var_51803_begin_0 = const()[name = tensor<string, []>("op_51803_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_51803_end_0 = const()[name = tensor<string, []>("op_51803_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_51803_end_mask_0 = const()[name = tensor<string, []>("op_51803_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_51803_cast_fp16 = slice_by_index(begin = var_51803_begin_0, end = var_51803_end_0, end_mask = var_51803_end_mask_0, x = k_459_cast_fp16)[name = tensor<string, []>("op_51803_cast_fp16")];
            tensor<int32, [4]> var_51807_begin_0 = const()[name = tensor<string, []>("op_51807_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_51807_end_0 = const()[name = tensor<string, []>("op_51807_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_51807_end_mask_0 = const()[name = tensor<string, []>("op_51807_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_51807_cast_fp16 = slice_by_index(begin = var_51807_begin_0, end = var_51807_end_0, end_mask = var_51807_end_mask_0, x = k_459_cast_fp16)[name = tensor<string, []>("op_51807_cast_fp16")];
            tensor<int32, [4]> var_51811_begin_0 = const()[name = tensor<string, []>("op_51811_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_51811_end_0 = const()[name = tensor<string, []>("op_51811_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_51811_end_mask_0 = const()[name = tensor<string, []>("op_51811_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_51811_cast_fp16 = slice_by_index(begin = var_51811_begin_0, end = var_51811_end_0, end_mask = var_51811_end_mask_0, x = k_459_cast_fp16)[name = tensor<string, []>("op_51811_cast_fp16")];
            tensor<int32, [4]> var_51815_begin_0 = const()[name = tensor<string, []>("op_51815_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_51815_end_0 = const()[name = tensor<string, []>("op_51815_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_51815_end_mask_0 = const()[name = tensor<string, []>("op_51815_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_51815_cast_fp16 = slice_by_index(begin = var_51815_begin_0, end = var_51815_end_0, end_mask = var_51815_end_mask_0, x = k_459_cast_fp16)[name = tensor<string, []>("op_51815_cast_fp16")];
            tensor<int32, [4]> var_51819_begin_0 = const()[name = tensor<string, []>("op_51819_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_51819_end_0 = const()[name = tensor<string, []>("op_51819_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_51819_end_mask_0 = const()[name = tensor<string, []>("op_51819_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_51819_cast_fp16 = slice_by_index(begin = var_51819_begin_0, end = var_51819_end_0, end_mask = var_51819_end_mask_0, x = k_459_cast_fp16)[name = tensor<string, []>("op_51819_cast_fp16")];
            tensor<int32, [4]> var_51821_begin_0 = const()[name = tensor<string, []>("op_51821_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_51821_end_0 = const()[name = tensor<string, []>("op_51821_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_51821_end_mask_0 = const()[name = tensor<string, []>("op_51821_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51821_cast_fp16 = slice_by_index(begin = var_51821_begin_0, end = var_51821_end_0, end_mask = var_51821_end_mask_0, x = v_229_cast_fp16)[name = tensor<string, []>("op_51821_cast_fp16")];
            tensor<int32, [4]> var_51825_begin_0 = const()[name = tensor<string, []>("op_51825_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_51825_end_0 = const()[name = tensor<string, []>("op_51825_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_51825_end_mask_0 = const()[name = tensor<string, []>("op_51825_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51825_cast_fp16 = slice_by_index(begin = var_51825_begin_0, end = var_51825_end_0, end_mask = var_51825_end_mask_0, x = v_229_cast_fp16)[name = tensor<string, []>("op_51825_cast_fp16")];
            tensor<int32, [4]> var_51829_begin_0 = const()[name = tensor<string, []>("op_51829_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_51829_end_0 = const()[name = tensor<string, []>("op_51829_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_51829_end_mask_0 = const()[name = tensor<string, []>("op_51829_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51829_cast_fp16 = slice_by_index(begin = var_51829_begin_0, end = var_51829_end_0, end_mask = var_51829_end_mask_0, x = v_229_cast_fp16)[name = tensor<string, []>("op_51829_cast_fp16")];
            tensor<int32, [4]> var_51833_begin_0 = const()[name = tensor<string, []>("op_51833_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_51833_end_0 = const()[name = tensor<string, []>("op_51833_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_51833_end_mask_0 = const()[name = tensor<string, []>("op_51833_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51833_cast_fp16 = slice_by_index(begin = var_51833_begin_0, end = var_51833_end_0, end_mask = var_51833_end_mask_0, x = v_229_cast_fp16)[name = tensor<string, []>("op_51833_cast_fp16")];
            tensor<int32, [4]> var_51837_begin_0 = const()[name = tensor<string, []>("op_51837_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_51837_end_0 = const()[name = tensor<string, []>("op_51837_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_51837_end_mask_0 = const()[name = tensor<string, []>("op_51837_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51837_cast_fp16 = slice_by_index(begin = var_51837_begin_0, end = var_51837_end_0, end_mask = var_51837_end_mask_0, x = v_229_cast_fp16)[name = tensor<string, []>("op_51837_cast_fp16")];
            tensor<int32, [4]> var_51841_begin_0 = const()[name = tensor<string, []>("op_51841_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_51841_end_0 = const()[name = tensor<string, []>("op_51841_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_51841_end_mask_0 = const()[name = tensor<string, []>("op_51841_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51841_cast_fp16 = slice_by_index(begin = var_51841_begin_0, end = var_51841_end_0, end_mask = var_51841_end_mask_0, x = v_229_cast_fp16)[name = tensor<string, []>("op_51841_cast_fp16")];
            tensor<int32, [4]> var_51845_begin_0 = const()[name = tensor<string, []>("op_51845_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_51845_end_0 = const()[name = tensor<string, []>("op_51845_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_51845_end_mask_0 = const()[name = tensor<string, []>("op_51845_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51845_cast_fp16 = slice_by_index(begin = var_51845_begin_0, end = var_51845_end_0, end_mask = var_51845_end_mask_0, x = v_229_cast_fp16)[name = tensor<string, []>("op_51845_cast_fp16")];
            tensor<int32, [4]> var_51849_begin_0 = const()[name = tensor<string, []>("op_51849_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_51849_end_0 = const()[name = tensor<string, []>("op_51849_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_51849_end_mask_0 = const()[name = tensor<string, []>("op_51849_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51849_cast_fp16 = slice_by_index(begin = var_51849_begin_0, end = var_51849_end_0, end_mask = var_51849_end_mask_0, x = v_229_cast_fp16)[name = tensor<string, []>("op_51849_cast_fp16")];
            tensor<int32, [4]> var_51853_begin_0 = const()[name = tensor<string, []>("op_51853_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_51853_end_0 = const()[name = tensor<string, []>("op_51853_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_51853_end_mask_0 = const()[name = tensor<string, []>("op_51853_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51853_cast_fp16 = slice_by_index(begin = var_51853_begin_0, end = var_51853_end_0, end_mask = var_51853_end_mask_0, x = v_229_cast_fp16)[name = tensor<string, []>("op_51853_cast_fp16")];
            tensor<int32, [4]> var_51857_begin_0 = const()[name = tensor<string, []>("op_51857_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_51857_end_0 = const()[name = tensor<string, []>("op_51857_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_51857_end_mask_0 = const()[name = tensor<string, []>("op_51857_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51857_cast_fp16 = slice_by_index(begin = var_51857_begin_0, end = var_51857_end_0, end_mask = var_51857_end_mask_0, x = v_229_cast_fp16)[name = tensor<string, []>("op_51857_cast_fp16")];
            tensor<int32, [4]> var_51861_begin_0 = const()[name = tensor<string, []>("op_51861_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_51861_end_0 = const()[name = tensor<string, []>("op_51861_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_51861_end_mask_0 = const()[name = tensor<string, []>("op_51861_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51861_cast_fp16 = slice_by_index(begin = var_51861_begin_0, end = var_51861_end_0, end_mask = var_51861_end_mask_0, x = v_229_cast_fp16)[name = tensor<string, []>("op_51861_cast_fp16")];
            tensor<int32, [4]> var_51865_begin_0 = const()[name = tensor<string, []>("op_51865_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_51865_end_0 = const()[name = tensor<string, []>("op_51865_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_51865_end_mask_0 = const()[name = tensor<string, []>("op_51865_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51865_cast_fp16 = slice_by_index(begin = var_51865_begin_0, end = var_51865_end_0, end_mask = var_51865_end_mask_0, x = v_229_cast_fp16)[name = tensor<string, []>("op_51865_cast_fp16")];
            tensor<int32, [4]> var_51869_begin_0 = const()[name = tensor<string, []>("op_51869_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_51869_end_0 = const()[name = tensor<string, []>("op_51869_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_51869_end_mask_0 = const()[name = tensor<string, []>("op_51869_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51869_cast_fp16 = slice_by_index(begin = var_51869_begin_0, end = var_51869_end_0, end_mask = var_51869_end_mask_0, x = v_229_cast_fp16)[name = tensor<string, []>("op_51869_cast_fp16")];
            tensor<int32, [4]> var_51873_begin_0 = const()[name = tensor<string, []>("op_51873_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_51873_end_0 = const()[name = tensor<string, []>("op_51873_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_51873_end_mask_0 = const()[name = tensor<string, []>("op_51873_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51873_cast_fp16 = slice_by_index(begin = var_51873_begin_0, end = var_51873_end_0, end_mask = var_51873_end_mask_0, x = v_229_cast_fp16)[name = tensor<string, []>("op_51873_cast_fp16")];
            tensor<int32, [4]> var_51877_begin_0 = const()[name = tensor<string, []>("op_51877_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_51877_end_0 = const()[name = tensor<string, []>("op_51877_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_51877_end_mask_0 = const()[name = tensor<string, []>("op_51877_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51877_cast_fp16 = slice_by_index(begin = var_51877_begin_0, end = var_51877_end_0, end_mask = var_51877_end_mask_0, x = v_229_cast_fp16)[name = tensor<string, []>("op_51877_cast_fp16")];
            tensor<int32, [4]> var_51881_begin_0 = const()[name = tensor<string, []>("op_51881_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_51881_end_0 = const()[name = tensor<string, []>("op_51881_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_51881_end_mask_0 = const()[name = tensor<string, []>("op_51881_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51881_cast_fp16 = slice_by_index(begin = var_51881_begin_0, end = var_51881_end_0, end_mask = var_51881_end_mask_0, x = v_229_cast_fp16)[name = tensor<string, []>("op_51881_cast_fp16")];
            tensor<int32, [4]> var_51885_begin_0 = const()[name = tensor<string, []>("op_51885_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_51885_end_0 = const()[name = tensor<string, []>("op_51885_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_51885_end_mask_0 = const()[name = tensor<string, []>("op_51885_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51885_cast_fp16 = slice_by_index(begin = var_51885_begin_0, end = var_51885_end_0, end_mask = var_51885_end_mask_0, x = v_229_cast_fp16)[name = tensor<string, []>("op_51885_cast_fp16")];
            tensor<int32, [4]> var_51889_begin_0 = const()[name = tensor<string, []>("op_51889_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_51889_end_0 = const()[name = tensor<string, []>("op_51889_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_51889_end_mask_0 = const()[name = tensor<string, []>("op_51889_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51889_cast_fp16 = slice_by_index(begin = var_51889_begin_0, end = var_51889_end_0, end_mask = var_51889_end_mask_0, x = v_229_cast_fp16)[name = tensor<string, []>("op_51889_cast_fp16")];
            tensor<int32, [4]> var_51893_begin_0 = const()[name = tensor<string, []>("op_51893_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_51893_end_0 = const()[name = tensor<string, []>("op_51893_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_51893_end_mask_0 = const()[name = tensor<string, []>("op_51893_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51893_cast_fp16 = slice_by_index(begin = var_51893_begin_0, end = var_51893_end_0, end_mask = var_51893_end_mask_0, x = v_229_cast_fp16)[name = tensor<string, []>("op_51893_cast_fp16")];
            tensor<int32, [4]> var_51897_begin_0 = const()[name = tensor<string, []>("op_51897_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_51897_end_0 = const()[name = tensor<string, []>("op_51897_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_51897_end_mask_0 = const()[name = tensor<string, []>("op_51897_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51897_cast_fp16 = slice_by_index(begin = var_51897_begin_0, end = var_51897_end_0, end_mask = var_51897_end_mask_0, x = v_229_cast_fp16)[name = tensor<string, []>("op_51897_cast_fp16")];
            tensor<string, []> var_51901_equation_0 = const()[name = tensor<string, []>("op_51901_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51901_cast_fp16 = einsum(equation = var_51901_equation_0, values = (var_51743_cast_fp16, var_51660_cast_fp16))[name = tensor<string, []>("op_51901_cast_fp16")];
            tensor<fp16, []> var_51902_to_fp16 = const()[name = tensor<string, []>("op_51902_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4401_cast_fp16 = mul(x = var_51901_cast_fp16, y = var_51902_to_fp16)[name = tensor<string, []>("aw_4401_cast_fp16")];
            tensor<string, []> var_51905_equation_0 = const()[name = tensor<string, []>("op_51905_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51905_cast_fp16 = einsum(equation = var_51905_equation_0, values = (var_51747_cast_fp16, var_51664_cast_fp16))[name = tensor<string, []>("op_51905_cast_fp16")];
            tensor<fp16, []> var_51906_to_fp16 = const()[name = tensor<string, []>("op_51906_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4403_cast_fp16 = mul(x = var_51905_cast_fp16, y = var_51906_to_fp16)[name = tensor<string, []>("aw_4403_cast_fp16")];
            tensor<string, []> var_51909_equation_0 = const()[name = tensor<string, []>("op_51909_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51909_cast_fp16 = einsum(equation = var_51909_equation_0, values = (var_51751_cast_fp16, var_51668_cast_fp16))[name = tensor<string, []>("op_51909_cast_fp16")];
            tensor<fp16, []> var_51910_to_fp16 = const()[name = tensor<string, []>("op_51910_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4405_cast_fp16 = mul(x = var_51909_cast_fp16, y = var_51910_to_fp16)[name = tensor<string, []>("aw_4405_cast_fp16")];
            tensor<string, []> var_51913_equation_0 = const()[name = tensor<string, []>("op_51913_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51913_cast_fp16 = einsum(equation = var_51913_equation_0, values = (var_51755_cast_fp16, var_51672_cast_fp16))[name = tensor<string, []>("op_51913_cast_fp16")];
            tensor<fp16, []> var_51914_to_fp16 = const()[name = tensor<string, []>("op_51914_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4407_cast_fp16 = mul(x = var_51913_cast_fp16, y = var_51914_to_fp16)[name = tensor<string, []>("aw_4407_cast_fp16")];
            tensor<string, []> var_51917_equation_0 = const()[name = tensor<string, []>("op_51917_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51917_cast_fp16 = einsum(equation = var_51917_equation_0, values = (var_51759_cast_fp16, var_51676_cast_fp16))[name = tensor<string, []>("op_51917_cast_fp16")];
            tensor<fp16, []> var_51918_to_fp16 = const()[name = tensor<string, []>("op_51918_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4409_cast_fp16 = mul(x = var_51917_cast_fp16, y = var_51918_to_fp16)[name = tensor<string, []>("aw_4409_cast_fp16")];
            tensor<string, []> var_51921_equation_0 = const()[name = tensor<string, []>("op_51921_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51921_cast_fp16 = einsum(equation = var_51921_equation_0, values = (var_51763_cast_fp16, var_51680_cast_fp16))[name = tensor<string, []>("op_51921_cast_fp16")];
            tensor<fp16, []> var_51922_to_fp16 = const()[name = tensor<string, []>("op_51922_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4411_cast_fp16 = mul(x = var_51921_cast_fp16, y = var_51922_to_fp16)[name = tensor<string, []>("aw_4411_cast_fp16")];
            tensor<string, []> var_51925_equation_0 = const()[name = tensor<string, []>("op_51925_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51925_cast_fp16 = einsum(equation = var_51925_equation_0, values = (var_51767_cast_fp16, var_51684_cast_fp16))[name = tensor<string, []>("op_51925_cast_fp16")];
            tensor<fp16, []> var_51926_to_fp16 = const()[name = tensor<string, []>("op_51926_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4413_cast_fp16 = mul(x = var_51925_cast_fp16, y = var_51926_to_fp16)[name = tensor<string, []>("aw_4413_cast_fp16")];
            tensor<string, []> var_51929_equation_0 = const()[name = tensor<string, []>("op_51929_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51929_cast_fp16 = einsum(equation = var_51929_equation_0, values = (var_51771_cast_fp16, var_51688_cast_fp16))[name = tensor<string, []>("op_51929_cast_fp16")];
            tensor<fp16, []> var_51930_to_fp16 = const()[name = tensor<string, []>("op_51930_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4415_cast_fp16 = mul(x = var_51929_cast_fp16, y = var_51930_to_fp16)[name = tensor<string, []>("aw_4415_cast_fp16")];
            tensor<string, []> var_51933_equation_0 = const()[name = tensor<string, []>("op_51933_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51933_cast_fp16 = einsum(equation = var_51933_equation_0, values = (var_51775_cast_fp16, var_51692_cast_fp16))[name = tensor<string, []>("op_51933_cast_fp16")];
            tensor<fp16, []> var_51934_to_fp16 = const()[name = tensor<string, []>("op_51934_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4417_cast_fp16 = mul(x = var_51933_cast_fp16, y = var_51934_to_fp16)[name = tensor<string, []>("aw_4417_cast_fp16")];
            tensor<string, []> var_51937_equation_0 = const()[name = tensor<string, []>("op_51937_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51937_cast_fp16 = einsum(equation = var_51937_equation_0, values = (var_51779_cast_fp16, var_51696_cast_fp16))[name = tensor<string, []>("op_51937_cast_fp16")];
            tensor<fp16, []> var_51938_to_fp16 = const()[name = tensor<string, []>("op_51938_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4419_cast_fp16 = mul(x = var_51937_cast_fp16, y = var_51938_to_fp16)[name = tensor<string, []>("aw_4419_cast_fp16")];
            tensor<string, []> var_51941_equation_0 = const()[name = tensor<string, []>("op_51941_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51941_cast_fp16 = einsum(equation = var_51941_equation_0, values = (var_51783_cast_fp16, var_51700_cast_fp16))[name = tensor<string, []>("op_51941_cast_fp16")];
            tensor<fp16, []> var_51942_to_fp16 = const()[name = tensor<string, []>("op_51942_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4421_cast_fp16 = mul(x = var_51941_cast_fp16, y = var_51942_to_fp16)[name = tensor<string, []>("aw_4421_cast_fp16")];
            tensor<string, []> var_51945_equation_0 = const()[name = tensor<string, []>("op_51945_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51945_cast_fp16 = einsum(equation = var_51945_equation_0, values = (var_51787_cast_fp16, var_51704_cast_fp16))[name = tensor<string, []>("op_51945_cast_fp16")];
            tensor<fp16, []> var_51946_to_fp16 = const()[name = tensor<string, []>("op_51946_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4423_cast_fp16 = mul(x = var_51945_cast_fp16, y = var_51946_to_fp16)[name = tensor<string, []>("aw_4423_cast_fp16")];
            tensor<string, []> var_51949_equation_0 = const()[name = tensor<string, []>("op_51949_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51949_cast_fp16 = einsum(equation = var_51949_equation_0, values = (var_51791_cast_fp16, var_51708_cast_fp16))[name = tensor<string, []>("op_51949_cast_fp16")];
            tensor<fp16, []> var_51950_to_fp16 = const()[name = tensor<string, []>("op_51950_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4425_cast_fp16 = mul(x = var_51949_cast_fp16, y = var_51950_to_fp16)[name = tensor<string, []>("aw_4425_cast_fp16")];
            tensor<string, []> var_51953_equation_0 = const()[name = tensor<string, []>("op_51953_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51953_cast_fp16 = einsum(equation = var_51953_equation_0, values = (var_51795_cast_fp16, var_51712_cast_fp16))[name = tensor<string, []>("op_51953_cast_fp16")];
            tensor<fp16, []> var_51954_to_fp16 = const()[name = tensor<string, []>("op_51954_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4427_cast_fp16 = mul(x = var_51953_cast_fp16, y = var_51954_to_fp16)[name = tensor<string, []>("aw_4427_cast_fp16")];
            tensor<string, []> var_51957_equation_0 = const()[name = tensor<string, []>("op_51957_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51957_cast_fp16 = einsum(equation = var_51957_equation_0, values = (var_51799_cast_fp16, var_51716_cast_fp16))[name = tensor<string, []>("op_51957_cast_fp16")];
            tensor<fp16, []> var_51958_to_fp16 = const()[name = tensor<string, []>("op_51958_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4429_cast_fp16 = mul(x = var_51957_cast_fp16, y = var_51958_to_fp16)[name = tensor<string, []>("aw_4429_cast_fp16")];
            tensor<string, []> var_51961_equation_0 = const()[name = tensor<string, []>("op_51961_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51961_cast_fp16 = einsum(equation = var_51961_equation_0, values = (var_51803_cast_fp16, var_51720_cast_fp16))[name = tensor<string, []>("op_51961_cast_fp16")];
            tensor<fp16, []> var_51962_to_fp16 = const()[name = tensor<string, []>("op_51962_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4431_cast_fp16 = mul(x = var_51961_cast_fp16, y = var_51962_to_fp16)[name = tensor<string, []>("aw_4431_cast_fp16")];
            tensor<string, []> var_51965_equation_0 = const()[name = tensor<string, []>("op_51965_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51965_cast_fp16 = einsum(equation = var_51965_equation_0, values = (var_51807_cast_fp16, var_51724_cast_fp16))[name = tensor<string, []>("op_51965_cast_fp16")];
            tensor<fp16, []> var_51966_to_fp16 = const()[name = tensor<string, []>("op_51966_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4433_cast_fp16 = mul(x = var_51965_cast_fp16, y = var_51966_to_fp16)[name = tensor<string, []>("aw_4433_cast_fp16")];
            tensor<string, []> var_51969_equation_0 = const()[name = tensor<string, []>("op_51969_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51969_cast_fp16 = einsum(equation = var_51969_equation_0, values = (var_51811_cast_fp16, var_51728_cast_fp16))[name = tensor<string, []>("op_51969_cast_fp16")];
            tensor<fp16, []> var_51970_to_fp16 = const()[name = tensor<string, []>("op_51970_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4435_cast_fp16 = mul(x = var_51969_cast_fp16, y = var_51970_to_fp16)[name = tensor<string, []>("aw_4435_cast_fp16")];
            tensor<string, []> var_51973_equation_0 = const()[name = tensor<string, []>("op_51973_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51973_cast_fp16 = einsum(equation = var_51973_equation_0, values = (var_51815_cast_fp16, var_51732_cast_fp16))[name = tensor<string, []>("op_51973_cast_fp16")];
            tensor<fp16, []> var_51974_to_fp16 = const()[name = tensor<string, []>("op_51974_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4437_cast_fp16 = mul(x = var_51973_cast_fp16, y = var_51974_to_fp16)[name = tensor<string, []>("aw_4437_cast_fp16")];
            tensor<string, []> var_51977_equation_0 = const()[name = tensor<string, []>("op_51977_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51977_cast_fp16 = einsum(equation = var_51977_equation_0, values = (var_51819_cast_fp16, var_51736_cast_fp16))[name = tensor<string, []>("op_51977_cast_fp16")];
            tensor<fp16, []> var_51978_to_fp16 = const()[name = tensor<string, []>("op_51978_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4439_cast_fp16 = mul(x = var_51977_cast_fp16, y = var_51978_to_fp16)[name = tensor<string, []>("aw_4439_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51980_cast_fp16 = softmax(axis = var_30355, x = aw_4401_cast_fp16)[name = tensor<string, []>("op_51980_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51981_cast_fp16 = softmax(axis = var_30355, x = aw_4403_cast_fp16)[name = tensor<string, []>("op_51981_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51982_cast_fp16 = softmax(axis = var_30355, x = aw_4405_cast_fp16)[name = tensor<string, []>("op_51982_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51983_cast_fp16 = softmax(axis = var_30355, x = aw_4407_cast_fp16)[name = tensor<string, []>("op_51983_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51984_cast_fp16 = softmax(axis = var_30355, x = aw_4409_cast_fp16)[name = tensor<string, []>("op_51984_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51985_cast_fp16 = softmax(axis = var_30355, x = aw_4411_cast_fp16)[name = tensor<string, []>("op_51985_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51986_cast_fp16 = softmax(axis = var_30355, x = aw_4413_cast_fp16)[name = tensor<string, []>("op_51986_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51987_cast_fp16 = softmax(axis = var_30355, x = aw_4415_cast_fp16)[name = tensor<string, []>("op_51987_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51988_cast_fp16 = softmax(axis = var_30355, x = aw_4417_cast_fp16)[name = tensor<string, []>("op_51988_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51989_cast_fp16 = softmax(axis = var_30355, x = aw_4419_cast_fp16)[name = tensor<string, []>("op_51989_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51990_cast_fp16 = softmax(axis = var_30355, x = aw_4421_cast_fp16)[name = tensor<string, []>("op_51990_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51991_cast_fp16 = softmax(axis = var_30355, x = aw_4423_cast_fp16)[name = tensor<string, []>("op_51991_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51992_cast_fp16 = softmax(axis = var_30355, x = aw_4425_cast_fp16)[name = tensor<string, []>("op_51992_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51993_cast_fp16 = softmax(axis = var_30355, x = aw_4427_cast_fp16)[name = tensor<string, []>("op_51993_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51994_cast_fp16 = softmax(axis = var_30355, x = aw_4429_cast_fp16)[name = tensor<string, []>("op_51994_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51995_cast_fp16 = softmax(axis = var_30355, x = aw_4431_cast_fp16)[name = tensor<string, []>("op_51995_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51996_cast_fp16 = softmax(axis = var_30355, x = aw_4433_cast_fp16)[name = tensor<string, []>("op_51996_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51997_cast_fp16 = softmax(axis = var_30355, x = aw_4435_cast_fp16)[name = tensor<string, []>("op_51997_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51998_cast_fp16 = softmax(axis = var_30355, x = aw_4437_cast_fp16)[name = tensor<string, []>("op_51998_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51999_cast_fp16 = softmax(axis = var_30355, x = aw_4439_cast_fp16)[name = tensor<string, []>("op_51999_cast_fp16")];
            tensor<string, []> var_52001_equation_0 = const()[name = tensor<string, []>("op_52001_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52001_cast_fp16 = einsum(equation = var_52001_equation_0, values = (var_51821_cast_fp16, var_51980_cast_fp16))[name = tensor<string, []>("op_52001_cast_fp16")];
            tensor<string, []> var_52003_equation_0 = const()[name = tensor<string, []>("op_52003_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52003_cast_fp16 = einsum(equation = var_52003_equation_0, values = (var_51825_cast_fp16, var_51981_cast_fp16))[name = tensor<string, []>("op_52003_cast_fp16")];
            tensor<string, []> var_52005_equation_0 = const()[name = tensor<string, []>("op_52005_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52005_cast_fp16 = einsum(equation = var_52005_equation_0, values = (var_51829_cast_fp16, var_51982_cast_fp16))[name = tensor<string, []>("op_52005_cast_fp16")];
            tensor<string, []> var_52007_equation_0 = const()[name = tensor<string, []>("op_52007_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52007_cast_fp16 = einsum(equation = var_52007_equation_0, values = (var_51833_cast_fp16, var_51983_cast_fp16))[name = tensor<string, []>("op_52007_cast_fp16")];
            tensor<string, []> var_52009_equation_0 = const()[name = tensor<string, []>("op_52009_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52009_cast_fp16 = einsum(equation = var_52009_equation_0, values = (var_51837_cast_fp16, var_51984_cast_fp16))[name = tensor<string, []>("op_52009_cast_fp16")];
            tensor<string, []> var_52011_equation_0 = const()[name = tensor<string, []>("op_52011_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52011_cast_fp16 = einsum(equation = var_52011_equation_0, values = (var_51841_cast_fp16, var_51985_cast_fp16))[name = tensor<string, []>("op_52011_cast_fp16")];
            tensor<string, []> var_52013_equation_0 = const()[name = tensor<string, []>("op_52013_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52013_cast_fp16 = einsum(equation = var_52013_equation_0, values = (var_51845_cast_fp16, var_51986_cast_fp16))[name = tensor<string, []>("op_52013_cast_fp16")];
            tensor<string, []> var_52015_equation_0 = const()[name = tensor<string, []>("op_52015_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52015_cast_fp16 = einsum(equation = var_52015_equation_0, values = (var_51849_cast_fp16, var_51987_cast_fp16))[name = tensor<string, []>("op_52015_cast_fp16")];
            tensor<string, []> var_52017_equation_0 = const()[name = tensor<string, []>("op_52017_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52017_cast_fp16 = einsum(equation = var_52017_equation_0, values = (var_51853_cast_fp16, var_51988_cast_fp16))[name = tensor<string, []>("op_52017_cast_fp16")];
            tensor<string, []> var_52019_equation_0 = const()[name = tensor<string, []>("op_52019_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52019_cast_fp16 = einsum(equation = var_52019_equation_0, values = (var_51857_cast_fp16, var_51989_cast_fp16))[name = tensor<string, []>("op_52019_cast_fp16")];
            tensor<string, []> var_52021_equation_0 = const()[name = tensor<string, []>("op_52021_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52021_cast_fp16 = einsum(equation = var_52021_equation_0, values = (var_51861_cast_fp16, var_51990_cast_fp16))[name = tensor<string, []>("op_52021_cast_fp16")];
            tensor<string, []> var_52023_equation_0 = const()[name = tensor<string, []>("op_52023_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52023_cast_fp16 = einsum(equation = var_52023_equation_0, values = (var_51865_cast_fp16, var_51991_cast_fp16))[name = tensor<string, []>("op_52023_cast_fp16")];
            tensor<string, []> var_52025_equation_0 = const()[name = tensor<string, []>("op_52025_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52025_cast_fp16 = einsum(equation = var_52025_equation_0, values = (var_51869_cast_fp16, var_51992_cast_fp16))[name = tensor<string, []>("op_52025_cast_fp16")];
            tensor<string, []> var_52027_equation_0 = const()[name = tensor<string, []>("op_52027_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52027_cast_fp16 = einsum(equation = var_52027_equation_0, values = (var_51873_cast_fp16, var_51993_cast_fp16))[name = tensor<string, []>("op_52027_cast_fp16")];
            tensor<string, []> var_52029_equation_0 = const()[name = tensor<string, []>("op_52029_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52029_cast_fp16 = einsum(equation = var_52029_equation_0, values = (var_51877_cast_fp16, var_51994_cast_fp16))[name = tensor<string, []>("op_52029_cast_fp16")];
            tensor<string, []> var_52031_equation_0 = const()[name = tensor<string, []>("op_52031_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52031_cast_fp16 = einsum(equation = var_52031_equation_0, values = (var_51881_cast_fp16, var_51995_cast_fp16))[name = tensor<string, []>("op_52031_cast_fp16")];
            tensor<string, []> var_52033_equation_0 = const()[name = tensor<string, []>("op_52033_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52033_cast_fp16 = einsum(equation = var_52033_equation_0, values = (var_51885_cast_fp16, var_51996_cast_fp16))[name = tensor<string, []>("op_52033_cast_fp16")];
            tensor<string, []> var_52035_equation_0 = const()[name = tensor<string, []>("op_52035_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52035_cast_fp16 = einsum(equation = var_52035_equation_0, values = (var_51889_cast_fp16, var_51997_cast_fp16))[name = tensor<string, []>("op_52035_cast_fp16")];
            tensor<string, []> var_52037_equation_0 = const()[name = tensor<string, []>("op_52037_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52037_cast_fp16 = einsum(equation = var_52037_equation_0, values = (var_51893_cast_fp16, var_51998_cast_fp16))[name = tensor<string, []>("op_52037_cast_fp16")];
            tensor<string, []> var_52039_equation_0 = const()[name = tensor<string, []>("op_52039_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52039_cast_fp16 = einsum(equation = var_52039_equation_0, values = (var_51897_cast_fp16, var_51999_cast_fp16))[name = tensor<string, []>("op_52039_cast_fp16")];
            tensor<bool, []> input_659_interleave_0 = const()[name = tensor<string, []>("input_659_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_659_cast_fp16 = concat(axis = var_30355, interleave = input_659_interleave_0, values = (var_52001_cast_fp16, var_52003_cast_fp16, var_52005_cast_fp16, var_52007_cast_fp16, var_52009_cast_fp16, var_52011_cast_fp16, var_52013_cast_fp16, var_52015_cast_fp16, var_52017_cast_fp16, var_52019_cast_fp16, var_52021_cast_fp16, var_52023_cast_fp16, var_52025_cast_fp16, var_52027_cast_fp16, var_52029_cast_fp16, var_52031_cast_fp16, var_52033_cast_fp16, var_52035_cast_fp16, var_52037_cast_fp16, var_52039_cast_fp16))[name = tensor<string, []>("input_659_cast_fp16")];
            tensor<string, []> var_52049_pad_type_0 = const()[name = tensor<string, []>("op_52049_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_52049_strides_0 = const()[name = tensor<string, []>("op_52049_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_52049_pad_0 = const()[name = tensor<string, []>("op_52049_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_52049_dilations_0 = const()[name = tensor<string, []>("op_52049_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_52049_groups_0 = const()[name = tensor<string, []>("op_52049_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_3_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(680227776))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(681456640))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_3_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_3_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_3_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(681456832)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_52049_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_3_attn1_to_out_0_bias_to_fp16, dilations = var_52049_dilations_0, groups = var_52049_groups_0, pad = var_52049_pad_0, pad_type = var_52049_pad_type_0, strides = var_52049_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_3_attn1_to_out_0_weight_to_fp16_palettized, x = input_659_cast_fp16)[name = tensor<string, []>("op_52049_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_345_cast_fp16 = add(x = var_52049_cast_fp16, y = inputs_343_cast_fp16)[name = tensor<string, []>("inputs_345_cast_fp16")];
            tensor<int32, [1]> hidden_states_455_axes_0 = const()[name = tensor<string, []>("hidden_states_455_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_455_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_455_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(681459456)))];
            tensor<fp16, [1280]> hidden_states_455_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_455_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(681462080)))];
            tensor<fp16, []> var_52059_to_fp16 = const()[name = tensor<string, []>("op_52059_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_455_cast_fp16 = layer_norm(axes = hidden_states_455_axes_0, beta = hidden_states_455_beta_0_to_fp16, epsilon = var_52059_to_fp16, gamma = hidden_states_455_gamma_0_to_fp16, x = inputs_345_cast_fp16)[name = tensor<string, []>("hidden_states_455_cast_fp16")];
            tensor<string, []> q_231_pad_type_0 = const()[name = tensor<string, []>("q_231_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_231_strides_0 = const()[name = tensor<string, []>("q_231_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_231_pad_0 = const()[name = tensor<string, []>("q_231_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_231_dilations_0 = const()[name = tensor<string, []>("q_231_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_231_groups_0 = const()[name = tensor<string, []>("q_231_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_3_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(681464704))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(682693568))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_3_attn2_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_231_cast_fp16 = conv(dilations = q_231_dilations_0, groups = q_231_groups_0, pad = q_231_pad_0, pad_type = q_231_pad_type_0, strides = q_231_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_3_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_455_cast_fp16)[name = tensor<string, []>("q_231_cast_fp16")];
            tensor<string, []> k_461_pad_type_0 = const()[name = tensor<string, []>("k_461_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_461_strides_0 = const()[name = tensor<string, []>("k_461_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_461_pad_0 = const()[name = tensor<string, []>("k_461_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_461_dilations_0 = const()[name = tensor<string, []>("k_461_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_461_groups_0 = const()[name = tensor<string, []>("k_461_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_3_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(682693760))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(684659904))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_3_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> k_461_cast_fp16 = conv(dilations = k_461_dilations_0, groups = k_461_groups_0, pad = k_461_pad_0, pad_type = k_461_pad_type_0, strides = k_461_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_3_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("k_461_cast_fp16")];
            tensor<string, []> v_231_pad_type_0 = const()[name = tensor<string, []>("v_231_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_231_strides_0 = const()[name = tensor<string, []>("v_231_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_231_pad_0 = const()[name = tensor<string, []>("v_231_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_231_dilations_0 = const()[name = tensor<string, []>("v_231_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_231_groups_0 = const()[name = tensor<string, []>("v_231_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_3_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(684660096))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(686626240))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_3_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> v_231_cast_fp16 = conv(dilations = v_231_dilations_0, groups = v_231_groups_0, pad = v_231_pad_0, pad_type = v_231_pad_type_0, strides = v_231_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_3_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("v_231_cast_fp16")];
            tensor<int32, [4]> var_52092_begin_0 = const()[name = tensor<string, []>("op_52092_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_52092_end_0 = const()[name = tensor<string, []>("op_52092_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_52092_end_mask_0 = const()[name = tensor<string, []>("op_52092_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52092_cast_fp16 = slice_by_index(begin = var_52092_begin_0, end = var_52092_end_0, end_mask = var_52092_end_mask_0, x = q_231_cast_fp16)[name = tensor<string, []>("op_52092_cast_fp16")];
            tensor<int32, [4]> var_52096_begin_0 = const()[name = tensor<string, []>("op_52096_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_52096_end_0 = const()[name = tensor<string, []>("op_52096_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_52096_end_mask_0 = const()[name = tensor<string, []>("op_52096_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52096_cast_fp16 = slice_by_index(begin = var_52096_begin_0, end = var_52096_end_0, end_mask = var_52096_end_mask_0, x = q_231_cast_fp16)[name = tensor<string, []>("op_52096_cast_fp16")];
            tensor<int32, [4]> var_52100_begin_0 = const()[name = tensor<string, []>("op_52100_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_52100_end_0 = const()[name = tensor<string, []>("op_52100_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_52100_end_mask_0 = const()[name = tensor<string, []>("op_52100_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52100_cast_fp16 = slice_by_index(begin = var_52100_begin_0, end = var_52100_end_0, end_mask = var_52100_end_mask_0, x = q_231_cast_fp16)[name = tensor<string, []>("op_52100_cast_fp16")];
            tensor<int32, [4]> var_52104_begin_0 = const()[name = tensor<string, []>("op_52104_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_52104_end_0 = const()[name = tensor<string, []>("op_52104_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_52104_end_mask_0 = const()[name = tensor<string, []>("op_52104_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52104_cast_fp16 = slice_by_index(begin = var_52104_begin_0, end = var_52104_end_0, end_mask = var_52104_end_mask_0, x = q_231_cast_fp16)[name = tensor<string, []>("op_52104_cast_fp16")];
            tensor<int32, [4]> var_52108_begin_0 = const()[name = tensor<string, []>("op_52108_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_52108_end_0 = const()[name = tensor<string, []>("op_52108_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_52108_end_mask_0 = const()[name = tensor<string, []>("op_52108_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52108_cast_fp16 = slice_by_index(begin = var_52108_begin_0, end = var_52108_end_0, end_mask = var_52108_end_mask_0, x = q_231_cast_fp16)[name = tensor<string, []>("op_52108_cast_fp16")];
            tensor<int32, [4]> var_52112_begin_0 = const()[name = tensor<string, []>("op_52112_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_52112_end_0 = const()[name = tensor<string, []>("op_52112_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_52112_end_mask_0 = const()[name = tensor<string, []>("op_52112_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52112_cast_fp16 = slice_by_index(begin = var_52112_begin_0, end = var_52112_end_0, end_mask = var_52112_end_mask_0, x = q_231_cast_fp16)[name = tensor<string, []>("op_52112_cast_fp16")];
            tensor<int32, [4]> var_52116_begin_0 = const()[name = tensor<string, []>("op_52116_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_52116_end_0 = const()[name = tensor<string, []>("op_52116_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_52116_end_mask_0 = const()[name = tensor<string, []>("op_52116_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52116_cast_fp16 = slice_by_index(begin = var_52116_begin_0, end = var_52116_end_0, end_mask = var_52116_end_mask_0, x = q_231_cast_fp16)[name = tensor<string, []>("op_52116_cast_fp16")];
            tensor<int32, [4]> var_52120_begin_0 = const()[name = tensor<string, []>("op_52120_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_52120_end_0 = const()[name = tensor<string, []>("op_52120_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_52120_end_mask_0 = const()[name = tensor<string, []>("op_52120_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52120_cast_fp16 = slice_by_index(begin = var_52120_begin_0, end = var_52120_end_0, end_mask = var_52120_end_mask_0, x = q_231_cast_fp16)[name = tensor<string, []>("op_52120_cast_fp16")];
            tensor<int32, [4]> var_52124_begin_0 = const()[name = tensor<string, []>("op_52124_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_52124_end_0 = const()[name = tensor<string, []>("op_52124_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_52124_end_mask_0 = const()[name = tensor<string, []>("op_52124_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52124_cast_fp16 = slice_by_index(begin = var_52124_begin_0, end = var_52124_end_0, end_mask = var_52124_end_mask_0, x = q_231_cast_fp16)[name = tensor<string, []>("op_52124_cast_fp16")];
            tensor<int32, [4]> var_52128_begin_0 = const()[name = tensor<string, []>("op_52128_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_52128_end_0 = const()[name = tensor<string, []>("op_52128_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_52128_end_mask_0 = const()[name = tensor<string, []>("op_52128_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52128_cast_fp16 = slice_by_index(begin = var_52128_begin_0, end = var_52128_end_0, end_mask = var_52128_end_mask_0, x = q_231_cast_fp16)[name = tensor<string, []>("op_52128_cast_fp16")];
            tensor<int32, [4]> var_52132_begin_0 = const()[name = tensor<string, []>("op_52132_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_52132_end_0 = const()[name = tensor<string, []>("op_52132_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_52132_end_mask_0 = const()[name = tensor<string, []>("op_52132_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52132_cast_fp16 = slice_by_index(begin = var_52132_begin_0, end = var_52132_end_0, end_mask = var_52132_end_mask_0, x = q_231_cast_fp16)[name = tensor<string, []>("op_52132_cast_fp16")];
            tensor<int32, [4]> var_52136_begin_0 = const()[name = tensor<string, []>("op_52136_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_52136_end_0 = const()[name = tensor<string, []>("op_52136_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_52136_end_mask_0 = const()[name = tensor<string, []>("op_52136_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52136_cast_fp16 = slice_by_index(begin = var_52136_begin_0, end = var_52136_end_0, end_mask = var_52136_end_mask_0, x = q_231_cast_fp16)[name = tensor<string, []>("op_52136_cast_fp16")];
            tensor<int32, [4]> var_52140_begin_0 = const()[name = tensor<string, []>("op_52140_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_52140_end_0 = const()[name = tensor<string, []>("op_52140_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_52140_end_mask_0 = const()[name = tensor<string, []>("op_52140_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52140_cast_fp16 = slice_by_index(begin = var_52140_begin_0, end = var_52140_end_0, end_mask = var_52140_end_mask_0, x = q_231_cast_fp16)[name = tensor<string, []>("op_52140_cast_fp16")];
            tensor<int32, [4]> var_52144_begin_0 = const()[name = tensor<string, []>("op_52144_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_52144_end_0 = const()[name = tensor<string, []>("op_52144_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_52144_end_mask_0 = const()[name = tensor<string, []>("op_52144_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52144_cast_fp16 = slice_by_index(begin = var_52144_begin_0, end = var_52144_end_0, end_mask = var_52144_end_mask_0, x = q_231_cast_fp16)[name = tensor<string, []>("op_52144_cast_fp16")];
            tensor<int32, [4]> var_52148_begin_0 = const()[name = tensor<string, []>("op_52148_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_52148_end_0 = const()[name = tensor<string, []>("op_52148_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_52148_end_mask_0 = const()[name = tensor<string, []>("op_52148_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52148_cast_fp16 = slice_by_index(begin = var_52148_begin_0, end = var_52148_end_0, end_mask = var_52148_end_mask_0, x = q_231_cast_fp16)[name = tensor<string, []>("op_52148_cast_fp16")];
            tensor<int32, [4]> var_52152_begin_0 = const()[name = tensor<string, []>("op_52152_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_52152_end_0 = const()[name = tensor<string, []>("op_52152_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_52152_end_mask_0 = const()[name = tensor<string, []>("op_52152_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52152_cast_fp16 = slice_by_index(begin = var_52152_begin_0, end = var_52152_end_0, end_mask = var_52152_end_mask_0, x = q_231_cast_fp16)[name = tensor<string, []>("op_52152_cast_fp16")];
            tensor<int32, [4]> var_52156_begin_0 = const()[name = tensor<string, []>("op_52156_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_52156_end_0 = const()[name = tensor<string, []>("op_52156_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_52156_end_mask_0 = const()[name = tensor<string, []>("op_52156_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52156_cast_fp16 = slice_by_index(begin = var_52156_begin_0, end = var_52156_end_0, end_mask = var_52156_end_mask_0, x = q_231_cast_fp16)[name = tensor<string, []>("op_52156_cast_fp16")];
            tensor<int32, [4]> var_52160_begin_0 = const()[name = tensor<string, []>("op_52160_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_52160_end_0 = const()[name = tensor<string, []>("op_52160_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_52160_end_mask_0 = const()[name = tensor<string, []>("op_52160_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52160_cast_fp16 = slice_by_index(begin = var_52160_begin_0, end = var_52160_end_0, end_mask = var_52160_end_mask_0, x = q_231_cast_fp16)[name = tensor<string, []>("op_52160_cast_fp16")];
            tensor<int32, [4]> var_52164_begin_0 = const()[name = tensor<string, []>("op_52164_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_52164_end_0 = const()[name = tensor<string, []>("op_52164_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_52164_end_mask_0 = const()[name = tensor<string, []>("op_52164_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52164_cast_fp16 = slice_by_index(begin = var_52164_begin_0, end = var_52164_end_0, end_mask = var_52164_end_mask_0, x = q_231_cast_fp16)[name = tensor<string, []>("op_52164_cast_fp16")];
            tensor<int32, [4]> var_52168_begin_0 = const()[name = tensor<string, []>("op_52168_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_52168_end_0 = const()[name = tensor<string, []>("op_52168_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_52168_end_mask_0 = const()[name = tensor<string, []>("op_52168_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52168_cast_fp16 = slice_by_index(begin = var_52168_begin_0, end = var_52168_end_0, end_mask = var_52168_end_mask_0, x = q_231_cast_fp16)[name = tensor<string, []>("op_52168_cast_fp16")];
            tensor<int32, [4]> k_463_perm_0 = const()[name = tensor<string, []>("k_463_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_52175_begin_0 = const()[name = tensor<string, []>("op_52175_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_52175_end_0 = const()[name = tensor<string, []>("op_52175_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_52175_end_mask_0 = const()[name = tensor<string, []>("op_52175_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> k_463_cast_fp16 = transpose(perm = k_463_perm_0, x = k_461_cast_fp16)[name = tensor<string, []>("transpose_24")];
            tensor<fp16, [2, 77, 1, 64]> var_52175_cast_fp16 = slice_by_index(begin = var_52175_begin_0, end = var_52175_end_0, end_mask = var_52175_end_mask_0, x = k_463_cast_fp16)[name = tensor<string, []>("op_52175_cast_fp16")];
            tensor<int32, [4]> var_52179_begin_0 = const()[name = tensor<string, []>("op_52179_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_52179_end_0 = const()[name = tensor<string, []>("op_52179_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_52179_end_mask_0 = const()[name = tensor<string, []>("op_52179_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_52179_cast_fp16 = slice_by_index(begin = var_52179_begin_0, end = var_52179_end_0, end_mask = var_52179_end_mask_0, x = k_463_cast_fp16)[name = tensor<string, []>("op_52179_cast_fp16")];
            tensor<int32, [4]> var_52183_begin_0 = const()[name = tensor<string, []>("op_52183_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_52183_end_0 = const()[name = tensor<string, []>("op_52183_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_52183_end_mask_0 = const()[name = tensor<string, []>("op_52183_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_52183_cast_fp16 = slice_by_index(begin = var_52183_begin_0, end = var_52183_end_0, end_mask = var_52183_end_mask_0, x = k_463_cast_fp16)[name = tensor<string, []>("op_52183_cast_fp16")];
            tensor<int32, [4]> var_52187_begin_0 = const()[name = tensor<string, []>("op_52187_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_52187_end_0 = const()[name = tensor<string, []>("op_52187_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_52187_end_mask_0 = const()[name = tensor<string, []>("op_52187_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_52187_cast_fp16 = slice_by_index(begin = var_52187_begin_0, end = var_52187_end_0, end_mask = var_52187_end_mask_0, x = k_463_cast_fp16)[name = tensor<string, []>("op_52187_cast_fp16")];
            tensor<int32, [4]> var_52191_begin_0 = const()[name = tensor<string, []>("op_52191_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_52191_end_0 = const()[name = tensor<string, []>("op_52191_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_52191_end_mask_0 = const()[name = tensor<string, []>("op_52191_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_52191_cast_fp16 = slice_by_index(begin = var_52191_begin_0, end = var_52191_end_0, end_mask = var_52191_end_mask_0, x = k_463_cast_fp16)[name = tensor<string, []>("op_52191_cast_fp16")];
            tensor<int32, [4]> var_52195_begin_0 = const()[name = tensor<string, []>("op_52195_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_52195_end_0 = const()[name = tensor<string, []>("op_52195_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_52195_end_mask_0 = const()[name = tensor<string, []>("op_52195_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_52195_cast_fp16 = slice_by_index(begin = var_52195_begin_0, end = var_52195_end_0, end_mask = var_52195_end_mask_0, x = k_463_cast_fp16)[name = tensor<string, []>("op_52195_cast_fp16")];
            tensor<int32, [4]> var_52199_begin_0 = const()[name = tensor<string, []>("op_52199_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_52199_end_0 = const()[name = tensor<string, []>("op_52199_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_52199_end_mask_0 = const()[name = tensor<string, []>("op_52199_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_52199_cast_fp16 = slice_by_index(begin = var_52199_begin_0, end = var_52199_end_0, end_mask = var_52199_end_mask_0, x = k_463_cast_fp16)[name = tensor<string, []>("op_52199_cast_fp16")];
            tensor<int32, [4]> var_52203_begin_0 = const()[name = tensor<string, []>("op_52203_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_52203_end_0 = const()[name = tensor<string, []>("op_52203_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_52203_end_mask_0 = const()[name = tensor<string, []>("op_52203_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_52203_cast_fp16 = slice_by_index(begin = var_52203_begin_0, end = var_52203_end_0, end_mask = var_52203_end_mask_0, x = k_463_cast_fp16)[name = tensor<string, []>("op_52203_cast_fp16")];
            tensor<int32, [4]> var_52207_begin_0 = const()[name = tensor<string, []>("op_52207_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_52207_end_0 = const()[name = tensor<string, []>("op_52207_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_52207_end_mask_0 = const()[name = tensor<string, []>("op_52207_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_52207_cast_fp16 = slice_by_index(begin = var_52207_begin_0, end = var_52207_end_0, end_mask = var_52207_end_mask_0, x = k_463_cast_fp16)[name = tensor<string, []>("op_52207_cast_fp16")];
            tensor<int32, [4]> var_52211_begin_0 = const()[name = tensor<string, []>("op_52211_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_52211_end_0 = const()[name = tensor<string, []>("op_52211_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_52211_end_mask_0 = const()[name = tensor<string, []>("op_52211_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_52211_cast_fp16 = slice_by_index(begin = var_52211_begin_0, end = var_52211_end_0, end_mask = var_52211_end_mask_0, x = k_463_cast_fp16)[name = tensor<string, []>("op_52211_cast_fp16")];
            tensor<int32, [4]> var_52215_begin_0 = const()[name = tensor<string, []>("op_52215_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_52215_end_0 = const()[name = tensor<string, []>("op_52215_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_52215_end_mask_0 = const()[name = tensor<string, []>("op_52215_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_52215_cast_fp16 = slice_by_index(begin = var_52215_begin_0, end = var_52215_end_0, end_mask = var_52215_end_mask_0, x = k_463_cast_fp16)[name = tensor<string, []>("op_52215_cast_fp16")];
            tensor<int32, [4]> var_52219_begin_0 = const()[name = tensor<string, []>("op_52219_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_52219_end_0 = const()[name = tensor<string, []>("op_52219_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_52219_end_mask_0 = const()[name = tensor<string, []>("op_52219_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_52219_cast_fp16 = slice_by_index(begin = var_52219_begin_0, end = var_52219_end_0, end_mask = var_52219_end_mask_0, x = k_463_cast_fp16)[name = tensor<string, []>("op_52219_cast_fp16")];
            tensor<int32, [4]> var_52223_begin_0 = const()[name = tensor<string, []>("op_52223_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_52223_end_0 = const()[name = tensor<string, []>("op_52223_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_52223_end_mask_0 = const()[name = tensor<string, []>("op_52223_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_52223_cast_fp16 = slice_by_index(begin = var_52223_begin_0, end = var_52223_end_0, end_mask = var_52223_end_mask_0, x = k_463_cast_fp16)[name = tensor<string, []>("op_52223_cast_fp16")];
            tensor<int32, [4]> var_52227_begin_0 = const()[name = tensor<string, []>("op_52227_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_52227_end_0 = const()[name = tensor<string, []>("op_52227_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_52227_end_mask_0 = const()[name = tensor<string, []>("op_52227_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_52227_cast_fp16 = slice_by_index(begin = var_52227_begin_0, end = var_52227_end_0, end_mask = var_52227_end_mask_0, x = k_463_cast_fp16)[name = tensor<string, []>("op_52227_cast_fp16")];
            tensor<int32, [4]> var_52231_begin_0 = const()[name = tensor<string, []>("op_52231_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_52231_end_0 = const()[name = tensor<string, []>("op_52231_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_52231_end_mask_0 = const()[name = tensor<string, []>("op_52231_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_52231_cast_fp16 = slice_by_index(begin = var_52231_begin_0, end = var_52231_end_0, end_mask = var_52231_end_mask_0, x = k_463_cast_fp16)[name = tensor<string, []>("op_52231_cast_fp16")];
            tensor<int32, [4]> var_52235_begin_0 = const()[name = tensor<string, []>("op_52235_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_52235_end_0 = const()[name = tensor<string, []>("op_52235_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_52235_end_mask_0 = const()[name = tensor<string, []>("op_52235_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_52235_cast_fp16 = slice_by_index(begin = var_52235_begin_0, end = var_52235_end_0, end_mask = var_52235_end_mask_0, x = k_463_cast_fp16)[name = tensor<string, []>("op_52235_cast_fp16")];
            tensor<int32, [4]> var_52239_begin_0 = const()[name = tensor<string, []>("op_52239_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_52239_end_0 = const()[name = tensor<string, []>("op_52239_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_52239_end_mask_0 = const()[name = tensor<string, []>("op_52239_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_52239_cast_fp16 = slice_by_index(begin = var_52239_begin_0, end = var_52239_end_0, end_mask = var_52239_end_mask_0, x = k_463_cast_fp16)[name = tensor<string, []>("op_52239_cast_fp16")];
            tensor<int32, [4]> var_52243_begin_0 = const()[name = tensor<string, []>("op_52243_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_52243_end_0 = const()[name = tensor<string, []>("op_52243_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_52243_end_mask_0 = const()[name = tensor<string, []>("op_52243_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_52243_cast_fp16 = slice_by_index(begin = var_52243_begin_0, end = var_52243_end_0, end_mask = var_52243_end_mask_0, x = k_463_cast_fp16)[name = tensor<string, []>("op_52243_cast_fp16")];
            tensor<int32, [4]> var_52247_begin_0 = const()[name = tensor<string, []>("op_52247_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_52247_end_0 = const()[name = tensor<string, []>("op_52247_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_52247_end_mask_0 = const()[name = tensor<string, []>("op_52247_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_52247_cast_fp16 = slice_by_index(begin = var_52247_begin_0, end = var_52247_end_0, end_mask = var_52247_end_mask_0, x = k_463_cast_fp16)[name = tensor<string, []>("op_52247_cast_fp16")];
            tensor<int32, [4]> var_52251_begin_0 = const()[name = tensor<string, []>("op_52251_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_52251_end_0 = const()[name = tensor<string, []>("op_52251_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_52251_end_mask_0 = const()[name = tensor<string, []>("op_52251_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_52251_cast_fp16 = slice_by_index(begin = var_52251_begin_0, end = var_52251_end_0, end_mask = var_52251_end_mask_0, x = k_463_cast_fp16)[name = tensor<string, []>("op_52251_cast_fp16")];
            tensor<int32, [4]> var_52253_begin_0 = const()[name = tensor<string, []>("op_52253_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_52253_end_0 = const()[name = tensor<string, []>("op_52253_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_52253_end_mask_0 = const()[name = tensor<string, []>("op_52253_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_52253_cast_fp16 = slice_by_index(begin = var_52253_begin_0, end = var_52253_end_0, end_mask = var_52253_end_mask_0, x = v_231_cast_fp16)[name = tensor<string, []>("op_52253_cast_fp16")];
            tensor<int32, [4]> var_52257_begin_0 = const()[name = tensor<string, []>("op_52257_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_52257_end_0 = const()[name = tensor<string, []>("op_52257_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_52257_end_mask_0 = const()[name = tensor<string, []>("op_52257_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_52257_cast_fp16 = slice_by_index(begin = var_52257_begin_0, end = var_52257_end_0, end_mask = var_52257_end_mask_0, x = v_231_cast_fp16)[name = tensor<string, []>("op_52257_cast_fp16")];
            tensor<int32, [4]> var_52261_begin_0 = const()[name = tensor<string, []>("op_52261_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_52261_end_0 = const()[name = tensor<string, []>("op_52261_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_52261_end_mask_0 = const()[name = tensor<string, []>("op_52261_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_52261_cast_fp16 = slice_by_index(begin = var_52261_begin_0, end = var_52261_end_0, end_mask = var_52261_end_mask_0, x = v_231_cast_fp16)[name = tensor<string, []>("op_52261_cast_fp16")];
            tensor<int32, [4]> var_52265_begin_0 = const()[name = tensor<string, []>("op_52265_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_52265_end_0 = const()[name = tensor<string, []>("op_52265_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_52265_end_mask_0 = const()[name = tensor<string, []>("op_52265_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_52265_cast_fp16 = slice_by_index(begin = var_52265_begin_0, end = var_52265_end_0, end_mask = var_52265_end_mask_0, x = v_231_cast_fp16)[name = tensor<string, []>("op_52265_cast_fp16")];
            tensor<int32, [4]> var_52269_begin_0 = const()[name = tensor<string, []>("op_52269_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_52269_end_0 = const()[name = tensor<string, []>("op_52269_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_52269_end_mask_0 = const()[name = tensor<string, []>("op_52269_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_52269_cast_fp16 = slice_by_index(begin = var_52269_begin_0, end = var_52269_end_0, end_mask = var_52269_end_mask_0, x = v_231_cast_fp16)[name = tensor<string, []>("op_52269_cast_fp16")];
            tensor<int32, [4]> var_52273_begin_0 = const()[name = tensor<string, []>("op_52273_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_52273_end_0 = const()[name = tensor<string, []>("op_52273_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_52273_end_mask_0 = const()[name = tensor<string, []>("op_52273_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_52273_cast_fp16 = slice_by_index(begin = var_52273_begin_0, end = var_52273_end_0, end_mask = var_52273_end_mask_0, x = v_231_cast_fp16)[name = tensor<string, []>("op_52273_cast_fp16")];
            tensor<int32, [4]> var_52277_begin_0 = const()[name = tensor<string, []>("op_52277_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_52277_end_0 = const()[name = tensor<string, []>("op_52277_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_52277_end_mask_0 = const()[name = tensor<string, []>("op_52277_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_52277_cast_fp16 = slice_by_index(begin = var_52277_begin_0, end = var_52277_end_0, end_mask = var_52277_end_mask_0, x = v_231_cast_fp16)[name = tensor<string, []>("op_52277_cast_fp16")];
            tensor<int32, [4]> var_52281_begin_0 = const()[name = tensor<string, []>("op_52281_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_52281_end_0 = const()[name = tensor<string, []>("op_52281_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_52281_end_mask_0 = const()[name = tensor<string, []>("op_52281_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_52281_cast_fp16 = slice_by_index(begin = var_52281_begin_0, end = var_52281_end_0, end_mask = var_52281_end_mask_0, x = v_231_cast_fp16)[name = tensor<string, []>("op_52281_cast_fp16")];
            tensor<int32, [4]> var_52285_begin_0 = const()[name = tensor<string, []>("op_52285_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_52285_end_0 = const()[name = tensor<string, []>("op_52285_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_52285_end_mask_0 = const()[name = tensor<string, []>("op_52285_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_52285_cast_fp16 = slice_by_index(begin = var_52285_begin_0, end = var_52285_end_0, end_mask = var_52285_end_mask_0, x = v_231_cast_fp16)[name = tensor<string, []>("op_52285_cast_fp16")];
            tensor<int32, [4]> var_52289_begin_0 = const()[name = tensor<string, []>("op_52289_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_52289_end_0 = const()[name = tensor<string, []>("op_52289_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_52289_end_mask_0 = const()[name = tensor<string, []>("op_52289_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_52289_cast_fp16 = slice_by_index(begin = var_52289_begin_0, end = var_52289_end_0, end_mask = var_52289_end_mask_0, x = v_231_cast_fp16)[name = tensor<string, []>("op_52289_cast_fp16")];
            tensor<int32, [4]> var_52293_begin_0 = const()[name = tensor<string, []>("op_52293_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_52293_end_0 = const()[name = tensor<string, []>("op_52293_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_52293_end_mask_0 = const()[name = tensor<string, []>("op_52293_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_52293_cast_fp16 = slice_by_index(begin = var_52293_begin_0, end = var_52293_end_0, end_mask = var_52293_end_mask_0, x = v_231_cast_fp16)[name = tensor<string, []>("op_52293_cast_fp16")];
            tensor<int32, [4]> var_52297_begin_0 = const()[name = tensor<string, []>("op_52297_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_52297_end_0 = const()[name = tensor<string, []>("op_52297_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_52297_end_mask_0 = const()[name = tensor<string, []>("op_52297_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_52297_cast_fp16 = slice_by_index(begin = var_52297_begin_0, end = var_52297_end_0, end_mask = var_52297_end_mask_0, x = v_231_cast_fp16)[name = tensor<string, []>("op_52297_cast_fp16")];
            tensor<int32, [4]> var_52301_begin_0 = const()[name = tensor<string, []>("op_52301_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_52301_end_0 = const()[name = tensor<string, []>("op_52301_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_52301_end_mask_0 = const()[name = tensor<string, []>("op_52301_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_52301_cast_fp16 = slice_by_index(begin = var_52301_begin_0, end = var_52301_end_0, end_mask = var_52301_end_mask_0, x = v_231_cast_fp16)[name = tensor<string, []>("op_52301_cast_fp16")];
            tensor<int32, [4]> var_52305_begin_0 = const()[name = tensor<string, []>("op_52305_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_52305_end_0 = const()[name = tensor<string, []>("op_52305_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_52305_end_mask_0 = const()[name = tensor<string, []>("op_52305_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_52305_cast_fp16 = slice_by_index(begin = var_52305_begin_0, end = var_52305_end_0, end_mask = var_52305_end_mask_0, x = v_231_cast_fp16)[name = tensor<string, []>("op_52305_cast_fp16")];
            tensor<int32, [4]> var_52309_begin_0 = const()[name = tensor<string, []>("op_52309_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_52309_end_0 = const()[name = tensor<string, []>("op_52309_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_52309_end_mask_0 = const()[name = tensor<string, []>("op_52309_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_52309_cast_fp16 = slice_by_index(begin = var_52309_begin_0, end = var_52309_end_0, end_mask = var_52309_end_mask_0, x = v_231_cast_fp16)[name = tensor<string, []>("op_52309_cast_fp16")];
            tensor<int32, [4]> var_52313_begin_0 = const()[name = tensor<string, []>("op_52313_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_52313_end_0 = const()[name = tensor<string, []>("op_52313_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_52313_end_mask_0 = const()[name = tensor<string, []>("op_52313_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_52313_cast_fp16 = slice_by_index(begin = var_52313_begin_0, end = var_52313_end_0, end_mask = var_52313_end_mask_0, x = v_231_cast_fp16)[name = tensor<string, []>("op_52313_cast_fp16")];
            tensor<int32, [4]> var_52317_begin_0 = const()[name = tensor<string, []>("op_52317_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_52317_end_0 = const()[name = tensor<string, []>("op_52317_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_52317_end_mask_0 = const()[name = tensor<string, []>("op_52317_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_52317_cast_fp16 = slice_by_index(begin = var_52317_begin_0, end = var_52317_end_0, end_mask = var_52317_end_mask_0, x = v_231_cast_fp16)[name = tensor<string, []>("op_52317_cast_fp16")];
            tensor<int32, [4]> var_52321_begin_0 = const()[name = tensor<string, []>("op_52321_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_52321_end_0 = const()[name = tensor<string, []>("op_52321_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_52321_end_mask_0 = const()[name = tensor<string, []>("op_52321_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_52321_cast_fp16 = slice_by_index(begin = var_52321_begin_0, end = var_52321_end_0, end_mask = var_52321_end_mask_0, x = v_231_cast_fp16)[name = tensor<string, []>("op_52321_cast_fp16")];
            tensor<int32, [4]> var_52325_begin_0 = const()[name = tensor<string, []>("op_52325_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_52325_end_0 = const()[name = tensor<string, []>("op_52325_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_52325_end_mask_0 = const()[name = tensor<string, []>("op_52325_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_52325_cast_fp16 = slice_by_index(begin = var_52325_begin_0, end = var_52325_end_0, end_mask = var_52325_end_mask_0, x = v_231_cast_fp16)[name = tensor<string, []>("op_52325_cast_fp16")];
            tensor<int32, [4]> var_52329_begin_0 = const()[name = tensor<string, []>("op_52329_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_52329_end_0 = const()[name = tensor<string, []>("op_52329_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_52329_end_mask_0 = const()[name = tensor<string, []>("op_52329_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_52329_cast_fp16 = slice_by_index(begin = var_52329_begin_0, end = var_52329_end_0, end_mask = var_52329_end_mask_0, x = v_231_cast_fp16)[name = tensor<string, []>("op_52329_cast_fp16")];
            tensor<string, []> var_52333_equation_0 = const()[name = tensor<string, []>("op_52333_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_52333_cast_fp16 = einsum(equation = var_52333_equation_0, values = (var_52175_cast_fp16, var_52092_cast_fp16))[name = tensor<string, []>("op_52333_cast_fp16")];
            tensor<fp16, []> var_52334_to_fp16 = const()[name = tensor<string, []>("op_52334_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4441_cast_fp16 = mul(x = var_52333_cast_fp16, y = var_52334_to_fp16)[name = tensor<string, []>("aw_4441_cast_fp16")];
            tensor<string, []> var_52337_equation_0 = const()[name = tensor<string, []>("op_52337_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_52337_cast_fp16 = einsum(equation = var_52337_equation_0, values = (var_52179_cast_fp16, var_52096_cast_fp16))[name = tensor<string, []>("op_52337_cast_fp16")];
            tensor<fp16, []> var_52338_to_fp16 = const()[name = tensor<string, []>("op_52338_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4443_cast_fp16 = mul(x = var_52337_cast_fp16, y = var_52338_to_fp16)[name = tensor<string, []>("aw_4443_cast_fp16")];
            tensor<string, []> var_52341_equation_0 = const()[name = tensor<string, []>("op_52341_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_52341_cast_fp16 = einsum(equation = var_52341_equation_0, values = (var_52183_cast_fp16, var_52100_cast_fp16))[name = tensor<string, []>("op_52341_cast_fp16")];
            tensor<fp16, []> var_52342_to_fp16 = const()[name = tensor<string, []>("op_52342_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4445_cast_fp16 = mul(x = var_52341_cast_fp16, y = var_52342_to_fp16)[name = tensor<string, []>("aw_4445_cast_fp16")];
            tensor<string, []> var_52345_equation_0 = const()[name = tensor<string, []>("op_52345_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_52345_cast_fp16 = einsum(equation = var_52345_equation_0, values = (var_52187_cast_fp16, var_52104_cast_fp16))[name = tensor<string, []>("op_52345_cast_fp16")];
            tensor<fp16, []> var_52346_to_fp16 = const()[name = tensor<string, []>("op_52346_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4447_cast_fp16 = mul(x = var_52345_cast_fp16, y = var_52346_to_fp16)[name = tensor<string, []>("aw_4447_cast_fp16")];
            tensor<string, []> var_52349_equation_0 = const()[name = tensor<string, []>("op_52349_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_52349_cast_fp16 = einsum(equation = var_52349_equation_0, values = (var_52191_cast_fp16, var_52108_cast_fp16))[name = tensor<string, []>("op_52349_cast_fp16")];
            tensor<fp16, []> var_52350_to_fp16 = const()[name = tensor<string, []>("op_52350_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4449_cast_fp16 = mul(x = var_52349_cast_fp16, y = var_52350_to_fp16)[name = tensor<string, []>("aw_4449_cast_fp16")];
            tensor<string, []> var_52353_equation_0 = const()[name = tensor<string, []>("op_52353_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_52353_cast_fp16 = einsum(equation = var_52353_equation_0, values = (var_52195_cast_fp16, var_52112_cast_fp16))[name = tensor<string, []>("op_52353_cast_fp16")];
            tensor<fp16, []> var_52354_to_fp16 = const()[name = tensor<string, []>("op_52354_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4451_cast_fp16 = mul(x = var_52353_cast_fp16, y = var_52354_to_fp16)[name = tensor<string, []>("aw_4451_cast_fp16")];
            tensor<string, []> var_52357_equation_0 = const()[name = tensor<string, []>("op_52357_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_52357_cast_fp16 = einsum(equation = var_52357_equation_0, values = (var_52199_cast_fp16, var_52116_cast_fp16))[name = tensor<string, []>("op_52357_cast_fp16")];
            tensor<fp16, []> var_52358_to_fp16 = const()[name = tensor<string, []>("op_52358_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4453_cast_fp16 = mul(x = var_52357_cast_fp16, y = var_52358_to_fp16)[name = tensor<string, []>("aw_4453_cast_fp16")];
            tensor<string, []> var_52361_equation_0 = const()[name = tensor<string, []>("op_52361_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_52361_cast_fp16 = einsum(equation = var_52361_equation_0, values = (var_52203_cast_fp16, var_52120_cast_fp16))[name = tensor<string, []>("op_52361_cast_fp16")];
            tensor<fp16, []> var_52362_to_fp16 = const()[name = tensor<string, []>("op_52362_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4455_cast_fp16 = mul(x = var_52361_cast_fp16, y = var_52362_to_fp16)[name = tensor<string, []>("aw_4455_cast_fp16")];
            tensor<string, []> var_52365_equation_0 = const()[name = tensor<string, []>("op_52365_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_52365_cast_fp16 = einsum(equation = var_52365_equation_0, values = (var_52207_cast_fp16, var_52124_cast_fp16))[name = tensor<string, []>("op_52365_cast_fp16")];
            tensor<fp16, []> var_52366_to_fp16 = const()[name = tensor<string, []>("op_52366_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4457_cast_fp16 = mul(x = var_52365_cast_fp16, y = var_52366_to_fp16)[name = tensor<string, []>("aw_4457_cast_fp16")];
            tensor<string, []> var_52369_equation_0 = const()[name = tensor<string, []>("op_52369_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_52369_cast_fp16 = einsum(equation = var_52369_equation_0, values = (var_52211_cast_fp16, var_52128_cast_fp16))[name = tensor<string, []>("op_52369_cast_fp16")];
            tensor<fp16, []> var_52370_to_fp16 = const()[name = tensor<string, []>("op_52370_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4459_cast_fp16 = mul(x = var_52369_cast_fp16, y = var_52370_to_fp16)[name = tensor<string, []>("aw_4459_cast_fp16")];
            tensor<string, []> var_52373_equation_0 = const()[name = tensor<string, []>("op_52373_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_52373_cast_fp16 = einsum(equation = var_52373_equation_0, values = (var_52215_cast_fp16, var_52132_cast_fp16))[name = tensor<string, []>("op_52373_cast_fp16")];
            tensor<fp16, []> var_52374_to_fp16 = const()[name = tensor<string, []>("op_52374_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4461_cast_fp16 = mul(x = var_52373_cast_fp16, y = var_52374_to_fp16)[name = tensor<string, []>("aw_4461_cast_fp16")];
            tensor<string, []> var_52377_equation_0 = const()[name = tensor<string, []>("op_52377_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_52377_cast_fp16 = einsum(equation = var_52377_equation_0, values = (var_52219_cast_fp16, var_52136_cast_fp16))[name = tensor<string, []>("op_52377_cast_fp16")];
            tensor<fp16, []> var_52378_to_fp16 = const()[name = tensor<string, []>("op_52378_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4463_cast_fp16 = mul(x = var_52377_cast_fp16, y = var_52378_to_fp16)[name = tensor<string, []>("aw_4463_cast_fp16")];
            tensor<string, []> var_52381_equation_0 = const()[name = tensor<string, []>("op_52381_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_52381_cast_fp16 = einsum(equation = var_52381_equation_0, values = (var_52223_cast_fp16, var_52140_cast_fp16))[name = tensor<string, []>("op_52381_cast_fp16")];
            tensor<fp16, []> var_52382_to_fp16 = const()[name = tensor<string, []>("op_52382_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4465_cast_fp16 = mul(x = var_52381_cast_fp16, y = var_52382_to_fp16)[name = tensor<string, []>("aw_4465_cast_fp16")];
            tensor<string, []> var_52385_equation_0 = const()[name = tensor<string, []>("op_52385_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_52385_cast_fp16 = einsum(equation = var_52385_equation_0, values = (var_52227_cast_fp16, var_52144_cast_fp16))[name = tensor<string, []>("op_52385_cast_fp16")];
            tensor<fp16, []> var_52386_to_fp16 = const()[name = tensor<string, []>("op_52386_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4467_cast_fp16 = mul(x = var_52385_cast_fp16, y = var_52386_to_fp16)[name = tensor<string, []>("aw_4467_cast_fp16")];
            tensor<string, []> var_52389_equation_0 = const()[name = tensor<string, []>("op_52389_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_52389_cast_fp16 = einsum(equation = var_52389_equation_0, values = (var_52231_cast_fp16, var_52148_cast_fp16))[name = tensor<string, []>("op_52389_cast_fp16")];
            tensor<fp16, []> var_52390_to_fp16 = const()[name = tensor<string, []>("op_52390_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4469_cast_fp16 = mul(x = var_52389_cast_fp16, y = var_52390_to_fp16)[name = tensor<string, []>("aw_4469_cast_fp16")];
            tensor<string, []> var_52393_equation_0 = const()[name = tensor<string, []>("op_52393_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_52393_cast_fp16 = einsum(equation = var_52393_equation_0, values = (var_52235_cast_fp16, var_52152_cast_fp16))[name = tensor<string, []>("op_52393_cast_fp16")];
            tensor<fp16, []> var_52394_to_fp16 = const()[name = tensor<string, []>("op_52394_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4471_cast_fp16 = mul(x = var_52393_cast_fp16, y = var_52394_to_fp16)[name = tensor<string, []>("aw_4471_cast_fp16")];
            tensor<string, []> var_52397_equation_0 = const()[name = tensor<string, []>("op_52397_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_52397_cast_fp16 = einsum(equation = var_52397_equation_0, values = (var_52239_cast_fp16, var_52156_cast_fp16))[name = tensor<string, []>("op_52397_cast_fp16")];
            tensor<fp16, []> var_52398_to_fp16 = const()[name = tensor<string, []>("op_52398_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4473_cast_fp16 = mul(x = var_52397_cast_fp16, y = var_52398_to_fp16)[name = tensor<string, []>("aw_4473_cast_fp16")];
            tensor<string, []> var_52401_equation_0 = const()[name = tensor<string, []>("op_52401_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_52401_cast_fp16 = einsum(equation = var_52401_equation_0, values = (var_52243_cast_fp16, var_52160_cast_fp16))[name = tensor<string, []>("op_52401_cast_fp16")];
            tensor<fp16, []> var_52402_to_fp16 = const()[name = tensor<string, []>("op_52402_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4475_cast_fp16 = mul(x = var_52401_cast_fp16, y = var_52402_to_fp16)[name = tensor<string, []>("aw_4475_cast_fp16")];
            tensor<string, []> var_52405_equation_0 = const()[name = tensor<string, []>("op_52405_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_52405_cast_fp16 = einsum(equation = var_52405_equation_0, values = (var_52247_cast_fp16, var_52164_cast_fp16))[name = tensor<string, []>("op_52405_cast_fp16")];
            tensor<fp16, []> var_52406_to_fp16 = const()[name = tensor<string, []>("op_52406_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4477_cast_fp16 = mul(x = var_52405_cast_fp16, y = var_52406_to_fp16)[name = tensor<string, []>("aw_4477_cast_fp16")];
            tensor<string, []> var_52409_equation_0 = const()[name = tensor<string, []>("op_52409_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_52409_cast_fp16 = einsum(equation = var_52409_equation_0, values = (var_52251_cast_fp16, var_52168_cast_fp16))[name = tensor<string, []>("op_52409_cast_fp16")];
            tensor<fp16, []> var_52410_to_fp16 = const()[name = tensor<string, []>("op_52410_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4479_cast_fp16 = mul(x = var_52409_cast_fp16, y = var_52410_to_fp16)[name = tensor<string, []>("aw_4479_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_52412_cast_fp16 = softmax(axis = var_30355, x = aw_4441_cast_fp16)[name = tensor<string, []>("op_52412_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_52413_cast_fp16 = softmax(axis = var_30355, x = aw_4443_cast_fp16)[name = tensor<string, []>("op_52413_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_52414_cast_fp16 = softmax(axis = var_30355, x = aw_4445_cast_fp16)[name = tensor<string, []>("op_52414_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_52415_cast_fp16 = softmax(axis = var_30355, x = aw_4447_cast_fp16)[name = tensor<string, []>("op_52415_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_52416_cast_fp16 = softmax(axis = var_30355, x = aw_4449_cast_fp16)[name = tensor<string, []>("op_52416_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_52417_cast_fp16 = softmax(axis = var_30355, x = aw_4451_cast_fp16)[name = tensor<string, []>("op_52417_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_52418_cast_fp16 = softmax(axis = var_30355, x = aw_4453_cast_fp16)[name = tensor<string, []>("op_52418_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_52419_cast_fp16 = softmax(axis = var_30355, x = aw_4455_cast_fp16)[name = tensor<string, []>("op_52419_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_52420_cast_fp16 = softmax(axis = var_30355, x = aw_4457_cast_fp16)[name = tensor<string, []>("op_52420_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_52421_cast_fp16 = softmax(axis = var_30355, x = aw_4459_cast_fp16)[name = tensor<string, []>("op_52421_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_52422_cast_fp16 = softmax(axis = var_30355, x = aw_4461_cast_fp16)[name = tensor<string, []>("op_52422_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_52423_cast_fp16 = softmax(axis = var_30355, x = aw_4463_cast_fp16)[name = tensor<string, []>("op_52423_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_52424_cast_fp16 = softmax(axis = var_30355, x = aw_4465_cast_fp16)[name = tensor<string, []>("op_52424_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_52425_cast_fp16 = softmax(axis = var_30355, x = aw_4467_cast_fp16)[name = tensor<string, []>("op_52425_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_52426_cast_fp16 = softmax(axis = var_30355, x = aw_4469_cast_fp16)[name = tensor<string, []>("op_52426_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_52427_cast_fp16 = softmax(axis = var_30355, x = aw_4471_cast_fp16)[name = tensor<string, []>("op_52427_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_52428_cast_fp16 = softmax(axis = var_30355, x = aw_4473_cast_fp16)[name = tensor<string, []>("op_52428_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_52429_cast_fp16 = softmax(axis = var_30355, x = aw_4475_cast_fp16)[name = tensor<string, []>("op_52429_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_52430_cast_fp16 = softmax(axis = var_30355, x = aw_4477_cast_fp16)[name = tensor<string, []>("op_52430_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_52431_cast_fp16 = softmax(axis = var_30355, x = aw_4479_cast_fp16)[name = tensor<string, []>("op_52431_cast_fp16")];
            tensor<string, []> var_52433_equation_0 = const()[name = tensor<string, []>("op_52433_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52433_cast_fp16 = einsum(equation = var_52433_equation_0, values = (var_52253_cast_fp16, var_52412_cast_fp16))[name = tensor<string, []>("op_52433_cast_fp16")];
            tensor<string, []> var_52435_equation_0 = const()[name = tensor<string, []>("op_52435_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52435_cast_fp16 = einsum(equation = var_52435_equation_0, values = (var_52257_cast_fp16, var_52413_cast_fp16))[name = tensor<string, []>("op_52435_cast_fp16")];
            tensor<string, []> var_52437_equation_0 = const()[name = tensor<string, []>("op_52437_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52437_cast_fp16 = einsum(equation = var_52437_equation_0, values = (var_52261_cast_fp16, var_52414_cast_fp16))[name = tensor<string, []>("op_52437_cast_fp16")];
            tensor<string, []> var_52439_equation_0 = const()[name = tensor<string, []>("op_52439_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52439_cast_fp16 = einsum(equation = var_52439_equation_0, values = (var_52265_cast_fp16, var_52415_cast_fp16))[name = tensor<string, []>("op_52439_cast_fp16")];
            tensor<string, []> var_52441_equation_0 = const()[name = tensor<string, []>("op_52441_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52441_cast_fp16 = einsum(equation = var_52441_equation_0, values = (var_52269_cast_fp16, var_52416_cast_fp16))[name = tensor<string, []>("op_52441_cast_fp16")];
            tensor<string, []> var_52443_equation_0 = const()[name = tensor<string, []>("op_52443_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52443_cast_fp16 = einsum(equation = var_52443_equation_0, values = (var_52273_cast_fp16, var_52417_cast_fp16))[name = tensor<string, []>("op_52443_cast_fp16")];
            tensor<string, []> var_52445_equation_0 = const()[name = tensor<string, []>("op_52445_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52445_cast_fp16 = einsum(equation = var_52445_equation_0, values = (var_52277_cast_fp16, var_52418_cast_fp16))[name = tensor<string, []>("op_52445_cast_fp16")];
            tensor<string, []> var_52447_equation_0 = const()[name = tensor<string, []>("op_52447_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52447_cast_fp16 = einsum(equation = var_52447_equation_0, values = (var_52281_cast_fp16, var_52419_cast_fp16))[name = tensor<string, []>("op_52447_cast_fp16")];
            tensor<string, []> var_52449_equation_0 = const()[name = tensor<string, []>("op_52449_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52449_cast_fp16 = einsum(equation = var_52449_equation_0, values = (var_52285_cast_fp16, var_52420_cast_fp16))[name = tensor<string, []>("op_52449_cast_fp16")];
            tensor<string, []> var_52451_equation_0 = const()[name = tensor<string, []>("op_52451_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52451_cast_fp16 = einsum(equation = var_52451_equation_0, values = (var_52289_cast_fp16, var_52421_cast_fp16))[name = tensor<string, []>("op_52451_cast_fp16")];
            tensor<string, []> var_52453_equation_0 = const()[name = tensor<string, []>("op_52453_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52453_cast_fp16 = einsum(equation = var_52453_equation_0, values = (var_52293_cast_fp16, var_52422_cast_fp16))[name = tensor<string, []>("op_52453_cast_fp16")];
            tensor<string, []> var_52455_equation_0 = const()[name = tensor<string, []>("op_52455_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52455_cast_fp16 = einsum(equation = var_52455_equation_0, values = (var_52297_cast_fp16, var_52423_cast_fp16))[name = tensor<string, []>("op_52455_cast_fp16")];
            tensor<string, []> var_52457_equation_0 = const()[name = tensor<string, []>("op_52457_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52457_cast_fp16 = einsum(equation = var_52457_equation_0, values = (var_52301_cast_fp16, var_52424_cast_fp16))[name = tensor<string, []>("op_52457_cast_fp16")];
            tensor<string, []> var_52459_equation_0 = const()[name = tensor<string, []>("op_52459_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52459_cast_fp16 = einsum(equation = var_52459_equation_0, values = (var_52305_cast_fp16, var_52425_cast_fp16))[name = tensor<string, []>("op_52459_cast_fp16")];
            tensor<string, []> var_52461_equation_0 = const()[name = tensor<string, []>("op_52461_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52461_cast_fp16 = einsum(equation = var_52461_equation_0, values = (var_52309_cast_fp16, var_52426_cast_fp16))[name = tensor<string, []>("op_52461_cast_fp16")];
            tensor<string, []> var_52463_equation_0 = const()[name = tensor<string, []>("op_52463_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52463_cast_fp16 = einsum(equation = var_52463_equation_0, values = (var_52313_cast_fp16, var_52427_cast_fp16))[name = tensor<string, []>("op_52463_cast_fp16")];
            tensor<string, []> var_52465_equation_0 = const()[name = tensor<string, []>("op_52465_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52465_cast_fp16 = einsum(equation = var_52465_equation_0, values = (var_52317_cast_fp16, var_52428_cast_fp16))[name = tensor<string, []>("op_52465_cast_fp16")];
            tensor<string, []> var_52467_equation_0 = const()[name = tensor<string, []>("op_52467_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52467_cast_fp16 = einsum(equation = var_52467_equation_0, values = (var_52321_cast_fp16, var_52429_cast_fp16))[name = tensor<string, []>("op_52467_cast_fp16")];
            tensor<string, []> var_52469_equation_0 = const()[name = tensor<string, []>("op_52469_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52469_cast_fp16 = einsum(equation = var_52469_equation_0, values = (var_52325_cast_fp16, var_52430_cast_fp16))[name = tensor<string, []>("op_52469_cast_fp16")];
            tensor<string, []> var_52471_equation_0 = const()[name = tensor<string, []>("op_52471_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52471_cast_fp16 = einsum(equation = var_52471_equation_0, values = (var_52329_cast_fp16, var_52431_cast_fp16))[name = tensor<string, []>("op_52471_cast_fp16")];
            tensor<bool, []> input_661_interleave_0 = const()[name = tensor<string, []>("input_661_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_661_cast_fp16 = concat(axis = var_30355, interleave = input_661_interleave_0, values = (var_52433_cast_fp16, var_52435_cast_fp16, var_52437_cast_fp16, var_52439_cast_fp16, var_52441_cast_fp16, var_52443_cast_fp16, var_52445_cast_fp16, var_52447_cast_fp16, var_52449_cast_fp16, var_52451_cast_fp16, var_52453_cast_fp16, var_52455_cast_fp16, var_52457_cast_fp16, var_52459_cast_fp16, var_52461_cast_fp16, var_52463_cast_fp16, var_52465_cast_fp16, var_52467_cast_fp16, var_52469_cast_fp16, var_52471_cast_fp16))[name = tensor<string, []>("input_661_cast_fp16")];
            tensor<string, []> var_52481_pad_type_0 = const()[name = tensor<string, []>("op_52481_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_52481_strides_0 = const()[name = tensor<string, []>("op_52481_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_52481_pad_0 = const()[name = tensor<string, []>("op_52481_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_52481_dilations_0 = const()[name = tensor<string, []>("op_52481_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_52481_groups_0 = const()[name = tensor<string, []>("op_52481_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_3_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(686626432))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(687855296))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_3_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_3_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_3_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(687855488)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_52481_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_3_attn2_to_out_0_bias_to_fp16, dilations = var_52481_dilations_0, groups = var_52481_groups_0, pad = var_52481_pad_0, pad_type = var_52481_pad_type_0, strides = var_52481_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_3_attn2_to_out_0_weight_to_fp16_palettized, x = input_661_cast_fp16)[name = tensor<string, []>("op_52481_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_347_cast_fp16 = add(x = var_52481_cast_fp16, y = inputs_345_cast_fp16)[name = tensor<string, []>("inputs_347_cast_fp16")];
            tensor<int32, [1]> input_663_axes_0 = const()[name = tensor<string, []>("input_663_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> input_663_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_663_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(687858112)))];
            tensor<fp16, [1280]> input_663_beta_0_to_fp16 = const()[name = tensor<string, []>("input_663_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(687860736)))];
            tensor<fp16, []> var_52491_to_fp16 = const()[name = tensor<string, []>("op_52491_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> input_663_cast_fp16 = layer_norm(axes = input_663_axes_0, beta = input_663_beta_0_to_fp16, epsilon = var_52491_to_fp16, gamma = input_663_gamma_0_to_fp16, x = inputs_347_cast_fp16)[name = tensor<string, []>("input_663_cast_fp16")];
            tensor<string, []> var_52511_pad_type_0 = const()[name = tensor<string, []>("op_52511_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_52511_strides_0 = const()[name = tensor<string, []>("op_52511_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_52511_pad_0 = const()[name = tensor<string, []>("op_52511_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_52511_dilations_0 = const()[name = tensor<string, []>("op_52511_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_52511_groups_0 = const()[name = tensor<string, []>("op_52511_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_3_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [9830400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(687863360))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(697693824))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_3_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([10240, 1280, 1, 1])];
            tensor<fp16, [10240]> up_blocks_0_attentions_2_transformer_blocks_3_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_3_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(697694016)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_52511_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_3_ff_net_0_proj_bias_to_fp16, dilations = var_52511_dilations_0, groups = var_52511_groups_0, pad = var_52511_pad_0, pad_type = var_52511_pad_type_0, strides = var_52511_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_3_ff_net_0_proj_weight_to_fp16_palettized, x = input_663_cast_fp16)[name = tensor<string, []>("op_52511_cast_fp16")];
            tensor<int32, [2]> var_52512_split_sizes_0 = const()[name = tensor<string, []>("op_52512_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_52512_axis_0 = const()[name = tensor<string, []>("op_52512_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_52512_cast_fp16_0, tensor<fp16, [2, 5120, 1, 1024]> var_52512_cast_fp16_1 = split(axis = var_52512_axis_0, split_sizes = var_52512_split_sizes_0, x = var_52511_cast_fp16)[name = tensor<string, []>("op_52512_cast_fp16")];
            tensor<string, []> var_52514_mode_0 = const()[name = tensor<string, []>("op_52514_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_52514_cast_fp16 = gelu(mode = var_52514_mode_0, x = var_52512_cast_fp16_1)[name = tensor<string, []>("op_52514_cast_fp16")];
            tensor<fp16, [2, 5120, 1, 1024]> input_665_cast_fp16 = mul(x = var_52512_cast_fp16_0, y = var_52514_cast_fp16)[name = tensor<string, []>("input_665_cast_fp16")];
            tensor<string, []> var_52522_pad_type_0 = const()[name = tensor<string, []>("op_52522_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_52522_strides_0 = const()[name = tensor<string, []>("op_52522_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_52522_pad_0 = const()[name = tensor<string, []>("op_52522_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_52522_dilations_0 = const()[name = tensor<string, []>("op_52522_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_52522_groups_0 = const()[name = tensor<string, []>("op_52522_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_3_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [4915200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(697714560))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(702629824))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_3_ff_net_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 5120, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_3_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_3_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(702630016)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_52522_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_3_ff_net_2_bias_to_fp16, dilations = var_52522_dilations_0, groups = var_52522_groups_0, pad = var_52522_pad_0, pad_type = var_52522_pad_type_0, strides = var_52522_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_3_ff_net_2_weight_to_fp16_palettized, x = input_665_cast_fp16)[name = tensor<string, []>("op_52522_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_349_cast_fp16 = add(x = var_52522_cast_fp16, y = inputs_347_cast_fp16)[name = tensor<string, []>("inputs_349_cast_fp16")];
            tensor<int32, [1]> hidden_states_459_axes_0 = const()[name = tensor<string, []>("hidden_states_459_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_459_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_459_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(702632640)))];
            tensor<fp16, [1280]> hidden_states_459_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_459_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(702635264)))];
            tensor<fp16, []> var_52538_to_fp16 = const()[name = tensor<string, []>("op_52538_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_459_cast_fp16 = layer_norm(axes = hidden_states_459_axes_0, beta = hidden_states_459_beta_0_to_fp16, epsilon = var_52538_to_fp16, gamma = hidden_states_459_gamma_0_to_fp16, x = inputs_349_cast_fp16)[name = tensor<string, []>("hidden_states_459_cast_fp16")];
            tensor<string, []> q_233_pad_type_0 = const()[name = tensor<string, []>("q_233_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_233_strides_0 = const()[name = tensor<string, []>("q_233_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_233_pad_0 = const()[name = tensor<string, []>("q_233_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_233_dilations_0 = const()[name = tensor<string, []>("q_233_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_233_groups_0 = const()[name = tensor<string, []>("q_233_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_4_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(702637888))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(703866752))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_4_attn1_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_233_cast_fp16 = conv(dilations = q_233_dilations_0, groups = q_233_groups_0, pad = q_233_pad_0, pad_type = q_233_pad_type_0, strides = q_233_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_4_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_459_cast_fp16)[name = tensor<string, []>("q_233_cast_fp16")];
            tensor<string, []> k_465_pad_type_0 = const()[name = tensor<string, []>("k_465_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_465_strides_0 = const()[name = tensor<string, []>("k_465_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_465_pad_0 = const()[name = tensor<string, []>("k_465_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_465_dilations_0 = const()[name = tensor<string, []>("k_465_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_465_groups_0 = const()[name = tensor<string, []>("k_465_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_4_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(703866944))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(705095808))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_4_attn1_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> k_465_cast_fp16 = conv(dilations = k_465_dilations_0, groups = k_465_groups_0, pad = k_465_pad_0, pad_type = k_465_pad_type_0, strides = k_465_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_4_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_459_cast_fp16)[name = tensor<string, []>("k_465_cast_fp16")];
            tensor<string, []> v_233_pad_type_0 = const()[name = tensor<string, []>("v_233_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_233_strides_0 = const()[name = tensor<string, []>("v_233_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_233_pad_0 = const()[name = tensor<string, []>("v_233_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_233_dilations_0 = const()[name = tensor<string, []>("v_233_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_233_groups_0 = const()[name = tensor<string, []>("v_233_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_4_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(705096000))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(706324864))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_4_attn1_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> v_233_cast_fp16 = conv(dilations = v_233_dilations_0, groups = v_233_groups_0, pad = v_233_pad_0, pad_type = v_233_pad_type_0, strides = v_233_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_4_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_459_cast_fp16)[name = tensor<string, []>("v_233_cast_fp16")];
            tensor<int32, [4]> var_52571_begin_0 = const()[name = tensor<string, []>("op_52571_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_52571_end_0 = const()[name = tensor<string, []>("op_52571_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_52571_end_mask_0 = const()[name = tensor<string, []>("op_52571_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52571_cast_fp16 = slice_by_index(begin = var_52571_begin_0, end = var_52571_end_0, end_mask = var_52571_end_mask_0, x = q_233_cast_fp16)[name = tensor<string, []>("op_52571_cast_fp16")];
            tensor<int32, [4]> var_52575_begin_0 = const()[name = tensor<string, []>("op_52575_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_52575_end_0 = const()[name = tensor<string, []>("op_52575_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_52575_end_mask_0 = const()[name = tensor<string, []>("op_52575_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52575_cast_fp16 = slice_by_index(begin = var_52575_begin_0, end = var_52575_end_0, end_mask = var_52575_end_mask_0, x = q_233_cast_fp16)[name = tensor<string, []>("op_52575_cast_fp16")];
            tensor<int32, [4]> var_52579_begin_0 = const()[name = tensor<string, []>("op_52579_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_52579_end_0 = const()[name = tensor<string, []>("op_52579_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_52579_end_mask_0 = const()[name = tensor<string, []>("op_52579_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52579_cast_fp16 = slice_by_index(begin = var_52579_begin_0, end = var_52579_end_0, end_mask = var_52579_end_mask_0, x = q_233_cast_fp16)[name = tensor<string, []>("op_52579_cast_fp16")];
            tensor<int32, [4]> var_52583_begin_0 = const()[name = tensor<string, []>("op_52583_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_52583_end_0 = const()[name = tensor<string, []>("op_52583_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_52583_end_mask_0 = const()[name = tensor<string, []>("op_52583_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52583_cast_fp16 = slice_by_index(begin = var_52583_begin_0, end = var_52583_end_0, end_mask = var_52583_end_mask_0, x = q_233_cast_fp16)[name = tensor<string, []>("op_52583_cast_fp16")];
            tensor<int32, [4]> var_52587_begin_0 = const()[name = tensor<string, []>("op_52587_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_52587_end_0 = const()[name = tensor<string, []>("op_52587_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_52587_end_mask_0 = const()[name = tensor<string, []>("op_52587_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52587_cast_fp16 = slice_by_index(begin = var_52587_begin_0, end = var_52587_end_0, end_mask = var_52587_end_mask_0, x = q_233_cast_fp16)[name = tensor<string, []>("op_52587_cast_fp16")];
            tensor<int32, [4]> var_52591_begin_0 = const()[name = tensor<string, []>("op_52591_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_52591_end_0 = const()[name = tensor<string, []>("op_52591_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_52591_end_mask_0 = const()[name = tensor<string, []>("op_52591_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52591_cast_fp16 = slice_by_index(begin = var_52591_begin_0, end = var_52591_end_0, end_mask = var_52591_end_mask_0, x = q_233_cast_fp16)[name = tensor<string, []>("op_52591_cast_fp16")];
            tensor<int32, [4]> var_52595_begin_0 = const()[name = tensor<string, []>("op_52595_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_52595_end_0 = const()[name = tensor<string, []>("op_52595_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_52595_end_mask_0 = const()[name = tensor<string, []>("op_52595_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52595_cast_fp16 = slice_by_index(begin = var_52595_begin_0, end = var_52595_end_0, end_mask = var_52595_end_mask_0, x = q_233_cast_fp16)[name = tensor<string, []>("op_52595_cast_fp16")];
            tensor<int32, [4]> var_52599_begin_0 = const()[name = tensor<string, []>("op_52599_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_52599_end_0 = const()[name = tensor<string, []>("op_52599_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_52599_end_mask_0 = const()[name = tensor<string, []>("op_52599_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52599_cast_fp16 = slice_by_index(begin = var_52599_begin_0, end = var_52599_end_0, end_mask = var_52599_end_mask_0, x = q_233_cast_fp16)[name = tensor<string, []>("op_52599_cast_fp16")];
            tensor<int32, [4]> var_52603_begin_0 = const()[name = tensor<string, []>("op_52603_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_52603_end_0 = const()[name = tensor<string, []>("op_52603_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_52603_end_mask_0 = const()[name = tensor<string, []>("op_52603_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52603_cast_fp16 = slice_by_index(begin = var_52603_begin_0, end = var_52603_end_0, end_mask = var_52603_end_mask_0, x = q_233_cast_fp16)[name = tensor<string, []>("op_52603_cast_fp16")];
            tensor<int32, [4]> var_52607_begin_0 = const()[name = tensor<string, []>("op_52607_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_52607_end_0 = const()[name = tensor<string, []>("op_52607_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_52607_end_mask_0 = const()[name = tensor<string, []>("op_52607_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52607_cast_fp16 = slice_by_index(begin = var_52607_begin_0, end = var_52607_end_0, end_mask = var_52607_end_mask_0, x = q_233_cast_fp16)[name = tensor<string, []>("op_52607_cast_fp16")];
            tensor<int32, [4]> var_52611_begin_0 = const()[name = tensor<string, []>("op_52611_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_52611_end_0 = const()[name = tensor<string, []>("op_52611_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_52611_end_mask_0 = const()[name = tensor<string, []>("op_52611_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52611_cast_fp16 = slice_by_index(begin = var_52611_begin_0, end = var_52611_end_0, end_mask = var_52611_end_mask_0, x = q_233_cast_fp16)[name = tensor<string, []>("op_52611_cast_fp16")];
            tensor<int32, [4]> var_52615_begin_0 = const()[name = tensor<string, []>("op_52615_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_52615_end_0 = const()[name = tensor<string, []>("op_52615_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_52615_end_mask_0 = const()[name = tensor<string, []>("op_52615_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52615_cast_fp16 = slice_by_index(begin = var_52615_begin_0, end = var_52615_end_0, end_mask = var_52615_end_mask_0, x = q_233_cast_fp16)[name = tensor<string, []>("op_52615_cast_fp16")];
            tensor<int32, [4]> var_52619_begin_0 = const()[name = tensor<string, []>("op_52619_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_52619_end_0 = const()[name = tensor<string, []>("op_52619_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_52619_end_mask_0 = const()[name = tensor<string, []>("op_52619_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52619_cast_fp16 = slice_by_index(begin = var_52619_begin_0, end = var_52619_end_0, end_mask = var_52619_end_mask_0, x = q_233_cast_fp16)[name = tensor<string, []>("op_52619_cast_fp16")];
            tensor<int32, [4]> var_52623_begin_0 = const()[name = tensor<string, []>("op_52623_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_52623_end_0 = const()[name = tensor<string, []>("op_52623_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_52623_end_mask_0 = const()[name = tensor<string, []>("op_52623_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52623_cast_fp16 = slice_by_index(begin = var_52623_begin_0, end = var_52623_end_0, end_mask = var_52623_end_mask_0, x = q_233_cast_fp16)[name = tensor<string, []>("op_52623_cast_fp16")];
            tensor<int32, [4]> var_52627_begin_0 = const()[name = tensor<string, []>("op_52627_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_52627_end_0 = const()[name = tensor<string, []>("op_52627_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_52627_end_mask_0 = const()[name = tensor<string, []>("op_52627_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52627_cast_fp16 = slice_by_index(begin = var_52627_begin_0, end = var_52627_end_0, end_mask = var_52627_end_mask_0, x = q_233_cast_fp16)[name = tensor<string, []>("op_52627_cast_fp16")];
            tensor<int32, [4]> var_52631_begin_0 = const()[name = tensor<string, []>("op_52631_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_52631_end_0 = const()[name = tensor<string, []>("op_52631_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_52631_end_mask_0 = const()[name = tensor<string, []>("op_52631_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52631_cast_fp16 = slice_by_index(begin = var_52631_begin_0, end = var_52631_end_0, end_mask = var_52631_end_mask_0, x = q_233_cast_fp16)[name = tensor<string, []>("op_52631_cast_fp16")];
            tensor<int32, [4]> var_52635_begin_0 = const()[name = tensor<string, []>("op_52635_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_52635_end_0 = const()[name = tensor<string, []>("op_52635_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_52635_end_mask_0 = const()[name = tensor<string, []>("op_52635_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52635_cast_fp16 = slice_by_index(begin = var_52635_begin_0, end = var_52635_end_0, end_mask = var_52635_end_mask_0, x = q_233_cast_fp16)[name = tensor<string, []>("op_52635_cast_fp16")];
            tensor<int32, [4]> var_52639_begin_0 = const()[name = tensor<string, []>("op_52639_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_52639_end_0 = const()[name = tensor<string, []>("op_52639_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_52639_end_mask_0 = const()[name = tensor<string, []>("op_52639_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52639_cast_fp16 = slice_by_index(begin = var_52639_begin_0, end = var_52639_end_0, end_mask = var_52639_end_mask_0, x = q_233_cast_fp16)[name = tensor<string, []>("op_52639_cast_fp16")];
            tensor<int32, [4]> var_52643_begin_0 = const()[name = tensor<string, []>("op_52643_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_52643_end_0 = const()[name = tensor<string, []>("op_52643_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_52643_end_mask_0 = const()[name = tensor<string, []>("op_52643_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52643_cast_fp16 = slice_by_index(begin = var_52643_begin_0, end = var_52643_end_0, end_mask = var_52643_end_mask_0, x = q_233_cast_fp16)[name = tensor<string, []>("op_52643_cast_fp16")];
            tensor<int32, [4]> var_52647_begin_0 = const()[name = tensor<string, []>("op_52647_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_52647_end_0 = const()[name = tensor<string, []>("op_52647_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_52647_end_mask_0 = const()[name = tensor<string, []>("op_52647_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52647_cast_fp16 = slice_by_index(begin = var_52647_begin_0, end = var_52647_end_0, end_mask = var_52647_end_mask_0, x = q_233_cast_fp16)[name = tensor<string, []>("op_52647_cast_fp16")];
            tensor<int32, [4]> k_467_perm_0 = const()[name = tensor<string, []>("k_467_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_52654_begin_0 = const()[name = tensor<string, []>("op_52654_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_52654_end_0 = const()[name = tensor<string, []>("op_52654_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_52654_end_mask_0 = const()[name = tensor<string, []>("op_52654_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> k_467_cast_fp16 = transpose(perm = k_467_perm_0, x = k_465_cast_fp16)[name = tensor<string, []>("transpose_23")];
            tensor<fp16, [2, 1024, 1, 64]> var_52654_cast_fp16 = slice_by_index(begin = var_52654_begin_0, end = var_52654_end_0, end_mask = var_52654_end_mask_0, x = k_467_cast_fp16)[name = tensor<string, []>("op_52654_cast_fp16")];
            tensor<int32, [4]> var_52658_begin_0 = const()[name = tensor<string, []>("op_52658_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_52658_end_0 = const()[name = tensor<string, []>("op_52658_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_52658_end_mask_0 = const()[name = tensor<string, []>("op_52658_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_52658_cast_fp16 = slice_by_index(begin = var_52658_begin_0, end = var_52658_end_0, end_mask = var_52658_end_mask_0, x = k_467_cast_fp16)[name = tensor<string, []>("op_52658_cast_fp16")];
            tensor<int32, [4]> var_52662_begin_0 = const()[name = tensor<string, []>("op_52662_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_52662_end_0 = const()[name = tensor<string, []>("op_52662_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_52662_end_mask_0 = const()[name = tensor<string, []>("op_52662_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_52662_cast_fp16 = slice_by_index(begin = var_52662_begin_0, end = var_52662_end_0, end_mask = var_52662_end_mask_0, x = k_467_cast_fp16)[name = tensor<string, []>("op_52662_cast_fp16")];
            tensor<int32, [4]> var_52666_begin_0 = const()[name = tensor<string, []>("op_52666_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_52666_end_0 = const()[name = tensor<string, []>("op_52666_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_52666_end_mask_0 = const()[name = tensor<string, []>("op_52666_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_52666_cast_fp16 = slice_by_index(begin = var_52666_begin_0, end = var_52666_end_0, end_mask = var_52666_end_mask_0, x = k_467_cast_fp16)[name = tensor<string, []>("op_52666_cast_fp16")];
            tensor<int32, [4]> var_52670_begin_0 = const()[name = tensor<string, []>("op_52670_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_52670_end_0 = const()[name = tensor<string, []>("op_52670_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_52670_end_mask_0 = const()[name = tensor<string, []>("op_52670_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_52670_cast_fp16 = slice_by_index(begin = var_52670_begin_0, end = var_52670_end_0, end_mask = var_52670_end_mask_0, x = k_467_cast_fp16)[name = tensor<string, []>("op_52670_cast_fp16")];
            tensor<int32, [4]> var_52674_begin_0 = const()[name = tensor<string, []>("op_52674_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_52674_end_0 = const()[name = tensor<string, []>("op_52674_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_52674_end_mask_0 = const()[name = tensor<string, []>("op_52674_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_52674_cast_fp16 = slice_by_index(begin = var_52674_begin_0, end = var_52674_end_0, end_mask = var_52674_end_mask_0, x = k_467_cast_fp16)[name = tensor<string, []>("op_52674_cast_fp16")];
            tensor<int32, [4]> var_52678_begin_0 = const()[name = tensor<string, []>("op_52678_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_52678_end_0 = const()[name = tensor<string, []>("op_52678_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_52678_end_mask_0 = const()[name = tensor<string, []>("op_52678_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_52678_cast_fp16 = slice_by_index(begin = var_52678_begin_0, end = var_52678_end_0, end_mask = var_52678_end_mask_0, x = k_467_cast_fp16)[name = tensor<string, []>("op_52678_cast_fp16")];
            tensor<int32, [4]> var_52682_begin_0 = const()[name = tensor<string, []>("op_52682_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_52682_end_0 = const()[name = tensor<string, []>("op_52682_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_52682_end_mask_0 = const()[name = tensor<string, []>("op_52682_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_52682_cast_fp16 = slice_by_index(begin = var_52682_begin_0, end = var_52682_end_0, end_mask = var_52682_end_mask_0, x = k_467_cast_fp16)[name = tensor<string, []>("op_52682_cast_fp16")];
            tensor<int32, [4]> var_52686_begin_0 = const()[name = tensor<string, []>("op_52686_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_52686_end_0 = const()[name = tensor<string, []>("op_52686_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_52686_end_mask_0 = const()[name = tensor<string, []>("op_52686_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_52686_cast_fp16 = slice_by_index(begin = var_52686_begin_0, end = var_52686_end_0, end_mask = var_52686_end_mask_0, x = k_467_cast_fp16)[name = tensor<string, []>("op_52686_cast_fp16")];
            tensor<int32, [4]> var_52690_begin_0 = const()[name = tensor<string, []>("op_52690_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_52690_end_0 = const()[name = tensor<string, []>("op_52690_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_52690_end_mask_0 = const()[name = tensor<string, []>("op_52690_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_52690_cast_fp16 = slice_by_index(begin = var_52690_begin_0, end = var_52690_end_0, end_mask = var_52690_end_mask_0, x = k_467_cast_fp16)[name = tensor<string, []>("op_52690_cast_fp16")];
            tensor<int32, [4]> var_52694_begin_0 = const()[name = tensor<string, []>("op_52694_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_52694_end_0 = const()[name = tensor<string, []>("op_52694_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_52694_end_mask_0 = const()[name = tensor<string, []>("op_52694_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_52694_cast_fp16 = slice_by_index(begin = var_52694_begin_0, end = var_52694_end_0, end_mask = var_52694_end_mask_0, x = k_467_cast_fp16)[name = tensor<string, []>("op_52694_cast_fp16")];
            tensor<int32, [4]> var_52698_begin_0 = const()[name = tensor<string, []>("op_52698_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_52698_end_0 = const()[name = tensor<string, []>("op_52698_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_52698_end_mask_0 = const()[name = tensor<string, []>("op_52698_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_52698_cast_fp16 = slice_by_index(begin = var_52698_begin_0, end = var_52698_end_0, end_mask = var_52698_end_mask_0, x = k_467_cast_fp16)[name = tensor<string, []>("op_52698_cast_fp16")];
            tensor<int32, [4]> var_52702_begin_0 = const()[name = tensor<string, []>("op_52702_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_52702_end_0 = const()[name = tensor<string, []>("op_52702_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_52702_end_mask_0 = const()[name = tensor<string, []>("op_52702_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_52702_cast_fp16 = slice_by_index(begin = var_52702_begin_0, end = var_52702_end_0, end_mask = var_52702_end_mask_0, x = k_467_cast_fp16)[name = tensor<string, []>("op_52702_cast_fp16")];
            tensor<int32, [4]> var_52706_begin_0 = const()[name = tensor<string, []>("op_52706_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_52706_end_0 = const()[name = tensor<string, []>("op_52706_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_52706_end_mask_0 = const()[name = tensor<string, []>("op_52706_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_52706_cast_fp16 = slice_by_index(begin = var_52706_begin_0, end = var_52706_end_0, end_mask = var_52706_end_mask_0, x = k_467_cast_fp16)[name = tensor<string, []>("op_52706_cast_fp16")];
            tensor<int32, [4]> var_52710_begin_0 = const()[name = tensor<string, []>("op_52710_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_52710_end_0 = const()[name = tensor<string, []>("op_52710_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_52710_end_mask_0 = const()[name = tensor<string, []>("op_52710_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_52710_cast_fp16 = slice_by_index(begin = var_52710_begin_0, end = var_52710_end_0, end_mask = var_52710_end_mask_0, x = k_467_cast_fp16)[name = tensor<string, []>("op_52710_cast_fp16")];
            tensor<int32, [4]> var_52714_begin_0 = const()[name = tensor<string, []>("op_52714_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_52714_end_0 = const()[name = tensor<string, []>("op_52714_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_52714_end_mask_0 = const()[name = tensor<string, []>("op_52714_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_52714_cast_fp16 = slice_by_index(begin = var_52714_begin_0, end = var_52714_end_0, end_mask = var_52714_end_mask_0, x = k_467_cast_fp16)[name = tensor<string, []>("op_52714_cast_fp16")];
            tensor<int32, [4]> var_52718_begin_0 = const()[name = tensor<string, []>("op_52718_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_52718_end_0 = const()[name = tensor<string, []>("op_52718_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_52718_end_mask_0 = const()[name = tensor<string, []>("op_52718_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_52718_cast_fp16 = slice_by_index(begin = var_52718_begin_0, end = var_52718_end_0, end_mask = var_52718_end_mask_0, x = k_467_cast_fp16)[name = tensor<string, []>("op_52718_cast_fp16")];
            tensor<int32, [4]> var_52722_begin_0 = const()[name = tensor<string, []>("op_52722_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_52722_end_0 = const()[name = tensor<string, []>("op_52722_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_52722_end_mask_0 = const()[name = tensor<string, []>("op_52722_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_52722_cast_fp16 = slice_by_index(begin = var_52722_begin_0, end = var_52722_end_0, end_mask = var_52722_end_mask_0, x = k_467_cast_fp16)[name = tensor<string, []>("op_52722_cast_fp16")];
            tensor<int32, [4]> var_52726_begin_0 = const()[name = tensor<string, []>("op_52726_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_52726_end_0 = const()[name = tensor<string, []>("op_52726_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_52726_end_mask_0 = const()[name = tensor<string, []>("op_52726_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_52726_cast_fp16 = slice_by_index(begin = var_52726_begin_0, end = var_52726_end_0, end_mask = var_52726_end_mask_0, x = k_467_cast_fp16)[name = tensor<string, []>("op_52726_cast_fp16")];
            tensor<int32, [4]> var_52730_begin_0 = const()[name = tensor<string, []>("op_52730_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_52730_end_0 = const()[name = tensor<string, []>("op_52730_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_52730_end_mask_0 = const()[name = tensor<string, []>("op_52730_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_52730_cast_fp16 = slice_by_index(begin = var_52730_begin_0, end = var_52730_end_0, end_mask = var_52730_end_mask_0, x = k_467_cast_fp16)[name = tensor<string, []>("op_52730_cast_fp16")];
            tensor<int32, [4]> var_52732_begin_0 = const()[name = tensor<string, []>("op_52732_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_52732_end_0 = const()[name = tensor<string, []>("op_52732_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_52732_end_mask_0 = const()[name = tensor<string, []>("op_52732_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52732_cast_fp16 = slice_by_index(begin = var_52732_begin_0, end = var_52732_end_0, end_mask = var_52732_end_mask_0, x = v_233_cast_fp16)[name = tensor<string, []>("op_52732_cast_fp16")];
            tensor<int32, [4]> var_52736_begin_0 = const()[name = tensor<string, []>("op_52736_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_52736_end_0 = const()[name = tensor<string, []>("op_52736_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_52736_end_mask_0 = const()[name = tensor<string, []>("op_52736_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52736_cast_fp16 = slice_by_index(begin = var_52736_begin_0, end = var_52736_end_0, end_mask = var_52736_end_mask_0, x = v_233_cast_fp16)[name = tensor<string, []>("op_52736_cast_fp16")];
            tensor<int32, [4]> var_52740_begin_0 = const()[name = tensor<string, []>("op_52740_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_52740_end_0 = const()[name = tensor<string, []>("op_52740_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_52740_end_mask_0 = const()[name = tensor<string, []>("op_52740_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52740_cast_fp16 = slice_by_index(begin = var_52740_begin_0, end = var_52740_end_0, end_mask = var_52740_end_mask_0, x = v_233_cast_fp16)[name = tensor<string, []>("op_52740_cast_fp16")];
            tensor<int32, [4]> var_52744_begin_0 = const()[name = tensor<string, []>("op_52744_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_52744_end_0 = const()[name = tensor<string, []>("op_52744_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_52744_end_mask_0 = const()[name = tensor<string, []>("op_52744_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52744_cast_fp16 = slice_by_index(begin = var_52744_begin_0, end = var_52744_end_0, end_mask = var_52744_end_mask_0, x = v_233_cast_fp16)[name = tensor<string, []>("op_52744_cast_fp16")];
            tensor<int32, [4]> var_52748_begin_0 = const()[name = tensor<string, []>("op_52748_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_52748_end_0 = const()[name = tensor<string, []>("op_52748_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_52748_end_mask_0 = const()[name = tensor<string, []>("op_52748_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52748_cast_fp16 = slice_by_index(begin = var_52748_begin_0, end = var_52748_end_0, end_mask = var_52748_end_mask_0, x = v_233_cast_fp16)[name = tensor<string, []>("op_52748_cast_fp16")];
            tensor<int32, [4]> var_52752_begin_0 = const()[name = tensor<string, []>("op_52752_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_52752_end_0 = const()[name = tensor<string, []>("op_52752_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_52752_end_mask_0 = const()[name = tensor<string, []>("op_52752_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52752_cast_fp16 = slice_by_index(begin = var_52752_begin_0, end = var_52752_end_0, end_mask = var_52752_end_mask_0, x = v_233_cast_fp16)[name = tensor<string, []>("op_52752_cast_fp16")];
            tensor<int32, [4]> var_52756_begin_0 = const()[name = tensor<string, []>("op_52756_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_52756_end_0 = const()[name = tensor<string, []>("op_52756_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_52756_end_mask_0 = const()[name = tensor<string, []>("op_52756_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52756_cast_fp16 = slice_by_index(begin = var_52756_begin_0, end = var_52756_end_0, end_mask = var_52756_end_mask_0, x = v_233_cast_fp16)[name = tensor<string, []>("op_52756_cast_fp16")];
            tensor<int32, [4]> var_52760_begin_0 = const()[name = tensor<string, []>("op_52760_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_52760_end_0 = const()[name = tensor<string, []>("op_52760_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_52760_end_mask_0 = const()[name = tensor<string, []>("op_52760_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52760_cast_fp16 = slice_by_index(begin = var_52760_begin_0, end = var_52760_end_0, end_mask = var_52760_end_mask_0, x = v_233_cast_fp16)[name = tensor<string, []>("op_52760_cast_fp16")];
            tensor<int32, [4]> var_52764_begin_0 = const()[name = tensor<string, []>("op_52764_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_52764_end_0 = const()[name = tensor<string, []>("op_52764_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_52764_end_mask_0 = const()[name = tensor<string, []>("op_52764_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52764_cast_fp16 = slice_by_index(begin = var_52764_begin_0, end = var_52764_end_0, end_mask = var_52764_end_mask_0, x = v_233_cast_fp16)[name = tensor<string, []>("op_52764_cast_fp16")];
            tensor<int32, [4]> var_52768_begin_0 = const()[name = tensor<string, []>("op_52768_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_52768_end_0 = const()[name = tensor<string, []>("op_52768_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_52768_end_mask_0 = const()[name = tensor<string, []>("op_52768_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52768_cast_fp16 = slice_by_index(begin = var_52768_begin_0, end = var_52768_end_0, end_mask = var_52768_end_mask_0, x = v_233_cast_fp16)[name = tensor<string, []>("op_52768_cast_fp16")];
            tensor<int32, [4]> var_52772_begin_0 = const()[name = tensor<string, []>("op_52772_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_52772_end_0 = const()[name = tensor<string, []>("op_52772_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_52772_end_mask_0 = const()[name = tensor<string, []>("op_52772_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52772_cast_fp16 = slice_by_index(begin = var_52772_begin_0, end = var_52772_end_0, end_mask = var_52772_end_mask_0, x = v_233_cast_fp16)[name = tensor<string, []>("op_52772_cast_fp16")];
            tensor<int32, [4]> var_52776_begin_0 = const()[name = tensor<string, []>("op_52776_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_52776_end_0 = const()[name = tensor<string, []>("op_52776_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_52776_end_mask_0 = const()[name = tensor<string, []>("op_52776_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52776_cast_fp16 = slice_by_index(begin = var_52776_begin_0, end = var_52776_end_0, end_mask = var_52776_end_mask_0, x = v_233_cast_fp16)[name = tensor<string, []>("op_52776_cast_fp16")];
            tensor<int32, [4]> var_52780_begin_0 = const()[name = tensor<string, []>("op_52780_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_52780_end_0 = const()[name = tensor<string, []>("op_52780_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_52780_end_mask_0 = const()[name = tensor<string, []>("op_52780_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52780_cast_fp16 = slice_by_index(begin = var_52780_begin_0, end = var_52780_end_0, end_mask = var_52780_end_mask_0, x = v_233_cast_fp16)[name = tensor<string, []>("op_52780_cast_fp16")];
            tensor<int32, [4]> var_52784_begin_0 = const()[name = tensor<string, []>("op_52784_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_52784_end_0 = const()[name = tensor<string, []>("op_52784_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_52784_end_mask_0 = const()[name = tensor<string, []>("op_52784_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52784_cast_fp16 = slice_by_index(begin = var_52784_begin_0, end = var_52784_end_0, end_mask = var_52784_end_mask_0, x = v_233_cast_fp16)[name = tensor<string, []>("op_52784_cast_fp16")];
            tensor<int32, [4]> var_52788_begin_0 = const()[name = tensor<string, []>("op_52788_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_52788_end_0 = const()[name = tensor<string, []>("op_52788_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_52788_end_mask_0 = const()[name = tensor<string, []>("op_52788_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52788_cast_fp16 = slice_by_index(begin = var_52788_begin_0, end = var_52788_end_0, end_mask = var_52788_end_mask_0, x = v_233_cast_fp16)[name = tensor<string, []>("op_52788_cast_fp16")];
            tensor<int32, [4]> var_52792_begin_0 = const()[name = tensor<string, []>("op_52792_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_52792_end_0 = const()[name = tensor<string, []>("op_52792_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_52792_end_mask_0 = const()[name = tensor<string, []>("op_52792_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52792_cast_fp16 = slice_by_index(begin = var_52792_begin_0, end = var_52792_end_0, end_mask = var_52792_end_mask_0, x = v_233_cast_fp16)[name = tensor<string, []>("op_52792_cast_fp16")];
            tensor<int32, [4]> var_52796_begin_0 = const()[name = tensor<string, []>("op_52796_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_52796_end_0 = const()[name = tensor<string, []>("op_52796_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_52796_end_mask_0 = const()[name = tensor<string, []>("op_52796_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52796_cast_fp16 = slice_by_index(begin = var_52796_begin_0, end = var_52796_end_0, end_mask = var_52796_end_mask_0, x = v_233_cast_fp16)[name = tensor<string, []>("op_52796_cast_fp16")];
            tensor<int32, [4]> var_52800_begin_0 = const()[name = tensor<string, []>("op_52800_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_52800_end_0 = const()[name = tensor<string, []>("op_52800_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_52800_end_mask_0 = const()[name = tensor<string, []>("op_52800_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52800_cast_fp16 = slice_by_index(begin = var_52800_begin_0, end = var_52800_end_0, end_mask = var_52800_end_mask_0, x = v_233_cast_fp16)[name = tensor<string, []>("op_52800_cast_fp16")];
            tensor<int32, [4]> var_52804_begin_0 = const()[name = tensor<string, []>("op_52804_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_52804_end_0 = const()[name = tensor<string, []>("op_52804_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_52804_end_mask_0 = const()[name = tensor<string, []>("op_52804_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52804_cast_fp16 = slice_by_index(begin = var_52804_begin_0, end = var_52804_end_0, end_mask = var_52804_end_mask_0, x = v_233_cast_fp16)[name = tensor<string, []>("op_52804_cast_fp16")];
            tensor<int32, [4]> var_52808_begin_0 = const()[name = tensor<string, []>("op_52808_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_52808_end_0 = const()[name = tensor<string, []>("op_52808_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_52808_end_mask_0 = const()[name = tensor<string, []>("op_52808_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52808_cast_fp16 = slice_by_index(begin = var_52808_begin_0, end = var_52808_end_0, end_mask = var_52808_end_mask_0, x = v_233_cast_fp16)[name = tensor<string, []>("op_52808_cast_fp16")];
            tensor<string, []> var_52812_equation_0 = const()[name = tensor<string, []>("op_52812_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52812_cast_fp16 = einsum(equation = var_52812_equation_0, values = (var_52654_cast_fp16, var_52571_cast_fp16))[name = tensor<string, []>("op_52812_cast_fp16")];
            tensor<fp16, []> var_52813_to_fp16 = const()[name = tensor<string, []>("op_52813_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4481_cast_fp16 = mul(x = var_52812_cast_fp16, y = var_52813_to_fp16)[name = tensor<string, []>("aw_4481_cast_fp16")];
            tensor<string, []> var_52816_equation_0 = const()[name = tensor<string, []>("op_52816_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52816_cast_fp16 = einsum(equation = var_52816_equation_0, values = (var_52658_cast_fp16, var_52575_cast_fp16))[name = tensor<string, []>("op_52816_cast_fp16")];
            tensor<fp16, []> var_52817_to_fp16 = const()[name = tensor<string, []>("op_52817_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4483_cast_fp16 = mul(x = var_52816_cast_fp16, y = var_52817_to_fp16)[name = tensor<string, []>("aw_4483_cast_fp16")];
            tensor<string, []> var_52820_equation_0 = const()[name = tensor<string, []>("op_52820_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52820_cast_fp16 = einsum(equation = var_52820_equation_0, values = (var_52662_cast_fp16, var_52579_cast_fp16))[name = tensor<string, []>("op_52820_cast_fp16")];
            tensor<fp16, []> var_52821_to_fp16 = const()[name = tensor<string, []>("op_52821_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4485_cast_fp16 = mul(x = var_52820_cast_fp16, y = var_52821_to_fp16)[name = tensor<string, []>("aw_4485_cast_fp16")];
            tensor<string, []> var_52824_equation_0 = const()[name = tensor<string, []>("op_52824_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52824_cast_fp16 = einsum(equation = var_52824_equation_0, values = (var_52666_cast_fp16, var_52583_cast_fp16))[name = tensor<string, []>("op_52824_cast_fp16")];
            tensor<fp16, []> var_52825_to_fp16 = const()[name = tensor<string, []>("op_52825_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4487_cast_fp16 = mul(x = var_52824_cast_fp16, y = var_52825_to_fp16)[name = tensor<string, []>("aw_4487_cast_fp16")];
            tensor<string, []> var_52828_equation_0 = const()[name = tensor<string, []>("op_52828_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52828_cast_fp16 = einsum(equation = var_52828_equation_0, values = (var_52670_cast_fp16, var_52587_cast_fp16))[name = tensor<string, []>("op_52828_cast_fp16")];
            tensor<fp16, []> var_52829_to_fp16 = const()[name = tensor<string, []>("op_52829_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4489_cast_fp16 = mul(x = var_52828_cast_fp16, y = var_52829_to_fp16)[name = tensor<string, []>("aw_4489_cast_fp16")];
            tensor<string, []> var_52832_equation_0 = const()[name = tensor<string, []>("op_52832_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52832_cast_fp16 = einsum(equation = var_52832_equation_0, values = (var_52674_cast_fp16, var_52591_cast_fp16))[name = tensor<string, []>("op_52832_cast_fp16")];
            tensor<fp16, []> var_52833_to_fp16 = const()[name = tensor<string, []>("op_52833_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4491_cast_fp16 = mul(x = var_52832_cast_fp16, y = var_52833_to_fp16)[name = tensor<string, []>("aw_4491_cast_fp16")];
            tensor<string, []> var_52836_equation_0 = const()[name = tensor<string, []>("op_52836_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52836_cast_fp16 = einsum(equation = var_52836_equation_0, values = (var_52678_cast_fp16, var_52595_cast_fp16))[name = tensor<string, []>("op_52836_cast_fp16")];
            tensor<fp16, []> var_52837_to_fp16 = const()[name = tensor<string, []>("op_52837_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4493_cast_fp16 = mul(x = var_52836_cast_fp16, y = var_52837_to_fp16)[name = tensor<string, []>("aw_4493_cast_fp16")];
            tensor<string, []> var_52840_equation_0 = const()[name = tensor<string, []>("op_52840_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52840_cast_fp16 = einsum(equation = var_52840_equation_0, values = (var_52682_cast_fp16, var_52599_cast_fp16))[name = tensor<string, []>("op_52840_cast_fp16")];
            tensor<fp16, []> var_52841_to_fp16 = const()[name = tensor<string, []>("op_52841_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4495_cast_fp16 = mul(x = var_52840_cast_fp16, y = var_52841_to_fp16)[name = tensor<string, []>("aw_4495_cast_fp16")];
            tensor<string, []> var_52844_equation_0 = const()[name = tensor<string, []>("op_52844_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52844_cast_fp16 = einsum(equation = var_52844_equation_0, values = (var_52686_cast_fp16, var_52603_cast_fp16))[name = tensor<string, []>("op_52844_cast_fp16")];
            tensor<fp16, []> var_52845_to_fp16 = const()[name = tensor<string, []>("op_52845_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4497_cast_fp16 = mul(x = var_52844_cast_fp16, y = var_52845_to_fp16)[name = tensor<string, []>("aw_4497_cast_fp16")];
            tensor<string, []> var_52848_equation_0 = const()[name = tensor<string, []>("op_52848_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52848_cast_fp16 = einsum(equation = var_52848_equation_0, values = (var_52690_cast_fp16, var_52607_cast_fp16))[name = tensor<string, []>("op_52848_cast_fp16")];
            tensor<fp16, []> var_52849_to_fp16 = const()[name = tensor<string, []>("op_52849_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4499_cast_fp16 = mul(x = var_52848_cast_fp16, y = var_52849_to_fp16)[name = tensor<string, []>("aw_4499_cast_fp16")];
            tensor<string, []> var_52852_equation_0 = const()[name = tensor<string, []>("op_52852_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52852_cast_fp16 = einsum(equation = var_52852_equation_0, values = (var_52694_cast_fp16, var_52611_cast_fp16))[name = tensor<string, []>("op_52852_cast_fp16")];
            tensor<fp16, []> var_52853_to_fp16 = const()[name = tensor<string, []>("op_52853_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4501_cast_fp16 = mul(x = var_52852_cast_fp16, y = var_52853_to_fp16)[name = tensor<string, []>("aw_4501_cast_fp16")];
            tensor<string, []> var_52856_equation_0 = const()[name = tensor<string, []>("op_52856_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52856_cast_fp16 = einsum(equation = var_52856_equation_0, values = (var_52698_cast_fp16, var_52615_cast_fp16))[name = tensor<string, []>("op_52856_cast_fp16")];
            tensor<fp16, []> var_52857_to_fp16 = const()[name = tensor<string, []>("op_52857_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4503_cast_fp16 = mul(x = var_52856_cast_fp16, y = var_52857_to_fp16)[name = tensor<string, []>("aw_4503_cast_fp16")];
            tensor<string, []> var_52860_equation_0 = const()[name = tensor<string, []>("op_52860_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52860_cast_fp16 = einsum(equation = var_52860_equation_0, values = (var_52702_cast_fp16, var_52619_cast_fp16))[name = tensor<string, []>("op_52860_cast_fp16")];
            tensor<fp16, []> var_52861_to_fp16 = const()[name = tensor<string, []>("op_52861_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4505_cast_fp16 = mul(x = var_52860_cast_fp16, y = var_52861_to_fp16)[name = tensor<string, []>("aw_4505_cast_fp16")];
            tensor<string, []> var_52864_equation_0 = const()[name = tensor<string, []>("op_52864_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52864_cast_fp16 = einsum(equation = var_52864_equation_0, values = (var_52706_cast_fp16, var_52623_cast_fp16))[name = tensor<string, []>("op_52864_cast_fp16")];
            tensor<fp16, []> var_52865_to_fp16 = const()[name = tensor<string, []>("op_52865_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4507_cast_fp16 = mul(x = var_52864_cast_fp16, y = var_52865_to_fp16)[name = tensor<string, []>("aw_4507_cast_fp16")];
            tensor<string, []> var_52868_equation_0 = const()[name = tensor<string, []>("op_52868_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52868_cast_fp16 = einsum(equation = var_52868_equation_0, values = (var_52710_cast_fp16, var_52627_cast_fp16))[name = tensor<string, []>("op_52868_cast_fp16")];
            tensor<fp16, []> var_52869_to_fp16 = const()[name = tensor<string, []>("op_52869_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4509_cast_fp16 = mul(x = var_52868_cast_fp16, y = var_52869_to_fp16)[name = tensor<string, []>("aw_4509_cast_fp16")];
            tensor<string, []> var_52872_equation_0 = const()[name = tensor<string, []>("op_52872_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52872_cast_fp16 = einsum(equation = var_52872_equation_0, values = (var_52714_cast_fp16, var_52631_cast_fp16))[name = tensor<string, []>("op_52872_cast_fp16")];
            tensor<fp16, []> var_52873_to_fp16 = const()[name = tensor<string, []>("op_52873_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4511_cast_fp16 = mul(x = var_52872_cast_fp16, y = var_52873_to_fp16)[name = tensor<string, []>("aw_4511_cast_fp16")];
            tensor<string, []> var_52876_equation_0 = const()[name = tensor<string, []>("op_52876_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52876_cast_fp16 = einsum(equation = var_52876_equation_0, values = (var_52718_cast_fp16, var_52635_cast_fp16))[name = tensor<string, []>("op_52876_cast_fp16")];
            tensor<fp16, []> var_52877_to_fp16 = const()[name = tensor<string, []>("op_52877_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4513_cast_fp16 = mul(x = var_52876_cast_fp16, y = var_52877_to_fp16)[name = tensor<string, []>("aw_4513_cast_fp16")];
            tensor<string, []> var_52880_equation_0 = const()[name = tensor<string, []>("op_52880_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52880_cast_fp16 = einsum(equation = var_52880_equation_0, values = (var_52722_cast_fp16, var_52639_cast_fp16))[name = tensor<string, []>("op_52880_cast_fp16")];
            tensor<fp16, []> var_52881_to_fp16 = const()[name = tensor<string, []>("op_52881_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4515_cast_fp16 = mul(x = var_52880_cast_fp16, y = var_52881_to_fp16)[name = tensor<string, []>("aw_4515_cast_fp16")];
            tensor<string, []> var_52884_equation_0 = const()[name = tensor<string, []>("op_52884_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52884_cast_fp16 = einsum(equation = var_52884_equation_0, values = (var_52726_cast_fp16, var_52643_cast_fp16))[name = tensor<string, []>("op_52884_cast_fp16")];
            tensor<fp16, []> var_52885_to_fp16 = const()[name = tensor<string, []>("op_52885_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4517_cast_fp16 = mul(x = var_52884_cast_fp16, y = var_52885_to_fp16)[name = tensor<string, []>("aw_4517_cast_fp16")];
            tensor<string, []> var_52888_equation_0 = const()[name = tensor<string, []>("op_52888_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52888_cast_fp16 = einsum(equation = var_52888_equation_0, values = (var_52730_cast_fp16, var_52647_cast_fp16))[name = tensor<string, []>("op_52888_cast_fp16")];
            tensor<fp16, []> var_52889_to_fp16 = const()[name = tensor<string, []>("op_52889_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4519_cast_fp16 = mul(x = var_52888_cast_fp16, y = var_52889_to_fp16)[name = tensor<string, []>("aw_4519_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52891_cast_fp16 = softmax(axis = var_30355, x = aw_4481_cast_fp16)[name = tensor<string, []>("op_52891_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52892_cast_fp16 = softmax(axis = var_30355, x = aw_4483_cast_fp16)[name = tensor<string, []>("op_52892_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52893_cast_fp16 = softmax(axis = var_30355, x = aw_4485_cast_fp16)[name = tensor<string, []>("op_52893_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52894_cast_fp16 = softmax(axis = var_30355, x = aw_4487_cast_fp16)[name = tensor<string, []>("op_52894_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52895_cast_fp16 = softmax(axis = var_30355, x = aw_4489_cast_fp16)[name = tensor<string, []>("op_52895_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52896_cast_fp16 = softmax(axis = var_30355, x = aw_4491_cast_fp16)[name = tensor<string, []>("op_52896_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52897_cast_fp16 = softmax(axis = var_30355, x = aw_4493_cast_fp16)[name = tensor<string, []>("op_52897_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52898_cast_fp16 = softmax(axis = var_30355, x = aw_4495_cast_fp16)[name = tensor<string, []>("op_52898_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52899_cast_fp16 = softmax(axis = var_30355, x = aw_4497_cast_fp16)[name = tensor<string, []>("op_52899_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52900_cast_fp16 = softmax(axis = var_30355, x = aw_4499_cast_fp16)[name = tensor<string, []>("op_52900_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52901_cast_fp16 = softmax(axis = var_30355, x = aw_4501_cast_fp16)[name = tensor<string, []>("op_52901_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52902_cast_fp16 = softmax(axis = var_30355, x = aw_4503_cast_fp16)[name = tensor<string, []>("op_52902_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52903_cast_fp16 = softmax(axis = var_30355, x = aw_4505_cast_fp16)[name = tensor<string, []>("op_52903_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52904_cast_fp16 = softmax(axis = var_30355, x = aw_4507_cast_fp16)[name = tensor<string, []>("op_52904_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52905_cast_fp16 = softmax(axis = var_30355, x = aw_4509_cast_fp16)[name = tensor<string, []>("op_52905_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52906_cast_fp16 = softmax(axis = var_30355, x = aw_4511_cast_fp16)[name = tensor<string, []>("op_52906_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52907_cast_fp16 = softmax(axis = var_30355, x = aw_4513_cast_fp16)[name = tensor<string, []>("op_52907_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52908_cast_fp16 = softmax(axis = var_30355, x = aw_4515_cast_fp16)[name = tensor<string, []>("op_52908_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52909_cast_fp16 = softmax(axis = var_30355, x = aw_4517_cast_fp16)[name = tensor<string, []>("op_52909_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52910_cast_fp16 = softmax(axis = var_30355, x = aw_4519_cast_fp16)[name = tensor<string, []>("op_52910_cast_fp16")];
            tensor<string, []> var_52912_equation_0 = const()[name = tensor<string, []>("op_52912_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52912_cast_fp16 = einsum(equation = var_52912_equation_0, values = (var_52732_cast_fp16, var_52891_cast_fp16))[name = tensor<string, []>("op_52912_cast_fp16")];
            tensor<string, []> var_52914_equation_0 = const()[name = tensor<string, []>("op_52914_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52914_cast_fp16 = einsum(equation = var_52914_equation_0, values = (var_52736_cast_fp16, var_52892_cast_fp16))[name = tensor<string, []>("op_52914_cast_fp16")];
            tensor<string, []> var_52916_equation_0 = const()[name = tensor<string, []>("op_52916_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52916_cast_fp16 = einsum(equation = var_52916_equation_0, values = (var_52740_cast_fp16, var_52893_cast_fp16))[name = tensor<string, []>("op_52916_cast_fp16")];
            tensor<string, []> var_52918_equation_0 = const()[name = tensor<string, []>("op_52918_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52918_cast_fp16 = einsum(equation = var_52918_equation_0, values = (var_52744_cast_fp16, var_52894_cast_fp16))[name = tensor<string, []>("op_52918_cast_fp16")];
            tensor<string, []> var_52920_equation_0 = const()[name = tensor<string, []>("op_52920_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52920_cast_fp16 = einsum(equation = var_52920_equation_0, values = (var_52748_cast_fp16, var_52895_cast_fp16))[name = tensor<string, []>("op_52920_cast_fp16")];
            tensor<string, []> var_52922_equation_0 = const()[name = tensor<string, []>("op_52922_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52922_cast_fp16 = einsum(equation = var_52922_equation_0, values = (var_52752_cast_fp16, var_52896_cast_fp16))[name = tensor<string, []>("op_52922_cast_fp16")];
            tensor<string, []> var_52924_equation_0 = const()[name = tensor<string, []>("op_52924_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52924_cast_fp16 = einsum(equation = var_52924_equation_0, values = (var_52756_cast_fp16, var_52897_cast_fp16))[name = tensor<string, []>("op_52924_cast_fp16")];
            tensor<string, []> var_52926_equation_0 = const()[name = tensor<string, []>("op_52926_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52926_cast_fp16 = einsum(equation = var_52926_equation_0, values = (var_52760_cast_fp16, var_52898_cast_fp16))[name = tensor<string, []>("op_52926_cast_fp16")];
            tensor<string, []> var_52928_equation_0 = const()[name = tensor<string, []>("op_52928_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52928_cast_fp16 = einsum(equation = var_52928_equation_0, values = (var_52764_cast_fp16, var_52899_cast_fp16))[name = tensor<string, []>("op_52928_cast_fp16")];
            tensor<string, []> var_52930_equation_0 = const()[name = tensor<string, []>("op_52930_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52930_cast_fp16 = einsum(equation = var_52930_equation_0, values = (var_52768_cast_fp16, var_52900_cast_fp16))[name = tensor<string, []>("op_52930_cast_fp16")];
            tensor<string, []> var_52932_equation_0 = const()[name = tensor<string, []>("op_52932_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52932_cast_fp16 = einsum(equation = var_52932_equation_0, values = (var_52772_cast_fp16, var_52901_cast_fp16))[name = tensor<string, []>("op_52932_cast_fp16")];
            tensor<string, []> var_52934_equation_0 = const()[name = tensor<string, []>("op_52934_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52934_cast_fp16 = einsum(equation = var_52934_equation_0, values = (var_52776_cast_fp16, var_52902_cast_fp16))[name = tensor<string, []>("op_52934_cast_fp16")];
            tensor<string, []> var_52936_equation_0 = const()[name = tensor<string, []>("op_52936_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52936_cast_fp16 = einsum(equation = var_52936_equation_0, values = (var_52780_cast_fp16, var_52903_cast_fp16))[name = tensor<string, []>("op_52936_cast_fp16")];
            tensor<string, []> var_52938_equation_0 = const()[name = tensor<string, []>("op_52938_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52938_cast_fp16 = einsum(equation = var_52938_equation_0, values = (var_52784_cast_fp16, var_52904_cast_fp16))[name = tensor<string, []>("op_52938_cast_fp16")];
            tensor<string, []> var_52940_equation_0 = const()[name = tensor<string, []>("op_52940_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52940_cast_fp16 = einsum(equation = var_52940_equation_0, values = (var_52788_cast_fp16, var_52905_cast_fp16))[name = tensor<string, []>("op_52940_cast_fp16")];
            tensor<string, []> var_52942_equation_0 = const()[name = tensor<string, []>("op_52942_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52942_cast_fp16 = einsum(equation = var_52942_equation_0, values = (var_52792_cast_fp16, var_52906_cast_fp16))[name = tensor<string, []>("op_52942_cast_fp16")];
            tensor<string, []> var_52944_equation_0 = const()[name = tensor<string, []>("op_52944_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52944_cast_fp16 = einsum(equation = var_52944_equation_0, values = (var_52796_cast_fp16, var_52907_cast_fp16))[name = tensor<string, []>("op_52944_cast_fp16")];
            tensor<string, []> var_52946_equation_0 = const()[name = tensor<string, []>("op_52946_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52946_cast_fp16 = einsum(equation = var_52946_equation_0, values = (var_52800_cast_fp16, var_52908_cast_fp16))[name = tensor<string, []>("op_52946_cast_fp16")];
            tensor<string, []> var_52948_equation_0 = const()[name = tensor<string, []>("op_52948_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52948_cast_fp16 = einsum(equation = var_52948_equation_0, values = (var_52804_cast_fp16, var_52909_cast_fp16))[name = tensor<string, []>("op_52948_cast_fp16")];
            tensor<string, []> var_52950_equation_0 = const()[name = tensor<string, []>("op_52950_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52950_cast_fp16 = einsum(equation = var_52950_equation_0, values = (var_52808_cast_fp16, var_52910_cast_fp16))[name = tensor<string, []>("op_52950_cast_fp16")];
            tensor<bool, []> input_667_interleave_0 = const()[name = tensor<string, []>("input_667_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_667_cast_fp16 = concat(axis = var_30355, interleave = input_667_interleave_0, values = (var_52912_cast_fp16, var_52914_cast_fp16, var_52916_cast_fp16, var_52918_cast_fp16, var_52920_cast_fp16, var_52922_cast_fp16, var_52924_cast_fp16, var_52926_cast_fp16, var_52928_cast_fp16, var_52930_cast_fp16, var_52932_cast_fp16, var_52934_cast_fp16, var_52936_cast_fp16, var_52938_cast_fp16, var_52940_cast_fp16, var_52942_cast_fp16, var_52944_cast_fp16, var_52946_cast_fp16, var_52948_cast_fp16, var_52950_cast_fp16))[name = tensor<string, []>("input_667_cast_fp16")];
            tensor<string, []> var_52960_pad_type_0 = const()[name = tensor<string, []>("op_52960_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_52960_strides_0 = const()[name = tensor<string, []>("op_52960_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_52960_pad_0 = const()[name = tensor<string, []>("op_52960_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_52960_dilations_0 = const()[name = tensor<string, []>("op_52960_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_52960_groups_0 = const()[name = tensor<string, []>("op_52960_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_4_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(706325056))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(707553920))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_4_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_4_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_4_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(707554112)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_52960_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_4_attn1_to_out_0_bias_to_fp16, dilations = var_52960_dilations_0, groups = var_52960_groups_0, pad = var_52960_pad_0, pad_type = var_52960_pad_type_0, strides = var_52960_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_4_attn1_to_out_0_weight_to_fp16_palettized, x = input_667_cast_fp16)[name = tensor<string, []>("op_52960_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_351_cast_fp16 = add(x = var_52960_cast_fp16, y = inputs_349_cast_fp16)[name = tensor<string, []>("inputs_351_cast_fp16")];
            tensor<int32, [1]> hidden_states_461_axes_0 = const()[name = tensor<string, []>("hidden_states_461_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_461_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_461_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(707556736)))];
            tensor<fp16, [1280]> hidden_states_461_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_461_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(707559360)))];
            tensor<fp16, []> var_52970_to_fp16 = const()[name = tensor<string, []>("op_52970_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_461_cast_fp16 = layer_norm(axes = hidden_states_461_axes_0, beta = hidden_states_461_beta_0_to_fp16, epsilon = var_52970_to_fp16, gamma = hidden_states_461_gamma_0_to_fp16, x = inputs_351_cast_fp16)[name = tensor<string, []>("hidden_states_461_cast_fp16")];
            tensor<string, []> q_235_pad_type_0 = const()[name = tensor<string, []>("q_235_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_235_strides_0 = const()[name = tensor<string, []>("q_235_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_235_pad_0 = const()[name = tensor<string, []>("q_235_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_235_dilations_0 = const()[name = tensor<string, []>("q_235_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_235_groups_0 = const()[name = tensor<string, []>("q_235_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_4_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(707561984))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(708790848))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_4_attn2_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_235_cast_fp16 = conv(dilations = q_235_dilations_0, groups = q_235_groups_0, pad = q_235_pad_0, pad_type = q_235_pad_type_0, strides = q_235_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_4_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_461_cast_fp16)[name = tensor<string, []>("q_235_cast_fp16")];
            tensor<string, []> k_469_pad_type_0 = const()[name = tensor<string, []>("k_469_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_469_strides_0 = const()[name = tensor<string, []>("k_469_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_469_pad_0 = const()[name = tensor<string, []>("k_469_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_469_dilations_0 = const()[name = tensor<string, []>("k_469_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_469_groups_0 = const()[name = tensor<string, []>("k_469_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_4_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(708791040))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(710757184))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_4_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> k_469_cast_fp16 = conv(dilations = k_469_dilations_0, groups = k_469_groups_0, pad = k_469_pad_0, pad_type = k_469_pad_type_0, strides = k_469_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_4_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("k_469_cast_fp16")];
            tensor<string, []> v_235_pad_type_0 = const()[name = tensor<string, []>("v_235_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_235_strides_0 = const()[name = tensor<string, []>("v_235_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_235_pad_0 = const()[name = tensor<string, []>("v_235_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_235_dilations_0 = const()[name = tensor<string, []>("v_235_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_235_groups_0 = const()[name = tensor<string, []>("v_235_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_4_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(710757376))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(712723520))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_4_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> v_235_cast_fp16 = conv(dilations = v_235_dilations_0, groups = v_235_groups_0, pad = v_235_pad_0, pad_type = v_235_pad_type_0, strides = v_235_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_4_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("v_235_cast_fp16")];
            tensor<int32, [4]> var_53003_begin_0 = const()[name = tensor<string, []>("op_53003_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_53003_end_0 = const()[name = tensor<string, []>("op_53003_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_53003_end_mask_0 = const()[name = tensor<string, []>("op_53003_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53003_cast_fp16 = slice_by_index(begin = var_53003_begin_0, end = var_53003_end_0, end_mask = var_53003_end_mask_0, x = q_235_cast_fp16)[name = tensor<string, []>("op_53003_cast_fp16")];
            tensor<int32, [4]> var_53007_begin_0 = const()[name = tensor<string, []>("op_53007_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_53007_end_0 = const()[name = tensor<string, []>("op_53007_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_53007_end_mask_0 = const()[name = tensor<string, []>("op_53007_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53007_cast_fp16 = slice_by_index(begin = var_53007_begin_0, end = var_53007_end_0, end_mask = var_53007_end_mask_0, x = q_235_cast_fp16)[name = tensor<string, []>("op_53007_cast_fp16")];
            tensor<int32, [4]> var_53011_begin_0 = const()[name = tensor<string, []>("op_53011_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_53011_end_0 = const()[name = tensor<string, []>("op_53011_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_53011_end_mask_0 = const()[name = tensor<string, []>("op_53011_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53011_cast_fp16 = slice_by_index(begin = var_53011_begin_0, end = var_53011_end_0, end_mask = var_53011_end_mask_0, x = q_235_cast_fp16)[name = tensor<string, []>("op_53011_cast_fp16")];
            tensor<int32, [4]> var_53015_begin_0 = const()[name = tensor<string, []>("op_53015_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_53015_end_0 = const()[name = tensor<string, []>("op_53015_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_53015_end_mask_0 = const()[name = tensor<string, []>("op_53015_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53015_cast_fp16 = slice_by_index(begin = var_53015_begin_0, end = var_53015_end_0, end_mask = var_53015_end_mask_0, x = q_235_cast_fp16)[name = tensor<string, []>("op_53015_cast_fp16")];
            tensor<int32, [4]> var_53019_begin_0 = const()[name = tensor<string, []>("op_53019_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_53019_end_0 = const()[name = tensor<string, []>("op_53019_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_53019_end_mask_0 = const()[name = tensor<string, []>("op_53019_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53019_cast_fp16 = slice_by_index(begin = var_53019_begin_0, end = var_53019_end_0, end_mask = var_53019_end_mask_0, x = q_235_cast_fp16)[name = tensor<string, []>("op_53019_cast_fp16")];
            tensor<int32, [4]> var_53023_begin_0 = const()[name = tensor<string, []>("op_53023_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_53023_end_0 = const()[name = tensor<string, []>("op_53023_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_53023_end_mask_0 = const()[name = tensor<string, []>("op_53023_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53023_cast_fp16 = slice_by_index(begin = var_53023_begin_0, end = var_53023_end_0, end_mask = var_53023_end_mask_0, x = q_235_cast_fp16)[name = tensor<string, []>("op_53023_cast_fp16")];
            tensor<int32, [4]> var_53027_begin_0 = const()[name = tensor<string, []>("op_53027_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_53027_end_0 = const()[name = tensor<string, []>("op_53027_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_53027_end_mask_0 = const()[name = tensor<string, []>("op_53027_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53027_cast_fp16 = slice_by_index(begin = var_53027_begin_0, end = var_53027_end_0, end_mask = var_53027_end_mask_0, x = q_235_cast_fp16)[name = tensor<string, []>("op_53027_cast_fp16")];
            tensor<int32, [4]> var_53031_begin_0 = const()[name = tensor<string, []>("op_53031_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_53031_end_0 = const()[name = tensor<string, []>("op_53031_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_53031_end_mask_0 = const()[name = tensor<string, []>("op_53031_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53031_cast_fp16 = slice_by_index(begin = var_53031_begin_0, end = var_53031_end_0, end_mask = var_53031_end_mask_0, x = q_235_cast_fp16)[name = tensor<string, []>("op_53031_cast_fp16")];
            tensor<int32, [4]> var_53035_begin_0 = const()[name = tensor<string, []>("op_53035_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_53035_end_0 = const()[name = tensor<string, []>("op_53035_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_53035_end_mask_0 = const()[name = tensor<string, []>("op_53035_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53035_cast_fp16 = slice_by_index(begin = var_53035_begin_0, end = var_53035_end_0, end_mask = var_53035_end_mask_0, x = q_235_cast_fp16)[name = tensor<string, []>("op_53035_cast_fp16")];
            tensor<int32, [4]> var_53039_begin_0 = const()[name = tensor<string, []>("op_53039_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_53039_end_0 = const()[name = tensor<string, []>("op_53039_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_53039_end_mask_0 = const()[name = tensor<string, []>("op_53039_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53039_cast_fp16 = slice_by_index(begin = var_53039_begin_0, end = var_53039_end_0, end_mask = var_53039_end_mask_0, x = q_235_cast_fp16)[name = tensor<string, []>("op_53039_cast_fp16")];
            tensor<int32, [4]> var_53043_begin_0 = const()[name = tensor<string, []>("op_53043_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_53043_end_0 = const()[name = tensor<string, []>("op_53043_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_53043_end_mask_0 = const()[name = tensor<string, []>("op_53043_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53043_cast_fp16 = slice_by_index(begin = var_53043_begin_0, end = var_53043_end_0, end_mask = var_53043_end_mask_0, x = q_235_cast_fp16)[name = tensor<string, []>("op_53043_cast_fp16")];
            tensor<int32, [4]> var_53047_begin_0 = const()[name = tensor<string, []>("op_53047_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_53047_end_0 = const()[name = tensor<string, []>("op_53047_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_53047_end_mask_0 = const()[name = tensor<string, []>("op_53047_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53047_cast_fp16 = slice_by_index(begin = var_53047_begin_0, end = var_53047_end_0, end_mask = var_53047_end_mask_0, x = q_235_cast_fp16)[name = tensor<string, []>("op_53047_cast_fp16")];
            tensor<int32, [4]> var_53051_begin_0 = const()[name = tensor<string, []>("op_53051_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_53051_end_0 = const()[name = tensor<string, []>("op_53051_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_53051_end_mask_0 = const()[name = tensor<string, []>("op_53051_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53051_cast_fp16 = slice_by_index(begin = var_53051_begin_0, end = var_53051_end_0, end_mask = var_53051_end_mask_0, x = q_235_cast_fp16)[name = tensor<string, []>("op_53051_cast_fp16")];
            tensor<int32, [4]> var_53055_begin_0 = const()[name = tensor<string, []>("op_53055_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_53055_end_0 = const()[name = tensor<string, []>("op_53055_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_53055_end_mask_0 = const()[name = tensor<string, []>("op_53055_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53055_cast_fp16 = slice_by_index(begin = var_53055_begin_0, end = var_53055_end_0, end_mask = var_53055_end_mask_0, x = q_235_cast_fp16)[name = tensor<string, []>("op_53055_cast_fp16")];
            tensor<int32, [4]> var_53059_begin_0 = const()[name = tensor<string, []>("op_53059_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_53059_end_0 = const()[name = tensor<string, []>("op_53059_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_53059_end_mask_0 = const()[name = tensor<string, []>("op_53059_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53059_cast_fp16 = slice_by_index(begin = var_53059_begin_0, end = var_53059_end_0, end_mask = var_53059_end_mask_0, x = q_235_cast_fp16)[name = tensor<string, []>("op_53059_cast_fp16")];
            tensor<int32, [4]> var_53063_begin_0 = const()[name = tensor<string, []>("op_53063_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_53063_end_0 = const()[name = tensor<string, []>("op_53063_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_53063_end_mask_0 = const()[name = tensor<string, []>("op_53063_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53063_cast_fp16 = slice_by_index(begin = var_53063_begin_0, end = var_53063_end_0, end_mask = var_53063_end_mask_0, x = q_235_cast_fp16)[name = tensor<string, []>("op_53063_cast_fp16")];
            tensor<int32, [4]> var_53067_begin_0 = const()[name = tensor<string, []>("op_53067_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_53067_end_0 = const()[name = tensor<string, []>("op_53067_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_53067_end_mask_0 = const()[name = tensor<string, []>("op_53067_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53067_cast_fp16 = slice_by_index(begin = var_53067_begin_0, end = var_53067_end_0, end_mask = var_53067_end_mask_0, x = q_235_cast_fp16)[name = tensor<string, []>("op_53067_cast_fp16")];
            tensor<int32, [4]> var_53071_begin_0 = const()[name = tensor<string, []>("op_53071_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_53071_end_0 = const()[name = tensor<string, []>("op_53071_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_53071_end_mask_0 = const()[name = tensor<string, []>("op_53071_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53071_cast_fp16 = slice_by_index(begin = var_53071_begin_0, end = var_53071_end_0, end_mask = var_53071_end_mask_0, x = q_235_cast_fp16)[name = tensor<string, []>("op_53071_cast_fp16")];
            tensor<int32, [4]> var_53075_begin_0 = const()[name = tensor<string, []>("op_53075_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_53075_end_0 = const()[name = tensor<string, []>("op_53075_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_53075_end_mask_0 = const()[name = tensor<string, []>("op_53075_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53075_cast_fp16 = slice_by_index(begin = var_53075_begin_0, end = var_53075_end_0, end_mask = var_53075_end_mask_0, x = q_235_cast_fp16)[name = tensor<string, []>("op_53075_cast_fp16")];
            tensor<int32, [4]> var_53079_begin_0 = const()[name = tensor<string, []>("op_53079_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_53079_end_0 = const()[name = tensor<string, []>("op_53079_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_53079_end_mask_0 = const()[name = tensor<string, []>("op_53079_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53079_cast_fp16 = slice_by_index(begin = var_53079_begin_0, end = var_53079_end_0, end_mask = var_53079_end_mask_0, x = q_235_cast_fp16)[name = tensor<string, []>("op_53079_cast_fp16")];
            tensor<int32, [4]> k_471_perm_0 = const()[name = tensor<string, []>("k_471_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_53086_begin_0 = const()[name = tensor<string, []>("op_53086_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_53086_end_0 = const()[name = tensor<string, []>("op_53086_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_53086_end_mask_0 = const()[name = tensor<string, []>("op_53086_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> k_471_cast_fp16 = transpose(perm = k_471_perm_0, x = k_469_cast_fp16)[name = tensor<string, []>("transpose_22")];
            tensor<fp16, [2, 77, 1, 64]> var_53086_cast_fp16 = slice_by_index(begin = var_53086_begin_0, end = var_53086_end_0, end_mask = var_53086_end_mask_0, x = k_471_cast_fp16)[name = tensor<string, []>("op_53086_cast_fp16")];
            tensor<int32, [4]> var_53090_begin_0 = const()[name = tensor<string, []>("op_53090_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_53090_end_0 = const()[name = tensor<string, []>("op_53090_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_53090_end_mask_0 = const()[name = tensor<string, []>("op_53090_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_53090_cast_fp16 = slice_by_index(begin = var_53090_begin_0, end = var_53090_end_0, end_mask = var_53090_end_mask_0, x = k_471_cast_fp16)[name = tensor<string, []>("op_53090_cast_fp16")];
            tensor<int32, [4]> var_53094_begin_0 = const()[name = tensor<string, []>("op_53094_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_53094_end_0 = const()[name = tensor<string, []>("op_53094_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_53094_end_mask_0 = const()[name = tensor<string, []>("op_53094_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_53094_cast_fp16 = slice_by_index(begin = var_53094_begin_0, end = var_53094_end_0, end_mask = var_53094_end_mask_0, x = k_471_cast_fp16)[name = tensor<string, []>("op_53094_cast_fp16")];
            tensor<int32, [4]> var_53098_begin_0 = const()[name = tensor<string, []>("op_53098_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_53098_end_0 = const()[name = tensor<string, []>("op_53098_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_53098_end_mask_0 = const()[name = tensor<string, []>("op_53098_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_53098_cast_fp16 = slice_by_index(begin = var_53098_begin_0, end = var_53098_end_0, end_mask = var_53098_end_mask_0, x = k_471_cast_fp16)[name = tensor<string, []>("op_53098_cast_fp16")];
            tensor<int32, [4]> var_53102_begin_0 = const()[name = tensor<string, []>("op_53102_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_53102_end_0 = const()[name = tensor<string, []>("op_53102_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_53102_end_mask_0 = const()[name = tensor<string, []>("op_53102_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_53102_cast_fp16 = slice_by_index(begin = var_53102_begin_0, end = var_53102_end_0, end_mask = var_53102_end_mask_0, x = k_471_cast_fp16)[name = tensor<string, []>("op_53102_cast_fp16")];
            tensor<int32, [4]> var_53106_begin_0 = const()[name = tensor<string, []>("op_53106_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_53106_end_0 = const()[name = tensor<string, []>("op_53106_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_53106_end_mask_0 = const()[name = tensor<string, []>("op_53106_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_53106_cast_fp16 = slice_by_index(begin = var_53106_begin_0, end = var_53106_end_0, end_mask = var_53106_end_mask_0, x = k_471_cast_fp16)[name = tensor<string, []>("op_53106_cast_fp16")];
            tensor<int32, [4]> var_53110_begin_0 = const()[name = tensor<string, []>("op_53110_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_53110_end_0 = const()[name = tensor<string, []>("op_53110_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_53110_end_mask_0 = const()[name = tensor<string, []>("op_53110_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_53110_cast_fp16 = slice_by_index(begin = var_53110_begin_0, end = var_53110_end_0, end_mask = var_53110_end_mask_0, x = k_471_cast_fp16)[name = tensor<string, []>("op_53110_cast_fp16")];
            tensor<int32, [4]> var_53114_begin_0 = const()[name = tensor<string, []>("op_53114_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_53114_end_0 = const()[name = tensor<string, []>("op_53114_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_53114_end_mask_0 = const()[name = tensor<string, []>("op_53114_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_53114_cast_fp16 = slice_by_index(begin = var_53114_begin_0, end = var_53114_end_0, end_mask = var_53114_end_mask_0, x = k_471_cast_fp16)[name = tensor<string, []>("op_53114_cast_fp16")];
            tensor<int32, [4]> var_53118_begin_0 = const()[name = tensor<string, []>("op_53118_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_53118_end_0 = const()[name = tensor<string, []>("op_53118_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_53118_end_mask_0 = const()[name = tensor<string, []>("op_53118_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_53118_cast_fp16 = slice_by_index(begin = var_53118_begin_0, end = var_53118_end_0, end_mask = var_53118_end_mask_0, x = k_471_cast_fp16)[name = tensor<string, []>("op_53118_cast_fp16")];
            tensor<int32, [4]> var_53122_begin_0 = const()[name = tensor<string, []>("op_53122_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_53122_end_0 = const()[name = tensor<string, []>("op_53122_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_53122_end_mask_0 = const()[name = tensor<string, []>("op_53122_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_53122_cast_fp16 = slice_by_index(begin = var_53122_begin_0, end = var_53122_end_0, end_mask = var_53122_end_mask_0, x = k_471_cast_fp16)[name = tensor<string, []>("op_53122_cast_fp16")];
            tensor<int32, [4]> var_53126_begin_0 = const()[name = tensor<string, []>("op_53126_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_53126_end_0 = const()[name = tensor<string, []>("op_53126_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_53126_end_mask_0 = const()[name = tensor<string, []>("op_53126_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_53126_cast_fp16 = slice_by_index(begin = var_53126_begin_0, end = var_53126_end_0, end_mask = var_53126_end_mask_0, x = k_471_cast_fp16)[name = tensor<string, []>("op_53126_cast_fp16")];
            tensor<int32, [4]> var_53130_begin_0 = const()[name = tensor<string, []>("op_53130_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_53130_end_0 = const()[name = tensor<string, []>("op_53130_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_53130_end_mask_0 = const()[name = tensor<string, []>("op_53130_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_53130_cast_fp16 = slice_by_index(begin = var_53130_begin_0, end = var_53130_end_0, end_mask = var_53130_end_mask_0, x = k_471_cast_fp16)[name = tensor<string, []>("op_53130_cast_fp16")];
            tensor<int32, [4]> var_53134_begin_0 = const()[name = tensor<string, []>("op_53134_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_53134_end_0 = const()[name = tensor<string, []>("op_53134_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_53134_end_mask_0 = const()[name = tensor<string, []>("op_53134_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_53134_cast_fp16 = slice_by_index(begin = var_53134_begin_0, end = var_53134_end_0, end_mask = var_53134_end_mask_0, x = k_471_cast_fp16)[name = tensor<string, []>("op_53134_cast_fp16")];
            tensor<int32, [4]> var_53138_begin_0 = const()[name = tensor<string, []>("op_53138_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_53138_end_0 = const()[name = tensor<string, []>("op_53138_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_53138_end_mask_0 = const()[name = tensor<string, []>("op_53138_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_53138_cast_fp16 = slice_by_index(begin = var_53138_begin_0, end = var_53138_end_0, end_mask = var_53138_end_mask_0, x = k_471_cast_fp16)[name = tensor<string, []>("op_53138_cast_fp16")];
            tensor<int32, [4]> var_53142_begin_0 = const()[name = tensor<string, []>("op_53142_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_53142_end_0 = const()[name = tensor<string, []>("op_53142_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_53142_end_mask_0 = const()[name = tensor<string, []>("op_53142_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_53142_cast_fp16 = slice_by_index(begin = var_53142_begin_0, end = var_53142_end_0, end_mask = var_53142_end_mask_0, x = k_471_cast_fp16)[name = tensor<string, []>("op_53142_cast_fp16")];
            tensor<int32, [4]> var_53146_begin_0 = const()[name = tensor<string, []>("op_53146_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_53146_end_0 = const()[name = tensor<string, []>("op_53146_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_53146_end_mask_0 = const()[name = tensor<string, []>("op_53146_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_53146_cast_fp16 = slice_by_index(begin = var_53146_begin_0, end = var_53146_end_0, end_mask = var_53146_end_mask_0, x = k_471_cast_fp16)[name = tensor<string, []>("op_53146_cast_fp16")];
            tensor<int32, [4]> var_53150_begin_0 = const()[name = tensor<string, []>("op_53150_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_53150_end_0 = const()[name = tensor<string, []>("op_53150_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_53150_end_mask_0 = const()[name = tensor<string, []>("op_53150_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_53150_cast_fp16 = slice_by_index(begin = var_53150_begin_0, end = var_53150_end_0, end_mask = var_53150_end_mask_0, x = k_471_cast_fp16)[name = tensor<string, []>("op_53150_cast_fp16")];
            tensor<int32, [4]> var_53154_begin_0 = const()[name = tensor<string, []>("op_53154_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_53154_end_0 = const()[name = tensor<string, []>("op_53154_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_53154_end_mask_0 = const()[name = tensor<string, []>("op_53154_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_53154_cast_fp16 = slice_by_index(begin = var_53154_begin_0, end = var_53154_end_0, end_mask = var_53154_end_mask_0, x = k_471_cast_fp16)[name = tensor<string, []>("op_53154_cast_fp16")];
            tensor<int32, [4]> var_53158_begin_0 = const()[name = tensor<string, []>("op_53158_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_53158_end_0 = const()[name = tensor<string, []>("op_53158_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_53158_end_mask_0 = const()[name = tensor<string, []>("op_53158_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_53158_cast_fp16 = slice_by_index(begin = var_53158_begin_0, end = var_53158_end_0, end_mask = var_53158_end_mask_0, x = k_471_cast_fp16)[name = tensor<string, []>("op_53158_cast_fp16")];
            tensor<int32, [4]> var_53162_begin_0 = const()[name = tensor<string, []>("op_53162_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_53162_end_0 = const()[name = tensor<string, []>("op_53162_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_53162_end_mask_0 = const()[name = tensor<string, []>("op_53162_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_53162_cast_fp16 = slice_by_index(begin = var_53162_begin_0, end = var_53162_end_0, end_mask = var_53162_end_mask_0, x = k_471_cast_fp16)[name = tensor<string, []>("op_53162_cast_fp16")];
            tensor<int32, [4]> var_53164_begin_0 = const()[name = tensor<string, []>("op_53164_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_53164_end_0 = const()[name = tensor<string, []>("op_53164_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_53164_end_mask_0 = const()[name = tensor<string, []>("op_53164_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_53164_cast_fp16 = slice_by_index(begin = var_53164_begin_0, end = var_53164_end_0, end_mask = var_53164_end_mask_0, x = v_235_cast_fp16)[name = tensor<string, []>("op_53164_cast_fp16")];
            tensor<int32, [4]> var_53168_begin_0 = const()[name = tensor<string, []>("op_53168_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_53168_end_0 = const()[name = tensor<string, []>("op_53168_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_53168_end_mask_0 = const()[name = tensor<string, []>("op_53168_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_53168_cast_fp16 = slice_by_index(begin = var_53168_begin_0, end = var_53168_end_0, end_mask = var_53168_end_mask_0, x = v_235_cast_fp16)[name = tensor<string, []>("op_53168_cast_fp16")];
            tensor<int32, [4]> var_53172_begin_0 = const()[name = tensor<string, []>("op_53172_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_53172_end_0 = const()[name = tensor<string, []>("op_53172_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_53172_end_mask_0 = const()[name = tensor<string, []>("op_53172_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_53172_cast_fp16 = slice_by_index(begin = var_53172_begin_0, end = var_53172_end_0, end_mask = var_53172_end_mask_0, x = v_235_cast_fp16)[name = tensor<string, []>("op_53172_cast_fp16")];
            tensor<int32, [4]> var_53176_begin_0 = const()[name = tensor<string, []>("op_53176_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_53176_end_0 = const()[name = tensor<string, []>("op_53176_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_53176_end_mask_0 = const()[name = tensor<string, []>("op_53176_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_53176_cast_fp16 = slice_by_index(begin = var_53176_begin_0, end = var_53176_end_0, end_mask = var_53176_end_mask_0, x = v_235_cast_fp16)[name = tensor<string, []>("op_53176_cast_fp16")];
            tensor<int32, [4]> var_53180_begin_0 = const()[name = tensor<string, []>("op_53180_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_53180_end_0 = const()[name = tensor<string, []>("op_53180_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_53180_end_mask_0 = const()[name = tensor<string, []>("op_53180_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_53180_cast_fp16 = slice_by_index(begin = var_53180_begin_0, end = var_53180_end_0, end_mask = var_53180_end_mask_0, x = v_235_cast_fp16)[name = tensor<string, []>("op_53180_cast_fp16")];
            tensor<int32, [4]> var_53184_begin_0 = const()[name = tensor<string, []>("op_53184_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_53184_end_0 = const()[name = tensor<string, []>("op_53184_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_53184_end_mask_0 = const()[name = tensor<string, []>("op_53184_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_53184_cast_fp16 = slice_by_index(begin = var_53184_begin_0, end = var_53184_end_0, end_mask = var_53184_end_mask_0, x = v_235_cast_fp16)[name = tensor<string, []>("op_53184_cast_fp16")];
            tensor<int32, [4]> var_53188_begin_0 = const()[name = tensor<string, []>("op_53188_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_53188_end_0 = const()[name = tensor<string, []>("op_53188_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_53188_end_mask_0 = const()[name = tensor<string, []>("op_53188_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_53188_cast_fp16 = slice_by_index(begin = var_53188_begin_0, end = var_53188_end_0, end_mask = var_53188_end_mask_0, x = v_235_cast_fp16)[name = tensor<string, []>("op_53188_cast_fp16")];
            tensor<int32, [4]> var_53192_begin_0 = const()[name = tensor<string, []>("op_53192_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_53192_end_0 = const()[name = tensor<string, []>("op_53192_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_53192_end_mask_0 = const()[name = tensor<string, []>("op_53192_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_53192_cast_fp16 = slice_by_index(begin = var_53192_begin_0, end = var_53192_end_0, end_mask = var_53192_end_mask_0, x = v_235_cast_fp16)[name = tensor<string, []>("op_53192_cast_fp16")];
            tensor<int32, [4]> var_53196_begin_0 = const()[name = tensor<string, []>("op_53196_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_53196_end_0 = const()[name = tensor<string, []>("op_53196_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_53196_end_mask_0 = const()[name = tensor<string, []>("op_53196_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_53196_cast_fp16 = slice_by_index(begin = var_53196_begin_0, end = var_53196_end_0, end_mask = var_53196_end_mask_0, x = v_235_cast_fp16)[name = tensor<string, []>("op_53196_cast_fp16")];
            tensor<int32, [4]> var_53200_begin_0 = const()[name = tensor<string, []>("op_53200_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_53200_end_0 = const()[name = tensor<string, []>("op_53200_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_53200_end_mask_0 = const()[name = tensor<string, []>("op_53200_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_53200_cast_fp16 = slice_by_index(begin = var_53200_begin_0, end = var_53200_end_0, end_mask = var_53200_end_mask_0, x = v_235_cast_fp16)[name = tensor<string, []>("op_53200_cast_fp16")];
            tensor<int32, [4]> var_53204_begin_0 = const()[name = tensor<string, []>("op_53204_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_53204_end_0 = const()[name = tensor<string, []>("op_53204_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_53204_end_mask_0 = const()[name = tensor<string, []>("op_53204_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_53204_cast_fp16 = slice_by_index(begin = var_53204_begin_0, end = var_53204_end_0, end_mask = var_53204_end_mask_0, x = v_235_cast_fp16)[name = tensor<string, []>("op_53204_cast_fp16")];
            tensor<int32, [4]> var_53208_begin_0 = const()[name = tensor<string, []>("op_53208_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_53208_end_0 = const()[name = tensor<string, []>("op_53208_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_53208_end_mask_0 = const()[name = tensor<string, []>("op_53208_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_53208_cast_fp16 = slice_by_index(begin = var_53208_begin_0, end = var_53208_end_0, end_mask = var_53208_end_mask_0, x = v_235_cast_fp16)[name = tensor<string, []>("op_53208_cast_fp16")];
            tensor<int32, [4]> var_53212_begin_0 = const()[name = tensor<string, []>("op_53212_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_53212_end_0 = const()[name = tensor<string, []>("op_53212_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_53212_end_mask_0 = const()[name = tensor<string, []>("op_53212_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_53212_cast_fp16 = slice_by_index(begin = var_53212_begin_0, end = var_53212_end_0, end_mask = var_53212_end_mask_0, x = v_235_cast_fp16)[name = tensor<string, []>("op_53212_cast_fp16")];
            tensor<int32, [4]> var_53216_begin_0 = const()[name = tensor<string, []>("op_53216_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_53216_end_0 = const()[name = tensor<string, []>("op_53216_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_53216_end_mask_0 = const()[name = tensor<string, []>("op_53216_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_53216_cast_fp16 = slice_by_index(begin = var_53216_begin_0, end = var_53216_end_0, end_mask = var_53216_end_mask_0, x = v_235_cast_fp16)[name = tensor<string, []>("op_53216_cast_fp16")];
            tensor<int32, [4]> var_53220_begin_0 = const()[name = tensor<string, []>("op_53220_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_53220_end_0 = const()[name = tensor<string, []>("op_53220_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_53220_end_mask_0 = const()[name = tensor<string, []>("op_53220_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_53220_cast_fp16 = slice_by_index(begin = var_53220_begin_0, end = var_53220_end_0, end_mask = var_53220_end_mask_0, x = v_235_cast_fp16)[name = tensor<string, []>("op_53220_cast_fp16")];
            tensor<int32, [4]> var_53224_begin_0 = const()[name = tensor<string, []>("op_53224_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_53224_end_0 = const()[name = tensor<string, []>("op_53224_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_53224_end_mask_0 = const()[name = tensor<string, []>("op_53224_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_53224_cast_fp16 = slice_by_index(begin = var_53224_begin_0, end = var_53224_end_0, end_mask = var_53224_end_mask_0, x = v_235_cast_fp16)[name = tensor<string, []>("op_53224_cast_fp16")];
            tensor<int32, [4]> var_53228_begin_0 = const()[name = tensor<string, []>("op_53228_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_53228_end_0 = const()[name = tensor<string, []>("op_53228_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_53228_end_mask_0 = const()[name = tensor<string, []>("op_53228_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_53228_cast_fp16 = slice_by_index(begin = var_53228_begin_0, end = var_53228_end_0, end_mask = var_53228_end_mask_0, x = v_235_cast_fp16)[name = tensor<string, []>("op_53228_cast_fp16")];
            tensor<int32, [4]> var_53232_begin_0 = const()[name = tensor<string, []>("op_53232_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_53232_end_0 = const()[name = tensor<string, []>("op_53232_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_53232_end_mask_0 = const()[name = tensor<string, []>("op_53232_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_53232_cast_fp16 = slice_by_index(begin = var_53232_begin_0, end = var_53232_end_0, end_mask = var_53232_end_mask_0, x = v_235_cast_fp16)[name = tensor<string, []>("op_53232_cast_fp16")];
            tensor<int32, [4]> var_53236_begin_0 = const()[name = tensor<string, []>("op_53236_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_53236_end_0 = const()[name = tensor<string, []>("op_53236_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_53236_end_mask_0 = const()[name = tensor<string, []>("op_53236_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_53236_cast_fp16 = slice_by_index(begin = var_53236_begin_0, end = var_53236_end_0, end_mask = var_53236_end_mask_0, x = v_235_cast_fp16)[name = tensor<string, []>("op_53236_cast_fp16")];
            tensor<int32, [4]> var_53240_begin_0 = const()[name = tensor<string, []>("op_53240_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_53240_end_0 = const()[name = tensor<string, []>("op_53240_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_53240_end_mask_0 = const()[name = tensor<string, []>("op_53240_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_53240_cast_fp16 = slice_by_index(begin = var_53240_begin_0, end = var_53240_end_0, end_mask = var_53240_end_mask_0, x = v_235_cast_fp16)[name = tensor<string, []>("op_53240_cast_fp16")];
            tensor<string, []> var_53244_equation_0 = const()[name = tensor<string, []>("op_53244_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_53244_cast_fp16 = einsum(equation = var_53244_equation_0, values = (var_53086_cast_fp16, var_53003_cast_fp16))[name = tensor<string, []>("op_53244_cast_fp16")];
            tensor<fp16, []> var_53245_to_fp16 = const()[name = tensor<string, []>("op_53245_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4521_cast_fp16 = mul(x = var_53244_cast_fp16, y = var_53245_to_fp16)[name = tensor<string, []>("aw_4521_cast_fp16")];
            tensor<string, []> var_53248_equation_0 = const()[name = tensor<string, []>("op_53248_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_53248_cast_fp16 = einsum(equation = var_53248_equation_0, values = (var_53090_cast_fp16, var_53007_cast_fp16))[name = tensor<string, []>("op_53248_cast_fp16")];
            tensor<fp16, []> var_53249_to_fp16 = const()[name = tensor<string, []>("op_53249_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4523_cast_fp16 = mul(x = var_53248_cast_fp16, y = var_53249_to_fp16)[name = tensor<string, []>("aw_4523_cast_fp16")];
            tensor<string, []> var_53252_equation_0 = const()[name = tensor<string, []>("op_53252_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_53252_cast_fp16 = einsum(equation = var_53252_equation_0, values = (var_53094_cast_fp16, var_53011_cast_fp16))[name = tensor<string, []>("op_53252_cast_fp16")];
            tensor<fp16, []> var_53253_to_fp16 = const()[name = tensor<string, []>("op_53253_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4525_cast_fp16 = mul(x = var_53252_cast_fp16, y = var_53253_to_fp16)[name = tensor<string, []>("aw_4525_cast_fp16")];
            tensor<string, []> var_53256_equation_0 = const()[name = tensor<string, []>("op_53256_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_53256_cast_fp16 = einsum(equation = var_53256_equation_0, values = (var_53098_cast_fp16, var_53015_cast_fp16))[name = tensor<string, []>("op_53256_cast_fp16")];
            tensor<fp16, []> var_53257_to_fp16 = const()[name = tensor<string, []>("op_53257_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4527_cast_fp16 = mul(x = var_53256_cast_fp16, y = var_53257_to_fp16)[name = tensor<string, []>("aw_4527_cast_fp16")];
            tensor<string, []> var_53260_equation_0 = const()[name = tensor<string, []>("op_53260_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_53260_cast_fp16 = einsum(equation = var_53260_equation_0, values = (var_53102_cast_fp16, var_53019_cast_fp16))[name = tensor<string, []>("op_53260_cast_fp16")];
            tensor<fp16, []> var_53261_to_fp16 = const()[name = tensor<string, []>("op_53261_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4529_cast_fp16 = mul(x = var_53260_cast_fp16, y = var_53261_to_fp16)[name = tensor<string, []>("aw_4529_cast_fp16")];
            tensor<string, []> var_53264_equation_0 = const()[name = tensor<string, []>("op_53264_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_53264_cast_fp16 = einsum(equation = var_53264_equation_0, values = (var_53106_cast_fp16, var_53023_cast_fp16))[name = tensor<string, []>("op_53264_cast_fp16")];
            tensor<fp16, []> var_53265_to_fp16 = const()[name = tensor<string, []>("op_53265_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4531_cast_fp16 = mul(x = var_53264_cast_fp16, y = var_53265_to_fp16)[name = tensor<string, []>("aw_4531_cast_fp16")];
            tensor<string, []> var_53268_equation_0 = const()[name = tensor<string, []>("op_53268_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_53268_cast_fp16 = einsum(equation = var_53268_equation_0, values = (var_53110_cast_fp16, var_53027_cast_fp16))[name = tensor<string, []>("op_53268_cast_fp16")];
            tensor<fp16, []> var_53269_to_fp16 = const()[name = tensor<string, []>("op_53269_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4533_cast_fp16 = mul(x = var_53268_cast_fp16, y = var_53269_to_fp16)[name = tensor<string, []>("aw_4533_cast_fp16")];
            tensor<string, []> var_53272_equation_0 = const()[name = tensor<string, []>("op_53272_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_53272_cast_fp16 = einsum(equation = var_53272_equation_0, values = (var_53114_cast_fp16, var_53031_cast_fp16))[name = tensor<string, []>("op_53272_cast_fp16")];
            tensor<fp16, []> var_53273_to_fp16 = const()[name = tensor<string, []>("op_53273_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4535_cast_fp16 = mul(x = var_53272_cast_fp16, y = var_53273_to_fp16)[name = tensor<string, []>("aw_4535_cast_fp16")];
            tensor<string, []> var_53276_equation_0 = const()[name = tensor<string, []>("op_53276_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_53276_cast_fp16 = einsum(equation = var_53276_equation_0, values = (var_53118_cast_fp16, var_53035_cast_fp16))[name = tensor<string, []>("op_53276_cast_fp16")];
            tensor<fp16, []> var_53277_to_fp16 = const()[name = tensor<string, []>("op_53277_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4537_cast_fp16 = mul(x = var_53276_cast_fp16, y = var_53277_to_fp16)[name = tensor<string, []>("aw_4537_cast_fp16")];
            tensor<string, []> var_53280_equation_0 = const()[name = tensor<string, []>("op_53280_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_53280_cast_fp16 = einsum(equation = var_53280_equation_0, values = (var_53122_cast_fp16, var_53039_cast_fp16))[name = tensor<string, []>("op_53280_cast_fp16")];
            tensor<fp16, []> var_53281_to_fp16 = const()[name = tensor<string, []>("op_53281_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4539_cast_fp16 = mul(x = var_53280_cast_fp16, y = var_53281_to_fp16)[name = tensor<string, []>("aw_4539_cast_fp16")];
            tensor<string, []> var_53284_equation_0 = const()[name = tensor<string, []>("op_53284_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_53284_cast_fp16 = einsum(equation = var_53284_equation_0, values = (var_53126_cast_fp16, var_53043_cast_fp16))[name = tensor<string, []>("op_53284_cast_fp16")];
            tensor<fp16, []> var_53285_to_fp16 = const()[name = tensor<string, []>("op_53285_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4541_cast_fp16 = mul(x = var_53284_cast_fp16, y = var_53285_to_fp16)[name = tensor<string, []>("aw_4541_cast_fp16")];
            tensor<string, []> var_53288_equation_0 = const()[name = tensor<string, []>("op_53288_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_53288_cast_fp16 = einsum(equation = var_53288_equation_0, values = (var_53130_cast_fp16, var_53047_cast_fp16))[name = tensor<string, []>("op_53288_cast_fp16")];
            tensor<fp16, []> var_53289_to_fp16 = const()[name = tensor<string, []>("op_53289_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4543_cast_fp16 = mul(x = var_53288_cast_fp16, y = var_53289_to_fp16)[name = tensor<string, []>("aw_4543_cast_fp16")];
            tensor<string, []> var_53292_equation_0 = const()[name = tensor<string, []>("op_53292_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_53292_cast_fp16 = einsum(equation = var_53292_equation_0, values = (var_53134_cast_fp16, var_53051_cast_fp16))[name = tensor<string, []>("op_53292_cast_fp16")];
            tensor<fp16, []> var_53293_to_fp16 = const()[name = tensor<string, []>("op_53293_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4545_cast_fp16 = mul(x = var_53292_cast_fp16, y = var_53293_to_fp16)[name = tensor<string, []>("aw_4545_cast_fp16")];
            tensor<string, []> var_53296_equation_0 = const()[name = tensor<string, []>("op_53296_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_53296_cast_fp16 = einsum(equation = var_53296_equation_0, values = (var_53138_cast_fp16, var_53055_cast_fp16))[name = tensor<string, []>("op_53296_cast_fp16")];
            tensor<fp16, []> var_53297_to_fp16 = const()[name = tensor<string, []>("op_53297_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4547_cast_fp16 = mul(x = var_53296_cast_fp16, y = var_53297_to_fp16)[name = tensor<string, []>("aw_4547_cast_fp16")];
            tensor<string, []> var_53300_equation_0 = const()[name = tensor<string, []>("op_53300_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_53300_cast_fp16 = einsum(equation = var_53300_equation_0, values = (var_53142_cast_fp16, var_53059_cast_fp16))[name = tensor<string, []>("op_53300_cast_fp16")];
            tensor<fp16, []> var_53301_to_fp16 = const()[name = tensor<string, []>("op_53301_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4549_cast_fp16 = mul(x = var_53300_cast_fp16, y = var_53301_to_fp16)[name = tensor<string, []>("aw_4549_cast_fp16")];
            tensor<string, []> var_53304_equation_0 = const()[name = tensor<string, []>("op_53304_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_53304_cast_fp16 = einsum(equation = var_53304_equation_0, values = (var_53146_cast_fp16, var_53063_cast_fp16))[name = tensor<string, []>("op_53304_cast_fp16")];
            tensor<fp16, []> var_53305_to_fp16 = const()[name = tensor<string, []>("op_53305_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4551_cast_fp16 = mul(x = var_53304_cast_fp16, y = var_53305_to_fp16)[name = tensor<string, []>("aw_4551_cast_fp16")];
            tensor<string, []> var_53308_equation_0 = const()[name = tensor<string, []>("op_53308_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_53308_cast_fp16 = einsum(equation = var_53308_equation_0, values = (var_53150_cast_fp16, var_53067_cast_fp16))[name = tensor<string, []>("op_53308_cast_fp16")];
            tensor<fp16, []> var_53309_to_fp16 = const()[name = tensor<string, []>("op_53309_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4553_cast_fp16 = mul(x = var_53308_cast_fp16, y = var_53309_to_fp16)[name = tensor<string, []>("aw_4553_cast_fp16")];
            tensor<string, []> var_53312_equation_0 = const()[name = tensor<string, []>("op_53312_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_53312_cast_fp16 = einsum(equation = var_53312_equation_0, values = (var_53154_cast_fp16, var_53071_cast_fp16))[name = tensor<string, []>("op_53312_cast_fp16")];
            tensor<fp16, []> var_53313_to_fp16 = const()[name = tensor<string, []>("op_53313_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4555_cast_fp16 = mul(x = var_53312_cast_fp16, y = var_53313_to_fp16)[name = tensor<string, []>("aw_4555_cast_fp16")];
            tensor<string, []> var_53316_equation_0 = const()[name = tensor<string, []>("op_53316_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_53316_cast_fp16 = einsum(equation = var_53316_equation_0, values = (var_53158_cast_fp16, var_53075_cast_fp16))[name = tensor<string, []>("op_53316_cast_fp16")];
            tensor<fp16, []> var_53317_to_fp16 = const()[name = tensor<string, []>("op_53317_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4557_cast_fp16 = mul(x = var_53316_cast_fp16, y = var_53317_to_fp16)[name = tensor<string, []>("aw_4557_cast_fp16")];
            tensor<string, []> var_53320_equation_0 = const()[name = tensor<string, []>("op_53320_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_53320_cast_fp16 = einsum(equation = var_53320_equation_0, values = (var_53162_cast_fp16, var_53079_cast_fp16))[name = tensor<string, []>("op_53320_cast_fp16")];
            tensor<fp16, []> var_53321_to_fp16 = const()[name = tensor<string, []>("op_53321_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4559_cast_fp16 = mul(x = var_53320_cast_fp16, y = var_53321_to_fp16)[name = tensor<string, []>("aw_4559_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_53323_cast_fp16 = softmax(axis = var_30355, x = aw_4521_cast_fp16)[name = tensor<string, []>("op_53323_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_53324_cast_fp16 = softmax(axis = var_30355, x = aw_4523_cast_fp16)[name = tensor<string, []>("op_53324_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_53325_cast_fp16 = softmax(axis = var_30355, x = aw_4525_cast_fp16)[name = tensor<string, []>("op_53325_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_53326_cast_fp16 = softmax(axis = var_30355, x = aw_4527_cast_fp16)[name = tensor<string, []>("op_53326_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_53327_cast_fp16 = softmax(axis = var_30355, x = aw_4529_cast_fp16)[name = tensor<string, []>("op_53327_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_53328_cast_fp16 = softmax(axis = var_30355, x = aw_4531_cast_fp16)[name = tensor<string, []>("op_53328_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_53329_cast_fp16 = softmax(axis = var_30355, x = aw_4533_cast_fp16)[name = tensor<string, []>("op_53329_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_53330_cast_fp16 = softmax(axis = var_30355, x = aw_4535_cast_fp16)[name = tensor<string, []>("op_53330_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_53331_cast_fp16 = softmax(axis = var_30355, x = aw_4537_cast_fp16)[name = tensor<string, []>("op_53331_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_53332_cast_fp16 = softmax(axis = var_30355, x = aw_4539_cast_fp16)[name = tensor<string, []>("op_53332_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_53333_cast_fp16 = softmax(axis = var_30355, x = aw_4541_cast_fp16)[name = tensor<string, []>("op_53333_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_53334_cast_fp16 = softmax(axis = var_30355, x = aw_4543_cast_fp16)[name = tensor<string, []>("op_53334_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_53335_cast_fp16 = softmax(axis = var_30355, x = aw_4545_cast_fp16)[name = tensor<string, []>("op_53335_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_53336_cast_fp16 = softmax(axis = var_30355, x = aw_4547_cast_fp16)[name = tensor<string, []>("op_53336_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_53337_cast_fp16 = softmax(axis = var_30355, x = aw_4549_cast_fp16)[name = tensor<string, []>("op_53337_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_53338_cast_fp16 = softmax(axis = var_30355, x = aw_4551_cast_fp16)[name = tensor<string, []>("op_53338_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_53339_cast_fp16 = softmax(axis = var_30355, x = aw_4553_cast_fp16)[name = tensor<string, []>("op_53339_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_53340_cast_fp16 = softmax(axis = var_30355, x = aw_4555_cast_fp16)[name = tensor<string, []>("op_53340_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_53341_cast_fp16 = softmax(axis = var_30355, x = aw_4557_cast_fp16)[name = tensor<string, []>("op_53341_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_53342_cast_fp16 = softmax(axis = var_30355, x = aw_4559_cast_fp16)[name = tensor<string, []>("op_53342_cast_fp16")];
            tensor<string, []> var_53344_equation_0 = const()[name = tensor<string, []>("op_53344_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53344_cast_fp16 = einsum(equation = var_53344_equation_0, values = (var_53164_cast_fp16, var_53323_cast_fp16))[name = tensor<string, []>("op_53344_cast_fp16")];
            tensor<string, []> var_53346_equation_0 = const()[name = tensor<string, []>("op_53346_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53346_cast_fp16 = einsum(equation = var_53346_equation_0, values = (var_53168_cast_fp16, var_53324_cast_fp16))[name = tensor<string, []>("op_53346_cast_fp16")];
            tensor<string, []> var_53348_equation_0 = const()[name = tensor<string, []>("op_53348_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53348_cast_fp16 = einsum(equation = var_53348_equation_0, values = (var_53172_cast_fp16, var_53325_cast_fp16))[name = tensor<string, []>("op_53348_cast_fp16")];
            tensor<string, []> var_53350_equation_0 = const()[name = tensor<string, []>("op_53350_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53350_cast_fp16 = einsum(equation = var_53350_equation_0, values = (var_53176_cast_fp16, var_53326_cast_fp16))[name = tensor<string, []>("op_53350_cast_fp16")];
            tensor<string, []> var_53352_equation_0 = const()[name = tensor<string, []>("op_53352_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53352_cast_fp16 = einsum(equation = var_53352_equation_0, values = (var_53180_cast_fp16, var_53327_cast_fp16))[name = tensor<string, []>("op_53352_cast_fp16")];
            tensor<string, []> var_53354_equation_0 = const()[name = tensor<string, []>("op_53354_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53354_cast_fp16 = einsum(equation = var_53354_equation_0, values = (var_53184_cast_fp16, var_53328_cast_fp16))[name = tensor<string, []>("op_53354_cast_fp16")];
            tensor<string, []> var_53356_equation_0 = const()[name = tensor<string, []>("op_53356_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53356_cast_fp16 = einsum(equation = var_53356_equation_0, values = (var_53188_cast_fp16, var_53329_cast_fp16))[name = tensor<string, []>("op_53356_cast_fp16")];
            tensor<string, []> var_53358_equation_0 = const()[name = tensor<string, []>("op_53358_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53358_cast_fp16 = einsum(equation = var_53358_equation_0, values = (var_53192_cast_fp16, var_53330_cast_fp16))[name = tensor<string, []>("op_53358_cast_fp16")];
            tensor<string, []> var_53360_equation_0 = const()[name = tensor<string, []>("op_53360_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53360_cast_fp16 = einsum(equation = var_53360_equation_0, values = (var_53196_cast_fp16, var_53331_cast_fp16))[name = tensor<string, []>("op_53360_cast_fp16")];
            tensor<string, []> var_53362_equation_0 = const()[name = tensor<string, []>("op_53362_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53362_cast_fp16 = einsum(equation = var_53362_equation_0, values = (var_53200_cast_fp16, var_53332_cast_fp16))[name = tensor<string, []>("op_53362_cast_fp16")];
            tensor<string, []> var_53364_equation_0 = const()[name = tensor<string, []>("op_53364_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53364_cast_fp16 = einsum(equation = var_53364_equation_0, values = (var_53204_cast_fp16, var_53333_cast_fp16))[name = tensor<string, []>("op_53364_cast_fp16")];
            tensor<string, []> var_53366_equation_0 = const()[name = tensor<string, []>("op_53366_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53366_cast_fp16 = einsum(equation = var_53366_equation_0, values = (var_53208_cast_fp16, var_53334_cast_fp16))[name = tensor<string, []>("op_53366_cast_fp16")];
            tensor<string, []> var_53368_equation_0 = const()[name = tensor<string, []>("op_53368_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53368_cast_fp16 = einsum(equation = var_53368_equation_0, values = (var_53212_cast_fp16, var_53335_cast_fp16))[name = tensor<string, []>("op_53368_cast_fp16")];
            tensor<string, []> var_53370_equation_0 = const()[name = tensor<string, []>("op_53370_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53370_cast_fp16 = einsum(equation = var_53370_equation_0, values = (var_53216_cast_fp16, var_53336_cast_fp16))[name = tensor<string, []>("op_53370_cast_fp16")];
            tensor<string, []> var_53372_equation_0 = const()[name = tensor<string, []>("op_53372_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53372_cast_fp16 = einsum(equation = var_53372_equation_0, values = (var_53220_cast_fp16, var_53337_cast_fp16))[name = tensor<string, []>("op_53372_cast_fp16")];
            tensor<string, []> var_53374_equation_0 = const()[name = tensor<string, []>("op_53374_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53374_cast_fp16 = einsum(equation = var_53374_equation_0, values = (var_53224_cast_fp16, var_53338_cast_fp16))[name = tensor<string, []>("op_53374_cast_fp16")];
            tensor<string, []> var_53376_equation_0 = const()[name = tensor<string, []>("op_53376_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53376_cast_fp16 = einsum(equation = var_53376_equation_0, values = (var_53228_cast_fp16, var_53339_cast_fp16))[name = tensor<string, []>("op_53376_cast_fp16")];
            tensor<string, []> var_53378_equation_0 = const()[name = tensor<string, []>("op_53378_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53378_cast_fp16 = einsum(equation = var_53378_equation_0, values = (var_53232_cast_fp16, var_53340_cast_fp16))[name = tensor<string, []>("op_53378_cast_fp16")];
            tensor<string, []> var_53380_equation_0 = const()[name = tensor<string, []>("op_53380_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53380_cast_fp16 = einsum(equation = var_53380_equation_0, values = (var_53236_cast_fp16, var_53341_cast_fp16))[name = tensor<string, []>("op_53380_cast_fp16")];
            tensor<string, []> var_53382_equation_0 = const()[name = tensor<string, []>("op_53382_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53382_cast_fp16 = einsum(equation = var_53382_equation_0, values = (var_53240_cast_fp16, var_53342_cast_fp16))[name = tensor<string, []>("op_53382_cast_fp16")];
            tensor<bool, []> input_669_interleave_0 = const()[name = tensor<string, []>("input_669_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_669_cast_fp16 = concat(axis = var_30355, interleave = input_669_interleave_0, values = (var_53344_cast_fp16, var_53346_cast_fp16, var_53348_cast_fp16, var_53350_cast_fp16, var_53352_cast_fp16, var_53354_cast_fp16, var_53356_cast_fp16, var_53358_cast_fp16, var_53360_cast_fp16, var_53362_cast_fp16, var_53364_cast_fp16, var_53366_cast_fp16, var_53368_cast_fp16, var_53370_cast_fp16, var_53372_cast_fp16, var_53374_cast_fp16, var_53376_cast_fp16, var_53378_cast_fp16, var_53380_cast_fp16, var_53382_cast_fp16))[name = tensor<string, []>("input_669_cast_fp16")];
            tensor<string, []> var_53392_pad_type_0 = const()[name = tensor<string, []>("op_53392_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_53392_strides_0 = const()[name = tensor<string, []>("op_53392_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_53392_pad_0 = const()[name = tensor<string, []>("op_53392_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_53392_dilations_0 = const()[name = tensor<string, []>("op_53392_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_53392_groups_0 = const()[name = tensor<string, []>("op_53392_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_4_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(712723712))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(713952576))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_4_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_4_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_4_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(713952768)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_53392_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_4_attn2_to_out_0_bias_to_fp16, dilations = var_53392_dilations_0, groups = var_53392_groups_0, pad = var_53392_pad_0, pad_type = var_53392_pad_type_0, strides = var_53392_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_4_attn2_to_out_0_weight_to_fp16_palettized, x = input_669_cast_fp16)[name = tensor<string, []>("op_53392_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_353_cast_fp16 = add(x = var_53392_cast_fp16, y = inputs_351_cast_fp16)[name = tensor<string, []>("inputs_353_cast_fp16")];
            tensor<int32, [1]> input_671_axes_0 = const()[name = tensor<string, []>("input_671_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> input_671_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_671_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(713955392)))];
            tensor<fp16, [1280]> input_671_beta_0_to_fp16 = const()[name = tensor<string, []>("input_671_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(713958016)))];
            tensor<fp16, []> var_53402_to_fp16 = const()[name = tensor<string, []>("op_53402_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> input_671_cast_fp16 = layer_norm(axes = input_671_axes_0, beta = input_671_beta_0_to_fp16, epsilon = var_53402_to_fp16, gamma = input_671_gamma_0_to_fp16, x = inputs_353_cast_fp16)[name = tensor<string, []>("input_671_cast_fp16")];
            tensor<string, []> var_53422_pad_type_0 = const()[name = tensor<string, []>("op_53422_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_53422_strides_0 = const()[name = tensor<string, []>("op_53422_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_53422_pad_0 = const()[name = tensor<string, []>("op_53422_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_53422_dilations_0 = const()[name = tensor<string, []>("op_53422_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_53422_groups_0 = const()[name = tensor<string, []>("op_53422_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_4_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [9830400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(713960640))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(723791104))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_4_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([10240, 1280, 1, 1])];
            tensor<fp16, [10240]> up_blocks_0_attentions_2_transformer_blocks_4_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_4_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(723791296)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_53422_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_4_ff_net_0_proj_bias_to_fp16, dilations = var_53422_dilations_0, groups = var_53422_groups_0, pad = var_53422_pad_0, pad_type = var_53422_pad_type_0, strides = var_53422_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_4_ff_net_0_proj_weight_to_fp16_palettized, x = input_671_cast_fp16)[name = tensor<string, []>("op_53422_cast_fp16")];
            tensor<int32, [2]> var_53423_split_sizes_0 = const()[name = tensor<string, []>("op_53423_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_53423_axis_0 = const()[name = tensor<string, []>("op_53423_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_53423_cast_fp16_0, tensor<fp16, [2, 5120, 1, 1024]> var_53423_cast_fp16_1 = split(axis = var_53423_axis_0, split_sizes = var_53423_split_sizes_0, x = var_53422_cast_fp16)[name = tensor<string, []>("op_53423_cast_fp16")];
            tensor<string, []> var_53425_mode_0 = const()[name = tensor<string, []>("op_53425_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_53425_cast_fp16 = gelu(mode = var_53425_mode_0, x = var_53423_cast_fp16_1)[name = tensor<string, []>("op_53425_cast_fp16")];
            tensor<fp16, [2, 5120, 1, 1024]> input_673_cast_fp16 = mul(x = var_53423_cast_fp16_0, y = var_53425_cast_fp16)[name = tensor<string, []>("input_673_cast_fp16")];
            tensor<string, []> var_53433_pad_type_0 = const()[name = tensor<string, []>("op_53433_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_53433_strides_0 = const()[name = tensor<string, []>("op_53433_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_53433_pad_0 = const()[name = tensor<string, []>("op_53433_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_53433_dilations_0 = const()[name = tensor<string, []>("op_53433_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_53433_groups_0 = const()[name = tensor<string, []>("op_53433_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_4_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [4915200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(723811840))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(728727104))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_4_ff_net_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 5120, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_4_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_4_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(728727296)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_53433_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_4_ff_net_2_bias_to_fp16, dilations = var_53433_dilations_0, groups = var_53433_groups_0, pad = var_53433_pad_0, pad_type = var_53433_pad_type_0, strides = var_53433_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_4_ff_net_2_weight_to_fp16_palettized, x = input_673_cast_fp16)[name = tensor<string, []>("op_53433_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_355_cast_fp16 = add(x = var_53433_cast_fp16, y = inputs_353_cast_fp16)[name = tensor<string, []>("inputs_355_cast_fp16")];
            tensor<int32, [1]> hidden_states_465_axes_0 = const()[name = tensor<string, []>("hidden_states_465_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_465_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_465_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(728729920)))];
            tensor<fp16, [1280]> hidden_states_465_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_465_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(728732544)))];
            tensor<fp16, []> var_53449_to_fp16 = const()[name = tensor<string, []>("op_53449_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_465_cast_fp16 = layer_norm(axes = hidden_states_465_axes_0, beta = hidden_states_465_beta_0_to_fp16, epsilon = var_53449_to_fp16, gamma = hidden_states_465_gamma_0_to_fp16, x = inputs_355_cast_fp16)[name = tensor<string, []>("hidden_states_465_cast_fp16")];
            tensor<string, []> q_237_pad_type_0 = const()[name = tensor<string, []>("q_237_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_237_strides_0 = const()[name = tensor<string, []>("q_237_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_237_pad_0 = const()[name = tensor<string, []>("q_237_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_237_dilations_0 = const()[name = tensor<string, []>("q_237_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_237_groups_0 = const()[name = tensor<string, []>("q_237_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_5_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(728735168))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(729964032))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_5_attn1_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_237_cast_fp16 = conv(dilations = q_237_dilations_0, groups = q_237_groups_0, pad = q_237_pad_0, pad_type = q_237_pad_type_0, strides = q_237_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_5_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_465_cast_fp16)[name = tensor<string, []>("q_237_cast_fp16")];
            tensor<string, []> k_473_pad_type_0 = const()[name = tensor<string, []>("k_473_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_473_strides_0 = const()[name = tensor<string, []>("k_473_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_473_pad_0 = const()[name = tensor<string, []>("k_473_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_473_dilations_0 = const()[name = tensor<string, []>("k_473_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_473_groups_0 = const()[name = tensor<string, []>("k_473_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_5_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(729964224))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(731193088))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_5_attn1_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> k_473_cast_fp16 = conv(dilations = k_473_dilations_0, groups = k_473_groups_0, pad = k_473_pad_0, pad_type = k_473_pad_type_0, strides = k_473_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_5_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_465_cast_fp16)[name = tensor<string, []>("k_473_cast_fp16")];
            tensor<string, []> v_237_pad_type_0 = const()[name = tensor<string, []>("v_237_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_237_strides_0 = const()[name = tensor<string, []>("v_237_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_237_pad_0 = const()[name = tensor<string, []>("v_237_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_237_dilations_0 = const()[name = tensor<string, []>("v_237_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_237_groups_0 = const()[name = tensor<string, []>("v_237_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_5_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(731193280))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(732422144))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_5_attn1_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> v_237_cast_fp16 = conv(dilations = v_237_dilations_0, groups = v_237_groups_0, pad = v_237_pad_0, pad_type = v_237_pad_type_0, strides = v_237_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_5_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_465_cast_fp16)[name = tensor<string, []>("v_237_cast_fp16")];
            tensor<int32, [4]> var_53482_begin_0 = const()[name = tensor<string, []>("op_53482_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_53482_end_0 = const()[name = tensor<string, []>("op_53482_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_53482_end_mask_0 = const()[name = tensor<string, []>("op_53482_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53482_cast_fp16 = slice_by_index(begin = var_53482_begin_0, end = var_53482_end_0, end_mask = var_53482_end_mask_0, x = q_237_cast_fp16)[name = tensor<string, []>("op_53482_cast_fp16")];
            tensor<int32, [4]> var_53486_begin_0 = const()[name = tensor<string, []>("op_53486_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_53486_end_0 = const()[name = tensor<string, []>("op_53486_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_53486_end_mask_0 = const()[name = tensor<string, []>("op_53486_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53486_cast_fp16 = slice_by_index(begin = var_53486_begin_0, end = var_53486_end_0, end_mask = var_53486_end_mask_0, x = q_237_cast_fp16)[name = tensor<string, []>("op_53486_cast_fp16")];
            tensor<int32, [4]> var_53490_begin_0 = const()[name = tensor<string, []>("op_53490_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_53490_end_0 = const()[name = tensor<string, []>("op_53490_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_53490_end_mask_0 = const()[name = tensor<string, []>("op_53490_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53490_cast_fp16 = slice_by_index(begin = var_53490_begin_0, end = var_53490_end_0, end_mask = var_53490_end_mask_0, x = q_237_cast_fp16)[name = tensor<string, []>("op_53490_cast_fp16")];
            tensor<int32, [4]> var_53494_begin_0 = const()[name = tensor<string, []>("op_53494_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_53494_end_0 = const()[name = tensor<string, []>("op_53494_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_53494_end_mask_0 = const()[name = tensor<string, []>("op_53494_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53494_cast_fp16 = slice_by_index(begin = var_53494_begin_0, end = var_53494_end_0, end_mask = var_53494_end_mask_0, x = q_237_cast_fp16)[name = tensor<string, []>("op_53494_cast_fp16")];
            tensor<int32, [4]> var_53498_begin_0 = const()[name = tensor<string, []>("op_53498_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_53498_end_0 = const()[name = tensor<string, []>("op_53498_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_53498_end_mask_0 = const()[name = tensor<string, []>("op_53498_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53498_cast_fp16 = slice_by_index(begin = var_53498_begin_0, end = var_53498_end_0, end_mask = var_53498_end_mask_0, x = q_237_cast_fp16)[name = tensor<string, []>("op_53498_cast_fp16")];
            tensor<int32, [4]> var_53502_begin_0 = const()[name = tensor<string, []>("op_53502_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_53502_end_0 = const()[name = tensor<string, []>("op_53502_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_53502_end_mask_0 = const()[name = tensor<string, []>("op_53502_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53502_cast_fp16 = slice_by_index(begin = var_53502_begin_0, end = var_53502_end_0, end_mask = var_53502_end_mask_0, x = q_237_cast_fp16)[name = tensor<string, []>("op_53502_cast_fp16")];
            tensor<int32, [4]> var_53506_begin_0 = const()[name = tensor<string, []>("op_53506_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_53506_end_0 = const()[name = tensor<string, []>("op_53506_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_53506_end_mask_0 = const()[name = tensor<string, []>("op_53506_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53506_cast_fp16 = slice_by_index(begin = var_53506_begin_0, end = var_53506_end_0, end_mask = var_53506_end_mask_0, x = q_237_cast_fp16)[name = tensor<string, []>("op_53506_cast_fp16")];
            tensor<int32, [4]> var_53510_begin_0 = const()[name = tensor<string, []>("op_53510_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_53510_end_0 = const()[name = tensor<string, []>("op_53510_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_53510_end_mask_0 = const()[name = tensor<string, []>("op_53510_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53510_cast_fp16 = slice_by_index(begin = var_53510_begin_0, end = var_53510_end_0, end_mask = var_53510_end_mask_0, x = q_237_cast_fp16)[name = tensor<string, []>("op_53510_cast_fp16")];
            tensor<int32, [4]> var_53514_begin_0 = const()[name = tensor<string, []>("op_53514_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_53514_end_0 = const()[name = tensor<string, []>("op_53514_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_53514_end_mask_0 = const()[name = tensor<string, []>("op_53514_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53514_cast_fp16 = slice_by_index(begin = var_53514_begin_0, end = var_53514_end_0, end_mask = var_53514_end_mask_0, x = q_237_cast_fp16)[name = tensor<string, []>("op_53514_cast_fp16")];
            tensor<int32, [4]> var_53518_begin_0 = const()[name = tensor<string, []>("op_53518_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_53518_end_0 = const()[name = tensor<string, []>("op_53518_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_53518_end_mask_0 = const()[name = tensor<string, []>("op_53518_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53518_cast_fp16 = slice_by_index(begin = var_53518_begin_0, end = var_53518_end_0, end_mask = var_53518_end_mask_0, x = q_237_cast_fp16)[name = tensor<string, []>("op_53518_cast_fp16")];
            tensor<int32, [4]> var_53522_begin_0 = const()[name = tensor<string, []>("op_53522_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_53522_end_0 = const()[name = tensor<string, []>("op_53522_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_53522_end_mask_0 = const()[name = tensor<string, []>("op_53522_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53522_cast_fp16 = slice_by_index(begin = var_53522_begin_0, end = var_53522_end_0, end_mask = var_53522_end_mask_0, x = q_237_cast_fp16)[name = tensor<string, []>("op_53522_cast_fp16")];
            tensor<int32, [4]> var_53526_begin_0 = const()[name = tensor<string, []>("op_53526_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_53526_end_0 = const()[name = tensor<string, []>("op_53526_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_53526_end_mask_0 = const()[name = tensor<string, []>("op_53526_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53526_cast_fp16 = slice_by_index(begin = var_53526_begin_0, end = var_53526_end_0, end_mask = var_53526_end_mask_0, x = q_237_cast_fp16)[name = tensor<string, []>("op_53526_cast_fp16")];
            tensor<int32, [4]> var_53530_begin_0 = const()[name = tensor<string, []>("op_53530_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_53530_end_0 = const()[name = tensor<string, []>("op_53530_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_53530_end_mask_0 = const()[name = tensor<string, []>("op_53530_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53530_cast_fp16 = slice_by_index(begin = var_53530_begin_0, end = var_53530_end_0, end_mask = var_53530_end_mask_0, x = q_237_cast_fp16)[name = tensor<string, []>("op_53530_cast_fp16")];
            tensor<int32, [4]> var_53534_begin_0 = const()[name = tensor<string, []>("op_53534_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_53534_end_0 = const()[name = tensor<string, []>("op_53534_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_53534_end_mask_0 = const()[name = tensor<string, []>("op_53534_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53534_cast_fp16 = slice_by_index(begin = var_53534_begin_0, end = var_53534_end_0, end_mask = var_53534_end_mask_0, x = q_237_cast_fp16)[name = tensor<string, []>("op_53534_cast_fp16")];
            tensor<int32, [4]> var_53538_begin_0 = const()[name = tensor<string, []>("op_53538_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_53538_end_0 = const()[name = tensor<string, []>("op_53538_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_53538_end_mask_0 = const()[name = tensor<string, []>("op_53538_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53538_cast_fp16 = slice_by_index(begin = var_53538_begin_0, end = var_53538_end_0, end_mask = var_53538_end_mask_0, x = q_237_cast_fp16)[name = tensor<string, []>("op_53538_cast_fp16")];
            tensor<int32, [4]> var_53542_begin_0 = const()[name = tensor<string, []>("op_53542_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_53542_end_0 = const()[name = tensor<string, []>("op_53542_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_53542_end_mask_0 = const()[name = tensor<string, []>("op_53542_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53542_cast_fp16 = slice_by_index(begin = var_53542_begin_0, end = var_53542_end_0, end_mask = var_53542_end_mask_0, x = q_237_cast_fp16)[name = tensor<string, []>("op_53542_cast_fp16")];
            tensor<int32, [4]> var_53546_begin_0 = const()[name = tensor<string, []>("op_53546_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_53546_end_0 = const()[name = tensor<string, []>("op_53546_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_53546_end_mask_0 = const()[name = tensor<string, []>("op_53546_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53546_cast_fp16 = slice_by_index(begin = var_53546_begin_0, end = var_53546_end_0, end_mask = var_53546_end_mask_0, x = q_237_cast_fp16)[name = tensor<string, []>("op_53546_cast_fp16")];
            tensor<int32, [4]> var_53550_begin_0 = const()[name = tensor<string, []>("op_53550_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_53550_end_0 = const()[name = tensor<string, []>("op_53550_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_53550_end_mask_0 = const()[name = tensor<string, []>("op_53550_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53550_cast_fp16 = slice_by_index(begin = var_53550_begin_0, end = var_53550_end_0, end_mask = var_53550_end_mask_0, x = q_237_cast_fp16)[name = tensor<string, []>("op_53550_cast_fp16")];
            tensor<int32, [4]> var_53554_begin_0 = const()[name = tensor<string, []>("op_53554_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_53554_end_0 = const()[name = tensor<string, []>("op_53554_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_53554_end_mask_0 = const()[name = tensor<string, []>("op_53554_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53554_cast_fp16 = slice_by_index(begin = var_53554_begin_0, end = var_53554_end_0, end_mask = var_53554_end_mask_0, x = q_237_cast_fp16)[name = tensor<string, []>("op_53554_cast_fp16")];
            tensor<int32, [4]> var_53558_begin_0 = const()[name = tensor<string, []>("op_53558_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_53558_end_0 = const()[name = tensor<string, []>("op_53558_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_53558_end_mask_0 = const()[name = tensor<string, []>("op_53558_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53558_cast_fp16 = slice_by_index(begin = var_53558_begin_0, end = var_53558_end_0, end_mask = var_53558_end_mask_0, x = q_237_cast_fp16)[name = tensor<string, []>("op_53558_cast_fp16")];
            tensor<int32, [4]> k_475_perm_0 = const()[name = tensor<string, []>("k_475_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_53565_begin_0 = const()[name = tensor<string, []>("op_53565_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_53565_end_0 = const()[name = tensor<string, []>("op_53565_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_53565_end_mask_0 = const()[name = tensor<string, []>("op_53565_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> k_475_cast_fp16 = transpose(perm = k_475_perm_0, x = k_473_cast_fp16)[name = tensor<string, []>("transpose_21")];
            tensor<fp16, [2, 1024, 1, 64]> var_53565_cast_fp16 = slice_by_index(begin = var_53565_begin_0, end = var_53565_end_0, end_mask = var_53565_end_mask_0, x = k_475_cast_fp16)[name = tensor<string, []>("op_53565_cast_fp16")];
            tensor<int32, [4]> var_53569_begin_0 = const()[name = tensor<string, []>("op_53569_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_53569_end_0 = const()[name = tensor<string, []>("op_53569_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_53569_end_mask_0 = const()[name = tensor<string, []>("op_53569_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_53569_cast_fp16 = slice_by_index(begin = var_53569_begin_0, end = var_53569_end_0, end_mask = var_53569_end_mask_0, x = k_475_cast_fp16)[name = tensor<string, []>("op_53569_cast_fp16")];
            tensor<int32, [4]> var_53573_begin_0 = const()[name = tensor<string, []>("op_53573_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_53573_end_0 = const()[name = tensor<string, []>("op_53573_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_53573_end_mask_0 = const()[name = tensor<string, []>("op_53573_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_53573_cast_fp16 = slice_by_index(begin = var_53573_begin_0, end = var_53573_end_0, end_mask = var_53573_end_mask_0, x = k_475_cast_fp16)[name = tensor<string, []>("op_53573_cast_fp16")];
            tensor<int32, [4]> var_53577_begin_0 = const()[name = tensor<string, []>("op_53577_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_53577_end_0 = const()[name = tensor<string, []>("op_53577_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_53577_end_mask_0 = const()[name = tensor<string, []>("op_53577_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_53577_cast_fp16 = slice_by_index(begin = var_53577_begin_0, end = var_53577_end_0, end_mask = var_53577_end_mask_0, x = k_475_cast_fp16)[name = tensor<string, []>("op_53577_cast_fp16")];
            tensor<int32, [4]> var_53581_begin_0 = const()[name = tensor<string, []>("op_53581_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_53581_end_0 = const()[name = tensor<string, []>("op_53581_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_53581_end_mask_0 = const()[name = tensor<string, []>("op_53581_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_53581_cast_fp16 = slice_by_index(begin = var_53581_begin_0, end = var_53581_end_0, end_mask = var_53581_end_mask_0, x = k_475_cast_fp16)[name = tensor<string, []>("op_53581_cast_fp16")];
            tensor<int32, [4]> var_53585_begin_0 = const()[name = tensor<string, []>("op_53585_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_53585_end_0 = const()[name = tensor<string, []>("op_53585_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_53585_end_mask_0 = const()[name = tensor<string, []>("op_53585_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_53585_cast_fp16 = slice_by_index(begin = var_53585_begin_0, end = var_53585_end_0, end_mask = var_53585_end_mask_0, x = k_475_cast_fp16)[name = tensor<string, []>("op_53585_cast_fp16")];
            tensor<int32, [4]> var_53589_begin_0 = const()[name = tensor<string, []>("op_53589_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_53589_end_0 = const()[name = tensor<string, []>("op_53589_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_53589_end_mask_0 = const()[name = tensor<string, []>("op_53589_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_53589_cast_fp16 = slice_by_index(begin = var_53589_begin_0, end = var_53589_end_0, end_mask = var_53589_end_mask_0, x = k_475_cast_fp16)[name = tensor<string, []>("op_53589_cast_fp16")];
            tensor<int32, [4]> var_53593_begin_0 = const()[name = tensor<string, []>("op_53593_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_53593_end_0 = const()[name = tensor<string, []>("op_53593_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_53593_end_mask_0 = const()[name = tensor<string, []>("op_53593_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_53593_cast_fp16 = slice_by_index(begin = var_53593_begin_0, end = var_53593_end_0, end_mask = var_53593_end_mask_0, x = k_475_cast_fp16)[name = tensor<string, []>("op_53593_cast_fp16")];
            tensor<int32, [4]> var_53597_begin_0 = const()[name = tensor<string, []>("op_53597_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_53597_end_0 = const()[name = tensor<string, []>("op_53597_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_53597_end_mask_0 = const()[name = tensor<string, []>("op_53597_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_53597_cast_fp16 = slice_by_index(begin = var_53597_begin_0, end = var_53597_end_0, end_mask = var_53597_end_mask_0, x = k_475_cast_fp16)[name = tensor<string, []>("op_53597_cast_fp16")];
            tensor<int32, [4]> var_53601_begin_0 = const()[name = tensor<string, []>("op_53601_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_53601_end_0 = const()[name = tensor<string, []>("op_53601_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_53601_end_mask_0 = const()[name = tensor<string, []>("op_53601_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_53601_cast_fp16 = slice_by_index(begin = var_53601_begin_0, end = var_53601_end_0, end_mask = var_53601_end_mask_0, x = k_475_cast_fp16)[name = tensor<string, []>("op_53601_cast_fp16")];
            tensor<int32, [4]> var_53605_begin_0 = const()[name = tensor<string, []>("op_53605_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_53605_end_0 = const()[name = tensor<string, []>("op_53605_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_53605_end_mask_0 = const()[name = tensor<string, []>("op_53605_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_53605_cast_fp16 = slice_by_index(begin = var_53605_begin_0, end = var_53605_end_0, end_mask = var_53605_end_mask_0, x = k_475_cast_fp16)[name = tensor<string, []>("op_53605_cast_fp16")];
            tensor<int32, [4]> var_53609_begin_0 = const()[name = tensor<string, []>("op_53609_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_53609_end_0 = const()[name = tensor<string, []>("op_53609_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_53609_end_mask_0 = const()[name = tensor<string, []>("op_53609_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_53609_cast_fp16 = slice_by_index(begin = var_53609_begin_0, end = var_53609_end_0, end_mask = var_53609_end_mask_0, x = k_475_cast_fp16)[name = tensor<string, []>("op_53609_cast_fp16")];
            tensor<int32, [4]> var_53613_begin_0 = const()[name = tensor<string, []>("op_53613_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_53613_end_0 = const()[name = tensor<string, []>("op_53613_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_53613_end_mask_0 = const()[name = tensor<string, []>("op_53613_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_53613_cast_fp16 = slice_by_index(begin = var_53613_begin_0, end = var_53613_end_0, end_mask = var_53613_end_mask_0, x = k_475_cast_fp16)[name = tensor<string, []>("op_53613_cast_fp16")];
            tensor<int32, [4]> var_53617_begin_0 = const()[name = tensor<string, []>("op_53617_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_53617_end_0 = const()[name = tensor<string, []>("op_53617_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_53617_end_mask_0 = const()[name = tensor<string, []>("op_53617_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_53617_cast_fp16 = slice_by_index(begin = var_53617_begin_0, end = var_53617_end_0, end_mask = var_53617_end_mask_0, x = k_475_cast_fp16)[name = tensor<string, []>("op_53617_cast_fp16")];
            tensor<int32, [4]> var_53621_begin_0 = const()[name = tensor<string, []>("op_53621_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_53621_end_0 = const()[name = tensor<string, []>("op_53621_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_53621_end_mask_0 = const()[name = tensor<string, []>("op_53621_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_53621_cast_fp16 = slice_by_index(begin = var_53621_begin_0, end = var_53621_end_0, end_mask = var_53621_end_mask_0, x = k_475_cast_fp16)[name = tensor<string, []>("op_53621_cast_fp16")];
            tensor<int32, [4]> var_53625_begin_0 = const()[name = tensor<string, []>("op_53625_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_53625_end_0 = const()[name = tensor<string, []>("op_53625_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_53625_end_mask_0 = const()[name = tensor<string, []>("op_53625_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_53625_cast_fp16 = slice_by_index(begin = var_53625_begin_0, end = var_53625_end_0, end_mask = var_53625_end_mask_0, x = k_475_cast_fp16)[name = tensor<string, []>("op_53625_cast_fp16")];
            tensor<int32, [4]> var_53629_begin_0 = const()[name = tensor<string, []>("op_53629_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_53629_end_0 = const()[name = tensor<string, []>("op_53629_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_53629_end_mask_0 = const()[name = tensor<string, []>("op_53629_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_53629_cast_fp16 = slice_by_index(begin = var_53629_begin_0, end = var_53629_end_0, end_mask = var_53629_end_mask_0, x = k_475_cast_fp16)[name = tensor<string, []>("op_53629_cast_fp16")];
            tensor<int32, [4]> var_53633_begin_0 = const()[name = tensor<string, []>("op_53633_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_53633_end_0 = const()[name = tensor<string, []>("op_53633_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_53633_end_mask_0 = const()[name = tensor<string, []>("op_53633_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_53633_cast_fp16 = slice_by_index(begin = var_53633_begin_0, end = var_53633_end_0, end_mask = var_53633_end_mask_0, x = k_475_cast_fp16)[name = tensor<string, []>("op_53633_cast_fp16")];
            tensor<int32, [4]> var_53637_begin_0 = const()[name = tensor<string, []>("op_53637_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_53637_end_0 = const()[name = tensor<string, []>("op_53637_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_53637_end_mask_0 = const()[name = tensor<string, []>("op_53637_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_53637_cast_fp16 = slice_by_index(begin = var_53637_begin_0, end = var_53637_end_0, end_mask = var_53637_end_mask_0, x = k_475_cast_fp16)[name = tensor<string, []>("op_53637_cast_fp16")];
            tensor<int32, [4]> var_53641_begin_0 = const()[name = tensor<string, []>("op_53641_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_53641_end_0 = const()[name = tensor<string, []>("op_53641_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_53641_end_mask_0 = const()[name = tensor<string, []>("op_53641_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_53641_cast_fp16 = slice_by_index(begin = var_53641_begin_0, end = var_53641_end_0, end_mask = var_53641_end_mask_0, x = k_475_cast_fp16)[name = tensor<string, []>("op_53641_cast_fp16")];
            tensor<int32, [4]> var_53643_begin_0 = const()[name = tensor<string, []>("op_53643_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_53643_end_0 = const()[name = tensor<string, []>("op_53643_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_53643_end_mask_0 = const()[name = tensor<string, []>("op_53643_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53643_cast_fp16 = slice_by_index(begin = var_53643_begin_0, end = var_53643_end_0, end_mask = var_53643_end_mask_0, x = v_237_cast_fp16)[name = tensor<string, []>("op_53643_cast_fp16")];
            tensor<int32, [4]> var_53647_begin_0 = const()[name = tensor<string, []>("op_53647_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_53647_end_0 = const()[name = tensor<string, []>("op_53647_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_53647_end_mask_0 = const()[name = tensor<string, []>("op_53647_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53647_cast_fp16 = slice_by_index(begin = var_53647_begin_0, end = var_53647_end_0, end_mask = var_53647_end_mask_0, x = v_237_cast_fp16)[name = tensor<string, []>("op_53647_cast_fp16")];
            tensor<int32, [4]> var_53651_begin_0 = const()[name = tensor<string, []>("op_53651_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_53651_end_0 = const()[name = tensor<string, []>("op_53651_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_53651_end_mask_0 = const()[name = tensor<string, []>("op_53651_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53651_cast_fp16 = slice_by_index(begin = var_53651_begin_0, end = var_53651_end_0, end_mask = var_53651_end_mask_0, x = v_237_cast_fp16)[name = tensor<string, []>("op_53651_cast_fp16")];
            tensor<int32, [4]> var_53655_begin_0 = const()[name = tensor<string, []>("op_53655_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_53655_end_0 = const()[name = tensor<string, []>("op_53655_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_53655_end_mask_0 = const()[name = tensor<string, []>("op_53655_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53655_cast_fp16 = slice_by_index(begin = var_53655_begin_0, end = var_53655_end_0, end_mask = var_53655_end_mask_0, x = v_237_cast_fp16)[name = tensor<string, []>("op_53655_cast_fp16")];
            tensor<int32, [4]> var_53659_begin_0 = const()[name = tensor<string, []>("op_53659_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_53659_end_0 = const()[name = tensor<string, []>("op_53659_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_53659_end_mask_0 = const()[name = tensor<string, []>("op_53659_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53659_cast_fp16 = slice_by_index(begin = var_53659_begin_0, end = var_53659_end_0, end_mask = var_53659_end_mask_0, x = v_237_cast_fp16)[name = tensor<string, []>("op_53659_cast_fp16")];
            tensor<int32, [4]> var_53663_begin_0 = const()[name = tensor<string, []>("op_53663_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_53663_end_0 = const()[name = tensor<string, []>("op_53663_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_53663_end_mask_0 = const()[name = tensor<string, []>("op_53663_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53663_cast_fp16 = slice_by_index(begin = var_53663_begin_0, end = var_53663_end_0, end_mask = var_53663_end_mask_0, x = v_237_cast_fp16)[name = tensor<string, []>("op_53663_cast_fp16")];
            tensor<int32, [4]> var_53667_begin_0 = const()[name = tensor<string, []>("op_53667_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_53667_end_0 = const()[name = tensor<string, []>("op_53667_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_53667_end_mask_0 = const()[name = tensor<string, []>("op_53667_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53667_cast_fp16 = slice_by_index(begin = var_53667_begin_0, end = var_53667_end_0, end_mask = var_53667_end_mask_0, x = v_237_cast_fp16)[name = tensor<string, []>("op_53667_cast_fp16")];
            tensor<int32, [4]> var_53671_begin_0 = const()[name = tensor<string, []>("op_53671_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_53671_end_0 = const()[name = tensor<string, []>("op_53671_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_53671_end_mask_0 = const()[name = tensor<string, []>("op_53671_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53671_cast_fp16 = slice_by_index(begin = var_53671_begin_0, end = var_53671_end_0, end_mask = var_53671_end_mask_0, x = v_237_cast_fp16)[name = tensor<string, []>("op_53671_cast_fp16")];
            tensor<int32, [4]> var_53675_begin_0 = const()[name = tensor<string, []>("op_53675_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_53675_end_0 = const()[name = tensor<string, []>("op_53675_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_53675_end_mask_0 = const()[name = tensor<string, []>("op_53675_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53675_cast_fp16 = slice_by_index(begin = var_53675_begin_0, end = var_53675_end_0, end_mask = var_53675_end_mask_0, x = v_237_cast_fp16)[name = tensor<string, []>("op_53675_cast_fp16")];
            tensor<int32, [4]> var_53679_begin_0 = const()[name = tensor<string, []>("op_53679_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_53679_end_0 = const()[name = tensor<string, []>("op_53679_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_53679_end_mask_0 = const()[name = tensor<string, []>("op_53679_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53679_cast_fp16 = slice_by_index(begin = var_53679_begin_0, end = var_53679_end_0, end_mask = var_53679_end_mask_0, x = v_237_cast_fp16)[name = tensor<string, []>("op_53679_cast_fp16")];
            tensor<int32, [4]> var_53683_begin_0 = const()[name = tensor<string, []>("op_53683_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_53683_end_0 = const()[name = tensor<string, []>("op_53683_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_53683_end_mask_0 = const()[name = tensor<string, []>("op_53683_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53683_cast_fp16 = slice_by_index(begin = var_53683_begin_0, end = var_53683_end_0, end_mask = var_53683_end_mask_0, x = v_237_cast_fp16)[name = tensor<string, []>("op_53683_cast_fp16")];
            tensor<int32, [4]> var_53687_begin_0 = const()[name = tensor<string, []>("op_53687_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_53687_end_0 = const()[name = tensor<string, []>("op_53687_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_53687_end_mask_0 = const()[name = tensor<string, []>("op_53687_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53687_cast_fp16 = slice_by_index(begin = var_53687_begin_0, end = var_53687_end_0, end_mask = var_53687_end_mask_0, x = v_237_cast_fp16)[name = tensor<string, []>("op_53687_cast_fp16")];
            tensor<int32, [4]> var_53691_begin_0 = const()[name = tensor<string, []>("op_53691_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_53691_end_0 = const()[name = tensor<string, []>("op_53691_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_53691_end_mask_0 = const()[name = tensor<string, []>("op_53691_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53691_cast_fp16 = slice_by_index(begin = var_53691_begin_0, end = var_53691_end_0, end_mask = var_53691_end_mask_0, x = v_237_cast_fp16)[name = tensor<string, []>("op_53691_cast_fp16")];
            tensor<int32, [4]> var_53695_begin_0 = const()[name = tensor<string, []>("op_53695_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_53695_end_0 = const()[name = tensor<string, []>("op_53695_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_53695_end_mask_0 = const()[name = tensor<string, []>("op_53695_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53695_cast_fp16 = slice_by_index(begin = var_53695_begin_0, end = var_53695_end_0, end_mask = var_53695_end_mask_0, x = v_237_cast_fp16)[name = tensor<string, []>("op_53695_cast_fp16")];
            tensor<int32, [4]> var_53699_begin_0 = const()[name = tensor<string, []>("op_53699_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_53699_end_0 = const()[name = tensor<string, []>("op_53699_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_53699_end_mask_0 = const()[name = tensor<string, []>("op_53699_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53699_cast_fp16 = slice_by_index(begin = var_53699_begin_0, end = var_53699_end_0, end_mask = var_53699_end_mask_0, x = v_237_cast_fp16)[name = tensor<string, []>("op_53699_cast_fp16")];
            tensor<int32, [4]> var_53703_begin_0 = const()[name = tensor<string, []>("op_53703_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_53703_end_0 = const()[name = tensor<string, []>("op_53703_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_53703_end_mask_0 = const()[name = tensor<string, []>("op_53703_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53703_cast_fp16 = slice_by_index(begin = var_53703_begin_0, end = var_53703_end_0, end_mask = var_53703_end_mask_0, x = v_237_cast_fp16)[name = tensor<string, []>("op_53703_cast_fp16")];
            tensor<int32, [4]> var_53707_begin_0 = const()[name = tensor<string, []>("op_53707_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_53707_end_0 = const()[name = tensor<string, []>("op_53707_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_53707_end_mask_0 = const()[name = tensor<string, []>("op_53707_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53707_cast_fp16 = slice_by_index(begin = var_53707_begin_0, end = var_53707_end_0, end_mask = var_53707_end_mask_0, x = v_237_cast_fp16)[name = tensor<string, []>("op_53707_cast_fp16")];
            tensor<int32, [4]> var_53711_begin_0 = const()[name = tensor<string, []>("op_53711_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_53711_end_0 = const()[name = tensor<string, []>("op_53711_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_53711_end_mask_0 = const()[name = tensor<string, []>("op_53711_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53711_cast_fp16 = slice_by_index(begin = var_53711_begin_0, end = var_53711_end_0, end_mask = var_53711_end_mask_0, x = v_237_cast_fp16)[name = tensor<string, []>("op_53711_cast_fp16")];
            tensor<int32, [4]> var_53715_begin_0 = const()[name = tensor<string, []>("op_53715_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_53715_end_0 = const()[name = tensor<string, []>("op_53715_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_53715_end_mask_0 = const()[name = tensor<string, []>("op_53715_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53715_cast_fp16 = slice_by_index(begin = var_53715_begin_0, end = var_53715_end_0, end_mask = var_53715_end_mask_0, x = v_237_cast_fp16)[name = tensor<string, []>("op_53715_cast_fp16")];
            tensor<int32, [4]> var_53719_begin_0 = const()[name = tensor<string, []>("op_53719_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_53719_end_0 = const()[name = tensor<string, []>("op_53719_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_53719_end_mask_0 = const()[name = tensor<string, []>("op_53719_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53719_cast_fp16 = slice_by_index(begin = var_53719_begin_0, end = var_53719_end_0, end_mask = var_53719_end_mask_0, x = v_237_cast_fp16)[name = tensor<string, []>("op_53719_cast_fp16")];
            tensor<string, []> var_53723_equation_0 = const()[name = tensor<string, []>("op_53723_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53723_cast_fp16 = einsum(equation = var_53723_equation_0, values = (var_53565_cast_fp16, var_53482_cast_fp16))[name = tensor<string, []>("op_53723_cast_fp16")];
            tensor<fp16, []> var_53724_to_fp16 = const()[name = tensor<string, []>("op_53724_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4561_cast_fp16 = mul(x = var_53723_cast_fp16, y = var_53724_to_fp16)[name = tensor<string, []>("aw_4561_cast_fp16")];
            tensor<string, []> var_53727_equation_0 = const()[name = tensor<string, []>("op_53727_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53727_cast_fp16 = einsum(equation = var_53727_equation_0, values = (var_53569_cast_fp16, var_53486_cast_fp16))[name = tensor<string, []>("op_53727_cast_fp16")];
            tensor<fp16, []> var_53728_to_fp16 = const()[name = tensor<string, []>("op_53728_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4563_cast_fp16 = mul(x = var_53727_cast_fp16, y = var_53728_to_fp16)[name = tensor<string, []>("aw_4563_cast_fp16")];
            tensor<string, []> var_53731_equation_0 = const()[name = tensor<string, []>("op_53731_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53731_cast_fp16 = einsum(equation = var_53731_equation_0, values = (var_53573_cast_fp16, var_53490_cast_fp16))[name = tensor<string, []>("op_53731_cast_fp16")];
            tensor<fp16, []> var_53732_to_fp16 = const()[name = tensor<string, []>("op_53732_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4565_cast_fp16 = mul(x = var_53731_cast_fp16, y = var_53732_to_fp16)[name = tensor<string, []>("aw_4565_cast_fp16")];
            tensor<string, []> var_53735_equation_0 = const()[name = tensor<string, []>("op_53735_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53735_cast_fp16 = einsum(equation = var_53735_equation_0, values = (var_53577_cast_fp16, var_53494_cast_fp16))[name = tensor<string, []>("op_53735_cast_fp16")];
            tensor<fp16, []> var_53736_to_fp16 = const()[name = tensor<string, []>("op_53736_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4567_cast_fp16 = mul(x = var_53735_cast_fp16, y = var_53736_to_fp16)[name = tensor<string, []>("aw_4567_cast_fp16")];
            tensor<string, []> var_53739_equation_0 = const()[name = tensor<string, []>("op_53739_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53739_cast_fp16 = einsum(equation = var_53739_equation_0, values = (var_53581_cast_fp16, var_53498_cast_fp16))[name = tensor<string, []>("op_53739_cast_fp16")];
            tensor<fp16, []> var_53740_to_fp16 = const()[name = tensor<string, []>("op_53740_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4569_cast_fp16 = mul(x = var_53739_cast_fp16, y = var_53740_to_fp16)[name = tensor<string, []>("aw_4569_cast_fp16")];
            tensor<string, []> var_53743_equation_0 = const()[name = tensor<string, []>("op_53743_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53743_cast_fp16 = einsum(equation = var_53743_equation_0, values = (var_53585_cast_fp16, var_53502_cast_fp16))[name = tensor<string, []>("op_53743_cast_fp16")];
            tensor<fp16, []> var_53744_to_fp16 = const()[name = tensor<string, []>("op_53744_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4571_cast_fp16 = mul(x = var_53743_cast_fp16, y = var_53744_to_fp16)[name = tensor<string, []>("aw_4571_cast_fp16")];
            tensor<string, []> var_53747_equation_0 = const()[name = tensor<string, []>("op_53747_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53747_cast_fp16 = einsum(equation = var_53747_equation_0, values = (var_53589_cast_fp16, var_53506_cast_fp16))[name = tensor<string, []>("op_53747_cast_fp16")];
            tensor<fp16, []> var_53748_to_fp16 = const()[name = tensor<string, []>("op_53748_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4573_cast_fp16 = mul(x = var_53747_cast_fp16, y = var_53748_to_fp16)[name = tensor<string, []>("aw_4573_cast_fp16")];
            tensor<string, []> var_53751_equation_0 = const()[name = tensor<string, []>("op_53751_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53751_cast_fp16 = einsum(equation = var_53751_equation_0, values = (var_53593_cast_fp16, var_53510_cast_fp16))[name = tensor<string, []>("op_53751_cast_fp16")];
            tensor<fp16, []> var_53752_to_fp16 = const()[name = tensor<string, []>("op_53752_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4575_cast_fp16 = mul(x = var_53751_cast_fp16, y = var_53752_to_fp16)[name = tensor<string, []>("aw_4575_cast_fp16")];
            tensor<string, []> var_53755_equation_0 = const()[name = tensor<string, []>("op_53755_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53755_cast_fp16 = einsum(equation = var_53755_equation_0, values = (var_53597_cast_fp16, var_53514_cast_fp16))[name = tensor<string, []>("op_53755_cast_fp16")];
            tensor<fp16, []> var_53756_to_fp16 = const()[name = tensor<string, []>("op_53756_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4577_cast_fp16 = mul(x = var_53755_cast_fp16, y = var_53756_to_fp16)[name = tensor<string, []>("aw_4577_cast_fp16")];
            tensor<string, []> var_53759_equation_0 = const()[name = tensor<string, []>("op_53759_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53759_cast_fp16 = einsum(equation = var_53759_equation_0, values = (var_53601_cast_fp16, var_53518_cast_fp16))[name = tensor<string, []>("op_53759_cast_fp16")];
            tensor<fp16, []> var_53760_to_fp16 = const()[name = tensor<string, []>("op_53760_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4579_cast_fp16 = mul(x = var_53759_cast_fp16, y = var_53760_to_fp16)[name = tensor<string, []>("aw_4579_cast_fp16")];
            tensor<string, []> var_53763_equation_0 = const()[name = tensor<string, []>("op_53763_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53763_cast_fp16 = einsum(equation = var_53763_equation_0, values = (var_53605_cast_fp16, var_53522_cast_fp16))[name = tensor<string, []>("op_53763_cast_fp16")];
            tensor<fp16, []> var_53764_to_fp16 = const()[name = tensor<string, []>("op_53764_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4581_cast_fp16 = mul(x = var_53763_cast_fp16, y = var_53764_to_fp16)[name = tensor<string, []>("aw_4581_cast_fp16")];
            tensor<string, []> var_53767_equation_0 = const()[name = tensor<string, []>("op_53767_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53767_cast_fp16 = einsum(equation = var_53767_equation_0, values = (var_53609_cast_fp16, var_53526_cast_fp16))[name = tensor<string, []>("op_53767_cast_fp16")];
            tensor<fp16, []> var_53768_to_fp16 = const()[name = tensor<string, []>("op_53768_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4583_cast_fp16 = mul(x = var_53767_cast_fp16, y = var_53768_to_fp16)[name = tensor<string, []>("aw_4583_cast_fp16")];
            tensor<string, []> var_53771_equation_0 = const()[name = tensor<string, []>("op_53771_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53771_cast_fp16 = einsum(equation = var_53771_equation_0, values = (var_53613_cast_fp16, var_53530_cast_fp16))[name = tensor<string, []>("op_53771_cast_fp16")];
            tensor<fp16, []> var_53772_to_fp16 = const()[name = tensor<string, []>("op_53772_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4585_cast_fp16 = mul(x = var_53771_cast_fp16, y = var_53772_to_fp16)[name = tensor<string, []>("aw_4585_cast_fp16")];
            tensor<string, []> var_53775_equation_0 = const()[name = tensor<string, []>("op_53775_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53775_cast_fp16 = einsum(equation = var_53775_equation_0, values = (var_53617_cast_fp16, var_53534_cast_fp16))[name = tensor<string, []>("op_53775_cast_fp16")];
            tensor<fp16, []> var_53776_to_fp16 = const()[name = tensor<string, []>("op_53776_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4587_cast_fp16 = mul(x = var_53775_cast_fp16, y = var_53776_to_fp16)[name = tensor<string, []>("aw_4587_cast_fp16")];
            tensor<string, []> var_53779_equation_0 = const()[name = tensor<string, []>("op_53779_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53779_cast_fp16 = einsum(equation = var_53779_equation_0, values = (var_53621_cast_fp16, var_53538_cast_fp16))[name = tensor<string, []>("op_53779_cast_fp16")];
            tensor<fp16, []> var_53780_to_fp16 = const()[name = tensor<string, []>("op_53780_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4589_cast_fp16 = mul(x = var_53779_cast_fp16, y = var_53780_to_fp16)[name = tensor<string, []>("aw_4589_cast_fp16")];
            tensor<string, []> var_53783_equation_0 = const()[name = tensor<string, []>("op_53783_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53783_cast_fp16 = einsum(equation = var_53783_equation_0, values = (var_53625_cast_fp16, var_53542_cast_fp16))[name = tensor<string, []>("op_53783_cast_fp16")];
            tensor<fp16, []> var_53784_to_fp16 = const()[name = tensor<string, []>("op_53784_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4591_cast_fp16 = mul(x = var_53783_cast_fp16, y = var_53784_to_fp16)[name = tensor<string, []>("aw_4591_cast_fp16")];
            tensor<string, []> var_53787_equation_0 = const()[name = tensor<string, []>("op_53787_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53787_cast_fp16 = einsum(equation = var_53787_equation_0, values = (var_53629_cast_fp16, var_53546_cast_fp16))[name = tensor<string, []>("op_53787_cast_fp16")];
            tensor<fp16, []> var_53788_to_fp16 = const()[name = tensor<string, []>("op_53788_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4593_cast_fp16 = mul(x = var_53787_cast_fp16, y = var_53788_to_fp16)[name = tensor<string, []>("aw_4593_cast_fp16")];
            tensor<string, []> var_53791_equation_0 = const()[name = tensor<string, []>("op_53791_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53791_cast_fp16 = einsum(equation = var_53791_equation_0, values = (var_53633_cast_fp16, var_53550_cast_fp16))[name = tensor<string, []>("op_53791_cast_fp16")];
            tensor<fp16, []> var_53792_to_fp16 = const()[name = tensor<string, []>("op_53792_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4595_cast_fp16 = mul(x = var_53791_cast_fp16, y = var_53792_to_fp16)[name = tensor<string, []>("aw_4595_cast_fp16")];
            tensor<string, []> var_53795_equation_0 = const()[name = tensor<string, []>("op_53795_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53795_cast_fp16 = einsum(equation = var_53795_equation_0, values = (var_53637_cast_fp16, var_53554_cast_fp16))[name = tensor<string, []>("op_53795_cast_fp16")];
            tensor<fp16, []> var_53796_to_fp16 = const()[name = tensor<string, []>("op_53796_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4597_cast_fp16 = mul(x = var_53795_cast_fp16, y = var_53796_to_fp16)[name = tensor<string, []>("aw_4597_cast_fp16")];
            tensor<string, []> var_53799_equation_0 = const()[name = tensor<string, []>("op_53799_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53799_cast_fp16 = einsum(equation = var_53799_equation_0, values = (var_53641_cast_fp16, var_53558_cast_fp16))[name = tensor<string, []>("op_53799_cast_fp16")];
            tensor<fp16, []> var_53800_to_fp16 = const()[name = tensor<string, []>("op_53800_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4599_cast_fp16 = mul(x = var_53799_cast_fp16, y = var_53800_to_fp16)[name = tensor<string, []>("aw_4599_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53802_cast_fp16 = softmax(axis = var_30355, x = aw_4561_cast_fp16)[name = tensor<string, []>("op_53802_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53803_cast_fp16 = softmax(axis = var_30355, x = aw_4563_cast_fp16)[name = tensor<string, []>("op_53803_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53804_cast_fp16 = softmax(axis = var_30355, x = aw_4565_cast_fp16)[name = tensor<string, []>("op_53804_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53805_cast_fp16 = softmax(axis = var_30355, x = aw_4567_cast_fp16)[name = tensor<string, []>("op_53805_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53806_cast_fp16 = softmax(axis = var_30355, x = aw_4569_cast_fp16)[name = tensor<string, []>("op_53806_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53807_cast_fp16 = softmax(axis = var_30355, x = aw_4571_cast_fp16)[name = tensor<string, []>("op_53807_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53808_cast_fp16 = softmax(axis = var_30355, x = aw_4573_cast_fp16)[name = tensor<string, []>("op_53808_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53809_cast_fp16 = softmax(axis = var_30355, x = aw_4575_cast_fp16)[name = tensor<string, []>("op_53809_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53810_cast_fp16 = softmax(axis = var_30355, x = aw_4577_cast_fp16)[name = tensor<string, []>("op_53810_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53811_cast_fp16 = softmax(axis = var_30355, x = aw_4579_cast_fp16)[name = tensor<string, []>("op_53811_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53812_cast_fp16 = softmax(axis = var_30355, x = aw_4581_cast_fp16)[name = tensor<string, []>("op_53812_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53813_cast_fp16 = softmax(axis = var_30355, x = aw_4583_cast_fp16)[name = tensor<string, []>("op_53813_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53814_cast_fp16 = softmax(axis = var_30355, x = aw_4585_cast_fp16)[name = tensor<string, []>("op_53814_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53815_cast_fp16 = softmax(axis = var_30355, x = aw_4587_cast_fp16)[name = tensor<string, []>("op_53815_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53816_cast_fp16 = softmax(axis = var_30355, x = aw_4589_cast_fp16)[name = tensor<string, []>("op_53816_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53817_cast_fp16 = softmax(axis = var_30355, x = aw_4591_cast_fp16)[name = tensor<string, []>("op_53817_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53818_cast_fp16 = softmax(axis = var_30355, x = aw_4593_cast_fp16)[name = tensor<string, []>("op_53818_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53819_cast_fp16 = softmax(axis = var_30355, x = aw_4595_cast_fp16)[name = tensor<string, []>("op_53819_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53820_cast_fp16 = softmax(axis = var_30355, x = aw_4597_cast_fp16)[name = tensor<string, []>("op_53820_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53821_cast_fp16 = softmax(axis = var_30355, x = aw_4599_cast_fp16)[name = tensor<string, []>("op_53821_cast_fp16")];
            tensor<string, []> var_53823_equation_0 = const()[name = tensor<string, []>("op_53823_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53823_cast_fp16 = einsum(equation = var_53823_equation_0, values = (var_53643_cast_fp16, var_53802_cast_fp16))[name = tensor<string, []>("op_53823_cast_fp16")];
            tensor<string, []> var_53825_equation_0 = const()[name = tensor<string, []>("op_53825_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53825_cast_fp16 = einsum(equation = var_53825_equation_0, values = (var_53647_cast_fp16, var_53803_cast_fp16))[name = tensor<string, []>("op_53825_cast_fp16")];
            tensor<string, []> var_53827_equation_0 = const()[name = tensor<string, []>("op_53827_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53827_cast_fp16 = einsum(equation = var_53827_equation_0, values = (var_53651_cast_fp16, var_53804_cast_fp16))[name = tensor<string, []>("op_53827_cast_fp16")];
            tensor<string, []> var_53829_equation_0 = const()[name = tensor<string, []>("op_53829_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53829_cast_fp16 = einsum(equation = var_53829_equation_0, values = (var_53655_cast_fp16, var_53805_cast_fp16))[name = tensor<string, []>("op_53829_cast_fp16")];
            tensor<string, []> var_53831_equation_0 = const()[name = tensor<string, []>("op_53831_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53831_cast_fp16 = einsum(equation = var_53831_equation_0, values = (var_53659_cast_fp16, var_53806_cast_fp16))[name = tensor<string, []>("op_53831_cast_fp16")];
            tensor<string, []> var_53833_equation_0 = const()[name = tensor<string, []>("op_53833_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53833_cast_fp16 = einsum(equation = var_53833_equation_0, values = (var_53663_cast_fp16, var_53807_cast_fp16))[name = tensor<string, []>("op_53833_cast_fp16")];
            tensor<string, []> var_53835_equation_0 = const()[name = tensor<string, []>("op_53835_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53835_cast_fp16 = einsum(equation = var_53835_equation_0, values = (var_53667_cast_fp16, var_53808_cast_fp16))[name = tensor<string, []>("op_53835_cast_fp16")];
            tensor<string, []> var_53837_equation_0 = const()[name = tensor<string, []>("op_53837_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53837_cast_fp16 = einsum(equation = var_53837_equation_0, values = (var_53671_cast_fp16, var_53809_cast_fp16))[name = tensor<string, []>("op_53837_cast_fp16")];
            tensor<string, []> var_53839_equation_0 = const()[name = tensor<string, []>("op_53839_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53839_cast_fp16 = einsum(equation = var_53839_equation_0, values = (var_53675_cast_fp16, var_53810_cast_fp16))[name = tensor<string, []>("op_53839_cast_fp16")];
            tensor<string, []> var_53841_equation_0 = const()[name = tensor<string, []>("op_53841_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53841_cast_fp16 = einsum(equation = var_53841_equation_0, values = (var_53679_cast_fp16, var_53811_cast_fp16))[name = tensor<string, []>("op_53841_cast_fp16")];
            tensor<string, []> var_53843_equation_0 = const()[name = tensor<string, []>("op_53843_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53843_cast_fp16 = einsum(equation = var_53843_equation_0, values = (var_53683_cast_fp16, var_53812_cast_fp16))[name = tensor<string, []>("op_53843_cast_fp16")];
            tensor<string, []> var_53845_equation_0 = const()[name = tensor<string, []>("op_53845_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53845_cast_fp16 = einsum(equation = var_53845_equation_0, values = (var_53687_cast_fp16, var_53813_cast_fp16))[name = tensor<string, []>("op_53845_cast_fp16")];
            tensor<string, []> var_53847_equation_0 = const()[name = tensor<string, []>("op_53847_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53847_cast_fp16 = einsum(equation = var_53847_equation_0, values = (var_53691_cast_fp16, var_53814_cast_fp16))[name = tensor<string, []>("op_53847_cast_fp16")];
            tensor<string, []> var_53849_equation_0 = const()[name = tensor<string, []>("op_53849_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53849_cast_fp16 = einsum(equation = var_53849_equation_0, values = (var_53695_cast_fp16, var_53815_cast_fp16))[name = tensor<string, []>("op_53849_cast_fp16")];
            tensor<string, []> var_53851_equation_0 = const()[name = tensor<string, []>("op_53851_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53851_cast_fp16 = einsum(equation = var_53851_equation_0, values = (var_53699_cast_fp16, var_53816_cast_fp16))[name = tensor<string, []>("op_53851_cast_fp16")];
            tensor<string, []> var_53853_equation_0 = const()[name = tensor<string, []>("op_53853_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53853_cast_fp16 = einsum(equation = var_53853_equation_0, values = (var_53703_cast_fp16, var_53817_cast_fp16))[name = tensor<string, []>("op_53853_cast_fp16")];
            tensor<string, []> var_53855_equation_0 = const()[name = tensor<string, []>("op_53855_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53855_cast_fp16 = einsum(equation = var_53855_equation_0, values = (var_53707_cast_fp16, var_53818_cast_fp16))[name = tensor<string, []>("op_53855_cast_fp16")];
            tensor<string, []> var_53857_equation_0 = const()[name = tensor<string, []>("op_53857_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53857_cast_fp16 = einsum(equation = var_53857_equation_0, values = (var_53711_cast_fp16, var_53819_cast_fp16))[name = tensor<string, []>("op_53857_cast_fp16")];
            tensor<string, []> var_53859_equation_0 = const()[name = tensor<string, []>("op_53859_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53859_cast_fp16 = einsum(equation = var_53859_equation_0, values = (var_53715_cast_fp16, var_53820_cast_fp16))[name = tensor<string, []>("op_53859_cast_fp16")];
            tensor<string, []> var_53861_equation_0 = const()[name = tensor<string, []>("op_53861_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53861_cast_fp16 = einsum(equation = var_53861_equation_0, values = (var_53719_cast_fp16, var_53821_cast_fp16))[name = tensor<string, []>("op_53861_cast_fp16")];
            tensor<bool, []> input_675_interleave_0 = const()[name = tensor<string, []>("input_675_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_675_cast_fp16 = concat(axis = var_30355, interleave = input_675_interleave_0, values = (var_53823_cast_fp16, var_53825_cast_fp16, var_53827_cast_fp16, var_53829_cast_fp16, var_53831_cast_fp16, var_53833_cast_fp16, var_53835_cast_fp16, var_53837_cast_fp16, var_53839_cast_fp16, var_53841_cast_fp16, var_53843_cast_fp16, var_53845_cast_fp16, var_53847_cast_fp16, var_53849_cast_fp16, var_53851_cast_fp16, var_53853_cast_fp16, var_53855_cast_fp16, var_53857_cast_fp16, var_53859_cast_fp16, var_53861_cast_fp16))[name = tensor<string, []>("input_675_cast_fp16")];
            tensor<string, []> var_53871_pad_type_0 = const()[name = tensor<string, []>("op_53871_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_53871_strides_0 = const()[name = tensor<string, []>("op_53871_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_53871_pad_0 = const()[name = tensor<string, []>("op_53871_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_53871_dilations_0 = const()[name = tensor<string, []>("op_53871_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_53871_groups_0 = const()[name = tensor<string, []>("op_53871_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_5_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(732422336))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(733651200))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_5_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_5_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_5_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(733651392)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_53871_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_5_attn1_to_out_0_bias_to_fp16, dilations = var_53871_dilations_0, groups = var_53871_groups_0, pad = var_53871_pad_0, pad_type = var_53871_pad_type_0, strides = var_53871_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_5_attn1_to_out_0_weight_to_fp16_palettized, x = input_675_cast_fp16)[name = tensor<string, []>("op_53871_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_357_cast_fp16 = add(x = var_53871_cast_fp16, y = inputs_355_cast_fp16)[name = tensor<string, []>("inputs_357_cast_fp16")];
            tensor<int32, [1]> hidden_states_467_axes_0 = const()[name = tensor<string, []>("hidden_states_467_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_467_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_467_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(733654016)))];
            tensor<fp16, [1280]> hidden_states_467_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_467_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(733656640)))];
            tensor<fp16, []> var_53881_to_fp16 = const()[name = tensor<string, []>("op_53881_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_467_cast_fp16 = layer_norm(axes = hidden_states_467_axes_0, beta = hidden_states_467_beta_0_to_fp16, epsilon = var_53881_to_fp16, gamma = hidden_states_467_gamma_0_to_fp16, x = inputs_357_cast_fp16)[name = tensor<string, []>("hidden_states_467_cast_fp16")];
            tensor<string, []> q_239_pad_type_0 = const()[name = tensor<string, []>("q_239_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_239_strides_0 = const()[name = tensor<string, []>("q_239_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_239_pad_0 = const()[name = tensor<string, []>("q_239_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_239_dilations_0 = const()[name = tensor<string, []>("q_239_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_239_groups_0 = const()[name = tensor<string, []>("q_239_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_5_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(733659264))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(734888128))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_5_attn2_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_239_cast_fp16 = conv(dilations = q_239_dilations_0, groups = q_239_groups_0, pad = q_239_pad_0, pad_type = q_239_pad_type_0, strides = q_239_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_5_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_467_cast_fp16)[name = tensor<string, []>("q_239_cast_fp16")];
            tensor<string, []> k_477_pad_type_0 = const()[name = tensor<string, []>("k_477_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_477_strides_0 = const()[name = tensor<string, []>("k_477_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_477_pad_0 = const()[name = tensor<string, []>("k_477_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_477_dilations_0 = const()[name = tensor<string, []>("k_477_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_477_groups_0 = const()[name = tensor<string, []>("k_477_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_5_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(734888320))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(736854464))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_5_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> k_477_cast_fp16 = conv(dilations = k_477_dilations_0, groups = k_477_groups_0, pad = k_477_pad_0, pad_type = k_477_pad_type_0, strides = k_477_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_5_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("k_477_cast_fp16")];
            tensor<string, []> v_239_pad_type_0 = const()[name = tensor<string, []>("v_239_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_239_strides_0 = const()[name = tensor<string, []>("v_239_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_239_pad_0 = const()[name = tensor<string, []>("v_239_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_239_dilations_0 = const()[name = tensor<string, []>("v_239_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_239_groups_0 = const()[name = tensor<string, []>("v_239_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_5_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(736854656))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(738820800))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_5_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> v_239_cast_fp16 = conv(dilations = v_239_dilations_0, groups = v_239_groups_0, pad = v_239_pad_0, pad_type = v_239_pad_type_0, strides = v_239_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_5_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("v_239_cast_fp16")];
            tensor<int32, [4]> var_53914_begin_0 = const()[name = tensor<string, []>("op_53914_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_53914_end_0 = const()[name = tensor<string, []>("op_53914_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_53914_end_mask_0 = const()[name = tensor<string, []>("op_53914_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53914_cast_fp16 = slice_by_index(begin = var_53914_begin_0, end = var_53914_end_0, end_mask = var_53914_end_mask_0, x = q_239_cast_fp16)[name = tensor<string, []>("op_53914_cast_fp16")];
            tensor<int32, [4]> var_53918_begin_0 = const()[name = tensor<string, []>("op_53918_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_53918_end_0 = const()[name = tensor<string, []>("op_53918_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_53918_end_mask_0 = const()[name = tensor<string, []>("op_53918_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53918_cast_fp16 = slice_by_index(begin = var_53918_begin_0, end = var_53918_end_0, end_mask = var_53918_end_mask_0, x = q_239_cast_fp16)[name = tensor<string, []>("op_53918_cast_fp16")];
            tensor<int32, [4]> var_53922_begin_0 = const()[name = tensor<string, []>("op_53922_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_53922_end_0 = const()[name = tensor<string, []>("op_53922_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_53922_end_mask_0 = const()[name = tensor<string, []>("op_53922_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53922_cast_fp16 = slice_by_index(begin = var_53922_begin_0, end = var_53922_end_0, end_mask = var_53922_end_mask_0, x = q_239_cast_fp16)[name = tensor<string, []>("op_53922_cast_fp16")];
            tensor<int32, [4]> var_53926_begin_0 = const()[name = tensor<string, []>("op_53926_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_53926_end_0 = const()[name = tensor<string, []>("op_53926_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_53926_end_mask_0 = const()[name = tensor<string, []>("op_53926_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53926_cast_fp16 = slice_by_index(begin = var_53926_begin_0, end = var_53926_end_0, end_mask = var_53926_end_mask_0, x = q_239_cast_fp16)[name = tensor<string, []>("op_53926_cast_fp16")];
            tensor<int32, [4]> var_53930_begin_0 = const()[name = tensor<string, []>("op_53930_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_53930_end_0 = const()[name = tensor<string, []>("op_53930_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_53930_end_mask_0 = const()[name = tensor<string, []>("op_53930_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53930_cast_fp16 = slice_by_index(begin = var_53930_begin_0, end = var_53930_end_0, end_mask = var_53930_end_mask_0, x = q_239_cast_fp16)[name = tensor<string, []>("op_53930_cast_fp16")];
            tensor<int32, [4]> var_53934_begin_0 = const()[name = tensor<string, []>("op_53934_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_53934_end_0 = const()[name = tensor<string, []>("op_53934_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_53934_end_mask_0 = const()[name = tensor<string, []>("op_53934_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53934_cast_fp16 = slice_by_index(begin = var_53934_begin_0, end = var_53934_end_0, end_mask = var_53934_end_mask_0, x = q_239_cast_fp16)[name = tensor<string, []>("op_53934_cast_fp16")];
            tensor<int32, [4]> var_53938_begin_0 = const()[name = tensor<string, []>("op_53938_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_53938_end_0 = const()[name = tensor<string, []>("op_53938_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_53938_end_mask_0 = const()[name = tensor<string, []>("op_53938_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53938_cast_fp16 = slice_by_index(begin = var_53938_begin_0, end = var_53938_end_0, end_mask = var_53938_end_mask_0, x = q_239_cast_fp16)[name = tensor<string, []>("op_53938_cast_fp16")];
            tensor<int32, [4]> var_53942_begin_0 = const()[name = tensor<string, []>("op_53942_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_53942_end_0 = const()[name = tensor<string, []>("op_53942_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_53942_end_mask_0 = const()[name = tensor<string, []>("op_53942_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53942_cast_fp16 = slice_by_index(begin = var_53942_begin_0, end = var_53942_end_0, end_mask = var_53942_end_mask_0, x = q_239_cast_fp16)[name = tensor<string, []>("op_53942_cast_fp16")];
            tensor<int32, [4]> var_53946_begin_0 = const()[name = tensor<string, []>("op_53946_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_53946_end_0 = const()[name = tensor<string, []>("op_53946_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_53946_end_mask_0 = const()[name = tensor<string, []>("op_53946_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53946_cast_fp16 = slice_by_index(begin = var_53946_begin_0, end = var_53946_end_0, end_mask = var_53946_end_mask_0, x = q_239_cast_fp16)[name = tensor<string, []>("op_53946_cast_fp16")];
            tensor<int32, [4]> var_53950_begin_0 = const()[name = tensor<string, []>("op_53950_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_53950_end_0 = const()[name = tensor<string, []>("op_53950_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_53950_end_mask_0 = const()[name = tensor<string, []>("op_53950_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53950_cast_fp16 = slice_by_index(begin = var_53950_begin_0, end = var_53950_end_0, end_mask = var_53950_end_mask_0, x = q_239_cast_fp16)[name = tensor<string, []>("op_53950_cast_fp16")];
            tensor<int32, [4]> var_53954_begin_0 = const()[name = tensor<string, []>("op_53954_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_53954_end_0 = const()[name = tensor<string, []>("op_53954_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_53954_end_mask_0 = const()[name = tensor<string, []>("op_53954_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53954_cast_fp16 = slice_by_index(begin = var_53954_begin_0, end = var_53954_end_0, end_mask = var_53954_end_mask_0, x = q_239_cast_fp16)[name = tensor<string, []>("op_53954_cast_fp16")];
            tensor<int32, [4]> var_53958_begin_0 = const()[name = tensor<string, []>("op_53958_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_53958_end_0 = const()[name = tensor<string, []>("op_53958_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_53958_end_mask_0 = const()[name = tensor<string, []>("op_53958_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53958_cast_fp16 = slice_by_index(begin = var_53958_begin_0, end = var_53958_end_0, end_mask = var_53958_end_mask_0, x = q_239_cast_fp16)[name = tensor<string, []>("op_53958_cast_fp16")];
            tensor<int32, [4]> var_53962_begin_0 = const()[name = tensor<string, []>("op_53962_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_53962_end_0 = const()[name = tensor<string, []>("op_53962_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_53962_end_mask_0 = const()[name = tensor<string, []>("op_53962_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53962_cast_fp16 = slice_by_index(begin = var_53962_begin_0, end = var_53962_end_0, end_mask = var_53962_end_mask_0, x = q_239_cast_fp16)[name = tensor<string, []>("op_53962_cast_fp16")];
            tensor<int32, [4]> var_53966_begin_0 = const()[name = tensor<string, []>("op_53966_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_53966_end_0 = const()[name = tensor<string, []>("op_53966_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_53966_end_mask_0 = const()[name = tensor<string, []>("op_53966_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53966_cast_fp16 = slice_by_index(begin = var_53966_begin_0, end = var_53966_end_0, end_mask = var_53966_end_mask_0, x = q_239_cast_fp16)[name = tensor<string, []>("op_53966_cast_fp16")];
            tensor<int32, [4]> var_53970_begin_0 = const()[name = tensor<string, []>("op_53970_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_53970_end_0 = const()[name = tensor<string, []>("op_53970_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_53970_end_mask_0 = const()[name = tensor<string, []>("op_53970_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53970_cast_fp16 = slice_by_index(begin = var_53970_begin_0, end = var_53970_end_0, end_mask = var_53970_end_mask_0, x = q_239_cast_fp16)[name = tensor<string, []>("op_53970_cast_fp16")];
            tensor<int32, [4]> var_53974_begin_0 = const()[name = tensor<string, []>("op_53974_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_53974_end_0 = const()[name = tensor<string, []>("op_53974_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_53974_end_mask_0 = const()[name = tensor<string, []>("op_53974_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53974_cast_fp16 = slice_by_index(begin = var_53974_begin_0, end = var_53974_end_0, end_mask = var_53974_end_mask_0, x = q_239_cast_fp16)[name = tensor<string, []>("op_53974_cast_fp16")];
            tensor<int32, [4]> var_53978_begin_0 = const()[name = tensor<string, []>("op_53978_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_53978_end_0 = const()[name = tensor<string, []>("op_53978_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_53978_end_mask_0 = const()[name = tensor<string, []>("op_53978_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53978_cast_fp16 = slice_by_index(begin = var_53978_begin_0, end = var_53978_end_0, end_mask = var_53978_end_mask_0, x = q_239_cast_fp16)[name = tensor<string, []>("op_53978_cast_fp16")];
            tensor<int32, [4]> var_53982_begin_0 = const()[name = tensor<string, []>("op_53982_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_53982_end_0 = const()[name = tensor<string, []>("op_53982_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_53982_end_mask_0 = const()[name = tensor<string, []>("op_53982_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53982_cast_fp16 = slice_by_index(begin = var_53982_begin_0, end = var_53982_end_0, end_mask = var_53982_end_mask_0, x = q_239_cast_fp16)[name = tensor<string, []>("op_53982_cast_fp16")];
            tensor<int32, [4]> var_53986_begin_0 = const()[name = tensor<string, []>("op_53986_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_53986_end_0 = const()[name = tensor<string, []>("op_53986_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_53986_end_mask_0 = const()[name = tensor<string, []>("op_53986_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53986_cast_fp16 = slice_by_index(begin = var_53986_begin_0, end = var_53986_end_0, end_mask = var_53986_end_mask_0, x = q_239_cast_fp16)[name = tensor<string, []>("op_53986_cast_fp16")];
            tensor<int32, [4]> var_53990_begin_0 = const()[name = tensor<string, []>("op_53990_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_53990_end_0 = const()[name = tensor<string, []>("op_53990_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_53990_end_mask_0 = const()[name = tensor<string, []>("op_53990_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53990_cast_fp16 = slice_by_index(begin = var_53990_begin_0, end = var_53990_end_0, end_mask = var_53990_end_mask_0, x = q_239_cast_fp16)[name = tensor<string, []>("op_53990_cast_fp16")];
            tensor<int32, [4]> k_479_perm_0 = const()[name = tensor<string, []>("k_479_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_53997_begin_0 = const()[name = tensor<string, []>("op_53997_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_53997_end_0 = const()[name = tensor<string, []>("op_53997_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_53997_end_mask_0 = const()[name = tensor<string, []>("op_53997_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> k_479_cast_fp16 = transpose(perm = k_479_perm_0, x = k_477_cast_fp16)[name = tensor<string, []>("transpose_20")];
            tensor<fp16, [2, 77, 1, 64]> var_53997_cast_fp16 = slice_by_index(begin = var_53997_begin_0, end = var_53997_end_0, end_mask = var_53997_end_mask_0, x = k_479_cast_fp16)[name = tensor<string, []>("op_53997_cast_fp16")];
            tensor<int32, [4]> var_54001_begin_0 = const()[name = tensor<string, []>("op_54001_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_54001_end_0 = const()[name = tensor<string, []>("op_54001_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_54001_end_mask_0 = const()[name = tensor<string, []>("op_54001_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54001_cast_fp16 = slice_by_index(begin = var_54001_begin_0, end = var_54001_end_0, end_mask = var_54001_end_mask_0, x = k_479_cast_fp16)[name = tensor<string, []>("op_54001_cast_fp16")];
            tensor<int32, [4]> var_54005_begin_0 = const()[name = tensor<string, []>("op_54005_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_54005_end_0 = const()[name = tensor<string, []>("op_54005_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_54005_end_mask_0 = const()[name = tensor<string, []>("op_54005_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54005_cast_fp16 = slice_by_index(begin = var_54005_begin_0, end = var_54005_end_0, end_mask = var_54005_end_mask_0, x = k_479_cast_fp16)[name = tensor<string, []>("op_54005_cast_fp16")];
            tensor<int32, [4]> var_54009_begin_0 = const()[name = tensor<string, []>("op_54009_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_54009_end_0 = const()[name = tensor<string, []>("op_54009_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_54009_end_mask_0 = const()[name = tensor<string, []>("op_54009_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54009_cast_fp16 = slice_by_index(begin = var_54009_begin_0, end = var_54009_end_0, end_mask = var_54009_end_mask_0, x = k_479_cast_fp16)[name = tensor<string, []>("op_54009_cast_fp16")];
            tensor<int32, [4]> var_54013_begin_0 = const()[name = tensor<string, []>("op_54013_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_54013_end_0 = const()[name = tensor<string, []>("op_54013_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_54013_end_mask_0 = const()[name = tensor<string, []>("op_54013_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54013_cast_fp16 = slice_by_index(begin = var_54013_begin_0, end = var_54013_end_0, end_mask = var_54013_end_mask_0, x = k_479_cast_fp16)[name = tensor<string, []>("op_54013_cast_fp16")];
            tensor<int32, [4]> var_54017_begin_0 = const()[name = tensor<string, []>("op_54017_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_54017_end_0 = const()[name = tensor<string, []>("op_54017_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_54017_end_mask_0 = const()[name = tensor<string, []>("op_54017_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54017_cast_fp16 = slice_by_index(begin = var_54017_begin_0, end = var_54017_end_0, end_mask = var_54017_end_mask_0, x = k_479_cast_fp16)[name = tensor<string, []>("op_54017_cast_fp16")];
            tensor<int32, [4]> var_54021_begin_0 = const()[name = tensor<string, []>("op_54021_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_54021_end_0 = const()[name = tensor<string, []>("op_54021_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_54021_end_mask_0 = const()[name = tensor<string, []>("op_54021_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54021_cast_fp16 = slice_by_index(begin = var_54021_begin_0, end = var_54021_end_0, end_mask = var_54021_end_mask_0, x = k_479_cast_fp16)[name = tensor<string, []>("op_54021_cast_fp16")];
            tensor<int32, [4]> var_54025_begin_0 = const()[name = tensor<string, []>("op_54025_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_54025_end_0 = const()[name = tensor<string, []>("op_54025_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_54025_end_mask_0 = const()[name = tensor<string, []>("op_54025_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54025_cast_fp16 = slice_by_index(begin = var_54025_begin_0, end = var_54025_end_0, end_mask = var_54025_end_mask_0, x = k_479_cast_fp16)[name = tensor<string, []>("op_54025_cast_fp16")];
            tensor<int32, [4]> var_54029_begin_0 = const()[name = tensor<string, []>("op_54029_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_54029_end_0 = const()[name = tensor<string, []>("op_54029_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_54029_end_mask_0 = const()[name = tensor<string, []>("op_54029_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54029_cast_fp16 = slice_by_index(begin = var_54029_begin_0, end = var_54029_end_0, end_mask = var_54029_end_mask_0, x = k_479_cast_fp16)[name = tensor<string, []>("op_54029_cast_fp16")];
            tensor<int32, [4]> var_54033_begin_0 = const()[name = tensor<string, []>("op_54033_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_54033_end_0 = const()[name = tensor<string, []>("op_54033_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_54033_end_mask_0 = const()[name = tensor<string, []>("op_54033_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54033_cast_fp16 = slice_by_index(begin = var_54033_begin_0, end = var_54033_end_0, end_mask = var_54033_end_mask_0, x = k_479_cast_fp16)[name = tensor<string, []>("op_54033_cast_fp16")];
            tensor<int32, [4]> var_54037_begin_0 = const()[name = tensor<string, []>("op_54037_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_54037_end_0 = const()[name = tensor<string, []>("op_54037_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_54037_end_mask_0 = const()[name = tensor<string, []>("op_54037_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54037_cast_fp16 = slice_by_index(begin = var_54037_begin_0, end = var_54037_end_0, end_mask = var_54037_end_mask_0, x = k_479_cast_fp16)[name = tensor<string, []>("op_54037_cast_fp16")];
            tensor<int32, [4]> var_54041_begin_0 = const()[name = tensor<string, []>("op_54041_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_54041_end_0 = const()[name = tensor<string, []>("op_54041_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_54041_end_mask_0 = const()[name = tensor<string, []>("op_54041_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54041_cast_fp16 = slice_by_index(begin = var_54041_begin_0, end = var_54041_end_0, end_mask = var_54041_end_mask_0, x = k_479_cast_fp16)[name = tensor<string, []>("op_54041_cast_fp16")];
            tensor<int32, [4]> var_54045_begin_0 = const()[name = tensor<string, []>("op_54045_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_54045_end_0 = const()[name = tensor<string, []>("op_54045_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_54045_end_mask_0 = const()[name = tensor<string, []>("op_54045_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54045_cast_fp16 = slice_by_index(begin = var_54045_begin_0, end = var_54045_end_0, end_mask = var_54045_end_mask_0, x = k_479_cast_fp16)[name = tensor<string, []>("op_54045_cast_fp16")];
            tensor<int32, [4]> var_54049_begin_0 = const()[name = tensor<string, []>("op_54049_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_54049_end_0 = const()[name = tensor<string, []>("op_54049_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_54049_end_mask_0 = const()[name = tensor<string, []>("op_54049_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54049_cast_fp16 = slice_by_index(begin = var_54049_begin_0, end = var_54049_end_0, end_mask = var_54049_end_mask_0, x = k_479_cast_fp16)[name = tensor<string, []>("op_54049_cast_fp16")];
            tensor<int32, [4]> var_54053_begin_0 = const()[name = tensor<string, []>("op_54053_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_54053_end_0 = const()[name = tensor<string, []>("op_54053_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_54053_end_mask_0 = const()[name = tensor<string, []>("op_54053_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54053_cast_fp16 = slice_by_index(begin = var_54053_begin_0, end = var_54053_end_0, end_mask = var_54053_end_mask_0, x = k_479_cast_fp16)[name = tensor<string, []>("op_54053_cast_fp16")];
            tensor<int32, [4]> var_54057_begin_0 = const()[name = tensor<string, []>("op_54057_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_54057_end_0 = const()[name = tensor<string, []>("op_54057_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_54057_end_mask_0 = const()[name = tensor<string, []>("op_54057_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54057_cast_fp16 = slice_by_index(begin = var_54057_begin_0, end = var_54057_end_0, end_mask = var_54057_end_mask_0, x = k_479_cast_fp16)[name = tensor<string, []>("op_54057_cast_fp16")];
            tensor<int32, [4]> var_54061_begin_0 = const()[name = tensor<string, []>("op_54061_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_54061_end_0 = const()[name = tensor<string, []>("op_54061_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_54061_end_mask_0 = const()[name = tensor<string, []>("op_54061_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54061_cast_fp16 = slice_by_index(begin = var_54061_begin_0, end = var_54061_end_0, end_mask = var_54061_end_mask_0, x = k_479_cast_fp16)[name = tensor<string, []>("op_54061_cast_fp16")];
            tensor<int32, [4]> var_54065_begin_0 = const()[name = tensor<string, []>("op_54065_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_54065_end_0 = const()[name = tensor<string, []>("op_54065_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_54065_end_mask_0 = const()[name = tensor<string, []>("op_54065_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54065_cast_fp16 = slice_by_index(begin = var_54065_begin_0, end = var_54065_end_0, end_mask = var_54065_end_mask_0, x = k_479_cast_fp16)[name = tensor<string, []>("op_54065_cast_fp16")];
            tensor<int32, [4]> var_54069_begin_0 = const()[name = tensor<string, []>("op_54069_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_54069_end_0 = const()[name = tensor<string, []>("op_54069_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_54069_end_mask_0 = const()[name = tensor<string, []>("op_54069_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54069_cast_fp16 = slice_by_index(begin = var_54069_begin_0, end = var_54069_end_0, end_mask = var_54069_end_mask_0, x = k_479_cast_fp16)[name = tensor<string, []>("op_54069_cast_fp16")];
            tensor<int32, [4]> var_54073_begin_0 = const()[name = tensor<string, []>("op_54073_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_54073_end_0 = const()[name = tensor<string, []>("op_54073_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_54073_end_mask_0 = const()[name = tensor<string, []>("op_54073_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54073_cast_fp16 = slice_by_index(begin = var_54073_begin_0, end = var_54073_end_0, end_mask = var_54073_end_mask_0, x = k_479_cast_fp16)[name = tensor<string, []>("op_54073_cast_fp16")];
            tensor<int32, [4]> var_54075_begin_0 = const()[name = tensor<string, []>("op_54075_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_54075_end_0 = const()[name = tensor<string, []>("op_54075_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_54075_end_mask_0 = const()[name = tensor<string, []>("op_54075_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_54075_cast_fp16 = slice_by_index(begin = var_54075_begin_0, end = var_54075_end_0, end_mask = var_54075_end_mask_0, x = v_239_cast_fp16)[name = tensor<string, []>("op_54075_cast_fp16")];
            tensor<int32, [4]> var_54079_begin_0 = const()[name = tensor<string, []>("op_54079_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_54079_end_0 = const()[name = tensor<string, []>("op_54079_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_54079_end_mask_0 = const()[name = tensor<string, []>("op_54079_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_54079_cast_fp16 = slice_by_index(begin = var_54079_begin_0, end = var_54079_end_0, end_mask = var_54079_end_mask_0, x = v_239_cast_fp16)[name = tensor<string, []>("op_54079_cast_fp16")];
            tensor<int32, [4]> var_54083_begin_0 = const()[name = tensor<string, []>("op_54083_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_54083_end_0 = const()[name = tensor<string, []>("op_54083_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_54083_end_mask_0 = const()[name = tensor<string, []>("op_54083_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_54083_cast_fp16 = slice_by_index(begin = var_54083_begin_0, end = var_54083_end_0, end_mask = var_54083_end_mask_0, x = v_239_cast_fp16)[name = tensor<string, []>("op_54083_cast_fp16")];
            tensor<int32, [4]> var_54087_begin_0 = const()[name = tensor<string, []>("op_54087_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_54087_end_0 = const()[name = tensor<string, []>("op_54087_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_54087_end_mask_0 = const()[name = tensor<string, []>("op_54087_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_54087_cast_fp16 = slice_by_index(begin = var_54087_begin_0, end = var_54087_end_0, end_mask = var_54087_end_mask_0, x = v_239_cast_fp16)[name = tensor<string, []>("op_54087_cast_fp16")];
            tensor<int32, [4]> var_54091_begin_0 = const()[name = tensor<string, []>("op_54091_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_54091_end_0 = const()[name = tensor<string, []>("op_54091_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_54091_end_mask_0 = const()[name = tensor<string, []>("op_54091_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_54091_cast_fp16 = slice_by_index(begin = var_54091_begin_0, end = var_54091_end_0, end_mask = var_54091_end_mask_0, x = v_239_cast_fp16)[name = tensor<string, []>("op_54091_cast_fp16")];
            tensor<int32, [4]> var_54095_begin_0 = const()[name = tensor<string, []>("op_54095_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_54095_end_0 = const()[name = tensor<string, []>("op_54095_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_54095_end_mask_0 = const()[name = tensor<string, []>("op_54095_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_54095_cast_fp16 = slice_by_index(begin = var_54095_begin_0, end = var_54095_end_0, end_mask = var_54095_end_mask_0, x = v_239_cast_fp16)[name = tensor<string, []>("op_54095_cast_fp16")];
            tensor<int32, [4]> var_54099_begin_0 = const()[name = tensor<string, []>("op_54099_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_54099_end_0 = const()[name = tensor<string, []>("op_54099_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_54099_end_mask_0 = const()[name = tensor<string, []>("op_54099_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_54099_cast_fp16 = slice_by_index(begin = var_54099_begin_0, end = var_54099_end_0, end_mask = var_54099_end_mask_0, x = v_239_cast_fp16)[name = tensor<string, []>("op_54099_cast_fp16")];
            tensor<int32, [4]> var_54103_begin_0 = const()[name = tensor<string, []>("op_54103_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_54103_end_0 = const()[name = tensor<string, []>("op_54103_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_54103_end_mask_0 = const()[name = tensor<string, []>("op_54103_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_54103_cast_fp16 = slice_by_index(begin = var_54103_begin_0, end = var_54103_end_0, end_mask = var_54103_end_mask_0, x = v_239_cast_fp16)[name = tensor<string, []>("op_54103_cast_fp16")];
            tensor<int32, [4]> var_54107_begin_0 = const()[name = tensor<string, []>("op_54107_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_54107_end_0 = const()[name = tensor<string, []>("op_54107_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_54107_end_mask_0 = const()[name = tensor<string, []>("op_54107_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_54107_cast_fp16 = slice_by_index(begin = var_54107_begin_0, end = var_54107_end_0, end_mask = var_54107_end_mask_0, x = v_239_cast_fp16)[name = tensor<string, []>("op_54107_cast_fp16")];
            tensor<int32, [4]> var_54111_begin_0 = const()[name = tensor<string, []>("op_54111_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_54111_end_0 = const()[name = tensor<string, []>("op_54111_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_54111_end_mask_0 = const()[name = tensor<string, []>("op_54111_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_54111_cast_fp16 = slice_by_index(begin = var_54111_begin_0, end = var_54111_end_0, end_mask = var_54111_end_mask_0, x = v_239_cast_fp16)[name = tensor<string, []>("op_54111_cast_fp16")];
            tensor<int32, [4]> var_54115_begin_0 = const()[name = tensor<string, []>("op_54115_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_54115_end_0 = const()[name = tensor<string, []>("op_54115_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_54115_end_mask_0 = const()[name = tensor<string, []>("op_54115_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_54115_cast_fp16 = slice_by_index(begin = var_54115_begin_0, end = var_54115_end_0, end_mask = var_54115_end_mask_0, x = v_239_cast_fp16)[name = tensor<string, []>("op_54115_cast_fp16")];
            tensor<int32, [4]> var_54119_begin_0 = const()[name = tensor<string, []>("op_54119_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_54119_end_0 = const()[name = tensor<string, []>("op_54119_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_54119_end_mask_0 = const()[name = tensor<string, []>("op_54119_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_54119_cast_fp16 = slice_by_index(begin = var_54119_begin_0, end = var_54119_end_0, end_mask = var_54119_end_mask_0, x = v_239_cast_fp16)[name = tensor<string, []>("op_54119_cast_fp16")];
            tensor<int32, [4]> var_54123_begin_0 = const()[name = tensor<string, []>("op_54123_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_54123_end_0 = const()[name = tensor<string, []>("op_54123_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_54123_end_mask_0 = const()[name = tensor<string, []>("op_54123_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_54123_cast_fp16 = slice_by_index(begin = var_54123_begin_0, end = var_54123_end_0, end_mask = var_54123_end_mask_0, x = v_239_cast_fp16)[name = tensor<string, []>("op_54123_cast_fp16")];
            tensor<int32, [4]> var_54127_begin_0 = const()[name = tensor<string, []>("op_54127_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_54127_end_0 = const()[name = tensor<string, []>("op_54127_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_54127_end_mask_0 = const()[name = tensor<string, []>("op_54127_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_54127_cast_fp16 = slice_by_index(begin = var_54127_begin_0, end = var_54127_end_0, end_mask = var_54127_end_mask_0, x = v_239_cast_fp16)[name = tensor<string, []>("op_54127_cast_fp16")];
            tensor<int32, [4]> var_54131_begin_0 = const()[name = tensor<string, []>("op_54131_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_54131_end_0 = const()[name = tensor<string, []>("op_54131_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_54131_end_mask_0 = const()[name = tensor<string, []>("op_54131_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_54131_cast_fp16 = slice_by_index(begin = var_54131_begin_0, end = var_54131_end_0, end_mask = var_54131_end_mask_0, x = v_239_cast_fp16)[name = tensor<string, []>("op_54131_cast_fp16")];
            tensor<int32, [4]> var_54135_begin_0 = const()[name = tensor<string, []>("op_54135_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_54135_end_0 = const()[name = tensor<string, []>("op_54135_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_54135_end_mask_0 = const()[name = tensor<string, []>("op_54135_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_54135_cast_fp16 = slice_by_index(begin = var_54135_begin_0, end = var_54135_end_0, end_mask = var_54135_end_mask_0, x = v_239_cast_fp16)[name = tensor<string, []>("op_54135_cast_fp16")];
            tensor<int32, [4]> var_54139_begin_0 = const()[name = tensor<string, []>("op_54139_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_54139_end_0 = const()[name = tensor<string, []>("op_54139_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_54139_end_mask_0 = const()[name = tensor<string, []>("op_54139_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_54139_cast_fp16 = slice_by_index(begin = var_54139_begin_0, end = var_54139_end_0, end_mask = var_54139_end_mask_0, x = v_239_cast_fp16)[name = tensor<string, []>("op_54139_cast_fp16")];
            tensor<int32, [4]> var_54143_begin_0 = const()[name = tensor<string, []>("op_54143_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_54143_end_0 = const()[name = tensor<string, []>("op_54143_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_54143_end_mask_0 = const()[name = tensor<string, []>("op_54143_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_54143_cast_fp16 = slice_by_index(begin = var_54143_begin_0, end = var_54143_end_0, end_mask = var_54143_end_mask_0, x = v_239_cast_fp16)[name = tensor<string, []>("op_54143_cast_fp16")];
            tensor<int32, [4]> var_54147_begin_0 = const()[name = tensor<string, []>("op_54147_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_54147_end_0 = const()[name = tensor<string, []>("op_54147_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_54147_end_mask_0 = const()[name = tensor<string, []>("op_54147_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_54147_cast_fp16 = slice_by_index(begin = var_54147_begin_0, end = var_54147_end_0, end_mask = var_54147_end_mask_0, x = v_239_cast_fp16)[name = tensor<string, []>("op_54147_cast_fp16")];
            tensor<int32, [4]> var_54151_begin_0 = const()[name = tensor<string, []>("op_54151_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_54151_end_0 = const()[name = tensor<string, []>("op_54151_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_54151_end_mask_0 = const()[name = tensor<string, []>("op_54151_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_54151_cast_fp16 = slice_by_index(begin = var_54151_begin_0, end = var_54151_end_0, end_mask = var_54151_end_mask_0, x = v_239_cast_fp16)[name = tensor<string, []>("op_54151_cast_fp16")];
            tensor<string, []> var_54155_equation_0 = const()[name = tensor<string, []>("op_54155_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_54155_cast_fp16 = einsum(equation = var_54155_equation_0, values = (var_53997_cast_fp16, var_53914_cast_fp16))[name = tensor<string, []>("op_54155_cast_fp16")];
            tensor<fp16, []> var_54156_to_fp16 = const()[name = tensor<string, []>("op_54156_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4601_cast_fp16 = mul(x = var_54155_cast_fp16, y = var_54156_to_fp16)[name = tensor<string, []>("aw_4601_cast_fp16")];
            tensor<string, []> var_54159_equation_0 = const()[name = tensor<string, []>("op_54159_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_54159_cast_fp16 = einsum(equation = var_54159_equation_0, values = (var_54001_cast_fp16, var_53918_cast_fp16))[name = tensor<string, []>("op_54159_cast_fp16")];
            tensor<fp16, []> var_54160_to_fp16 = const()[name = tensor<string, []>("op_54160_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4603_cast_fp16 = mul(x = var_54159_cast_fp16, y = var_54160_to_fp16)[name = tensor<string, []>("aw_4603_cast_fp16")];
            tensor<string, []> var_54163_equation_0 = const()[name = tensor<string, []>("op_54163_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_54163_cast_fp16 = einsum(equation = var_54163_equation_0, values = (var_54005_cast_fp16, var_53922_cast_fp16))[name = tensor<string, []>("op_54163_cast_fp16")];
            tensor<fp16, []> var_54164_to_fp16 = const()[name = tensor<string, []>("op_54164_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4605_cast_fp16 = mul(x = var_54163_cast_fp16, y = var_54164_to_fp16)[name = tensor<string, []>("aw_4605_cast_fp16")];
            tensor<string, []> var_54167_equation_0 = const()[name = tensor<string, []>("op_54167_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_54167_cast_fp16 = einsum(equation = var_54167_equation_0, values = (var_54009_cast_fp16, var_53926_cast_fp16))[name = tensor<string, []>("op_54167_cast_fp16")];
            tensor<fp16, []> var_54168_to_fp16 = const()[name = tensor<string, []>("op_54168_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4607_cast_fp16 = mul(x = var_54167_cast_fp16, y = var_54168_to_fp16)[name = tensor<string, []>("aw_4607_cast_fp16")];
            tensor<string, []> var_54171_equation_0 = const()[name = tensor<string, []>("op_54171_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_54171_cast_fp16 = einsum(equation = var_54171_equation_0, values = (var_54013_cast_fp16, var_53930_cast_fp16))[name = tensor<string, []>("op_54171_cast_fp16")];
            tensor<fp16, []> var_54172_to_fp16 = const()[name = tensor<string, []>("op_54172_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4609_cast_fp16 = mul(x = var_54171_cast_fp16, y = var_54172_to_fp16)[name = tensor<string, []>("aw_4609_cast_fp16")];
            tensor<string, []> var_54175_equation_0 = const()[name = tensor<string, []>("op_54175_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_54175_cast_fp16 = einsum(equation = var_54175_equation_0, values = (var_54017_cast_fp16, var_53934_cast_fp16))[name = tensor<string, []>("op_54175_cast_fp16")];
            tensor<fp16, []> var_54176_to_fp16 = const()[name = tensor<string, []>("op_54176_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4611_cast_fp16 = mul(x = var_54175_cast_fp16, y = var_54176_to_fp16)[name = tensor<string, []>("aw_4611_cast_fp16")];
            tensor<string, []> var_54179_equation_0 = const()[name = tensor<string, []>("op_54179_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_54179_cast_fp16 = einsum(equation = var_54179_equation_0, values = (var_54021_cast_fp16, var_53938_cast_fp16))[name = tensor<string, []>("op_54179_cast_fp16")];
            tensor<fp16, []> var_54180_to_fp16 = const()[name = tensor<string, []>("op_54180_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4613_cast_fp16 = mul(x = var_54179_cast_fp16, y = var_54180_to_fp16)[name = tensor<string, []>("aw_4613_cast_fp16")];
            tensor<string, []> var_54183_equation_0 = const()[name = tensor<string, []>("op_54183_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_54183_cast_fp16 = einsum(equation = var_54183_equation_0, values = (var_54025_cast_fp16, var_53942_cast_fp16))[name = tensor<string, []>("op_54183_cast_fp16")];
            tensor<fp16, []> var_54184_to_fp16 = const()[name = tensor<string, []>("op_54184_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4615_cast_fp16 = mul(x = var_54183_cast_fp16, y = var_54184_to_fp16)[name = tensor<string, []>("aw_4615_cast_fp16")];
            tensor<string, []> var_54187_equation_0 = const()[name = tensor<string, []>("op_54187_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_54187_cast_fp16 = einsum(equation = var_54187_equation_0, values = (var_54029_cast_fp16, var_53946_cast_fp16))[name = tensor<string, []>("op_54187_cast_fp16")];
            tensor<fp16, []> var_54188_to_fp16 = const()[name = tensor<string, []>("op_54188_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4617_cast_fp16 = mul(x = var_54187_cast_fp16, y = var_54188_to_fp16)[name = tensor<string, []>("aw_4617_cast_fp16")];
            tensor<string, []> var_54191_equation_0 = const()[name = tensor<string, []>("op_54191_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_54191_cast_fp16 = einsum(equation = var_54191_equation_0, values = (var_54033_cast_fp16, var_53950_cast_fp16))[name = tensor<string, []>("op_54191_cast_fp16")];
            tensor<fp16, []> var_54192_to_fp16 = const()[name = tensor<string, []>("op_54192_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4619_cast_fp16 = mul(x = var_54191_cast_fp16, y = var_54192_to_fp16)[name = tensor<string, []>("aw_4619_cast_fp16")];
            tensor<string, []> var_54195_equation_0 = const()[name = tensor<string, []>("op_54195_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_54195_cast_fp16 = einsum(equation = var_54195_equation_0, values = (var_54037_cast_fp16, var_53954_cast_fp16))[name = tensor<string, []>("op_54195_cast_fp16")];
            tensor<fp16, []> var_54196_to_fp16 = const()[name = tensor<string, []>("op_54196_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4621_cast_fp16 = mul(x = var_54195_cast_fp16, y = var_54196_to_fp16)[name = tensor<string, []>("aw_4621_cast_fp16")];
            tensor<string, []> var_54199_equation_0 = const()[name = tensor<string, []>("op_54199_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_54199_cast_fp16 = einsum(equation = var_54199_equation_0, values = (var_54041_cast_fp16, var_53958_cast_fp16))[name = tensor<string, []>("op_54199_cast_fp16")];
            tensor<fp16, []> var_54200_to_fp16 = const()[name = tensor<string, []>("op_54200_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4623_cast_fp16 = mul(x = var_54199_cast_fp16, y = var_54200_to_fp16)[name = tensor<string, []>("aw_4623_cast_fp16")];
            tensor<string, []> var_54203_equation_0 = const()[name = tensor<string, []>("op_54203_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_54203_cast_fp16 = einsum(equation = var_54203_equation_0, values = (var_54045_cast_fp16, var_53962_cast_fp16))[name = tensor<string, []>("op_54203_cast_fp16")];
            tensor<fp16, []> var_54204_to_fp16 = const()[name = tensor<string, []>("op_54204_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4625_cast_fp16 = mul(x = var_54203_cast_fp16, y = var_54204_to_fp16)[name = tensor<string, []>("aw_4625_cast_fp16")];
            tensor<string, []> var_54207_equation_0 = const()[name = tensor<string, []>("op_54207_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_54207_cast_fp16 = einsum(equation = var_54207_equation_0, values = (var_54049_cast_fp16, var_53966_cast_fp16))[name = tensor<string, []>("op_54207_cast_fp16")];
            tensor<fp16, []> var_54208_to_fp16 = const()[name = tensor<string, []>("op_54208_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4627_cast_fp16 = mul(x = var_54207_cast_fp16, y = var_54208_to_fp16)[name = tensor<string, []>("aw_4627_cast_fp16")];
            tensor<string, []> var_54211_equation_0 = const()[name = tensor<string, []>("op_54211_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_54211_cast_fp16 = einsum(equation = var_54211_equation_0, values = (var_54053_cast_fp16, var_53970_cast_fp16))[name = tensor<string, []>("op_54211_cast_fp16")];
            tensor<fp16, []> var_54212_to_fp16 = const()[name = tensor<string, []>("op_54212_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4629_cast_fp16 = mul(x = var_54211_cast_fp16, y = var_54212_to_fp16)[name = tensor<string, []>("aw_4629_cast_fp16")];
            tensor<string, []> var_54215_equation_0 = const()[name = tensor<string, []>("op_54215_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_54215_cast_fp16 = einsum(equation = var_54215_equation_0, values = (var_54057_cast_fp16, var_53974_cast_fp16))[name = tensor<string, []>("op_54215_cast_fp16")];
            tensor<fp16, []> var_54216_to_fp16 = const()[name = tensor<string, []>("op_54216_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4631_cast_fp16 = mul(x = var_54215_cast_fp16, y = var_54216_to_fp16)[name = tensor<string, []>("aw_4631_cast_fp16")];
            tensor<string, []> var_54219_equation_0 = const()[name = tensor<string, []>("op_54219_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_54219_cast_fp16 = einsum(equation = var_54219_equation_0, values = (var_54061_cast_fp16, var_53978_cast_fp16))[name = tensor<string, []>("op_54219_cast_fp16")];
            tensor<fp16, []> var_54220_to_fp16 = const()[name = tensor<string, []>("op_54220_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4633_cast_fp16 = mul(x = var_54219_cast_fp16, y = var_54220_to_fp16)[name = tensor<string, []>("aw_4633_cast_fp16")];
            tensor<string, []> var_54223_equation_0 = const()[name = tensor<string, []>("op_54223_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_54223_cast_fp16 = einsum(equation = var_54223_equation_0, values = (var_54065_cast_fp16, var_53982_cast_fp16))[name = tensor<string, []>("op_54223_cast_fp16")];
            tensor<fp16, []> var_54224_to_fp16 = const()[name = tensor<string, []>("op_54224_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4635_cast_fp16 = mul(x = var_54223_cast_fp16, y = var_54224_to_fp16)[name = tensor<string, []>("aw_4635_cast_fp16")];
            tensor<string, []> var_54227_equation_0 = const()[name = tensor<string, []>("op_54227_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_54227_cast_fp16 = einsum(equation = var_54227_equation_0, values = (var_54069_cast_fp16, var_53986_cast_fp16))[name = tensor<string, []>("op_54227_cast_fp16")];
            tensor<fp16, []> var_54228_to_fp16 = const()[name = tensor<string, []>("op_54228_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4637_cast_fp16 = mul(x = var_54227_cast_fp16, y = var_54228_to_fp16)[name = tensor<string, []>("aw_4637_cast_fp16")];
            tensor<string, []> var_54231_equation_0 = const()[name = tensor<string, []>("op_54231_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_54231_cast_fp16 = einsum(equation = var_54231_equation_0, values = (var_54073_cast_fp16, var_53990_cast_fp16))[name = tensor<string, []>("op_54231_cast_fp16")];
            tensor<fp16, []> var_54232_to_fp16 = const()[name = tensor<string, []>("op_54232_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4639_cast_fp16 = mul(x = var_54231_cast_fp16, y = var_54232_to_fp16)[name = tensor<string, []>("aw_4639_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_54234_cast_fp16 = softmax(axis = var_30355, x = aw_4601_cast_fp16)[name = tensor<string, []>("op_54234_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_54235_cast_fp16 = softmax(axis = var_30355, x = aw_4603_cast_fp16)[name = tensor<string, []>("op_54235_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_54236_cast_fp16 = softmax(axis = var_30355, x = aw_4605_cast_fp16)[name = tensor<string, []>("op_54236_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_54237_cast_fp16 = softmax(axis = var_30355, x = aw_4607_cast_fp16)[name = tensor<string, []>("op_54237_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_54238_cast_fp16 = softmax(axis = var_30355, x = aw_4609_cast_fp16)[name = tensor<string, []>("op_54238_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_54239_cast_fp16 = softmax(axis = var_30355, x = aw_4611_cast_fp16)[name = tensor<string, []>("op_54239_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_54240_cast_fp16 = softmax(axis = var_30355, x = aw_4613_cast_fp16)[name = tensor<string, []>("op_54240_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_54241_cast_fp16 = softmax(axis = var_30355, x = aw_4615_cast_fp16)[name = tensor<string, []>("op_54241_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_54242_cast_fp16 = softmax(axis = var_30355, x = aw_4617_cast_fp16)[name = tensor<string, []>("op_54242_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_54243_cast_fp16 = softmax(axis = var_30355, x = aw_4619_cast_fp16)[name = tensor<string, []>("op_54243_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_54244_cast_fp16 = softmax(axis = var_30355, x = aw_4621_cast_fp16)[name = tensor<string, []>("op_54244_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_54245_cast_fp16 = softmax(axis = var_30355, x = aw_4623_cast_fp16)[name = tensor<string, []>("op_54245_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_54246_cast_fp16 = softmax(axis = var_30355, x = aw_4625_cast_fp16)[name = tensor<string, []>("op_54246_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_54247_cast_fp16 = softmax(axis = var_30355, x = aw_4627_cast_fp16)[name = tensor<string, []>("op_54247_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_54248_cast_fp16 = softmax(axis = var_30355, x = aw_4629_cast_fp16)[name = tensor<string, []>("op_54248_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_54249_cast_fp16 = softmax(axis = var_30355, x = aw_4631_cast_fp16)[name = tensor<string, []>("op_54249_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_54250_cast_fp16 = softmax(axis = var_30355, x = aw_4633_cast_fp16)[name = tensor<string, []>("op_54250_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_54251_cast_fp16 = softmax(axis = var_30355, x = aw_4635_cast_fp16)[name = tensor<string, []>("op_54251_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_54252_cast_fp16 = softmax(axis = var_30355, x = aw_4637_cast_fp16)[name = tensor<string, []>("op_54252_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_54253_cast_fp16 = softmax(axis = var_30355, x = aw_4639_cast_fp16)[name = tensor<string, []>("op_54253_cast_fp16")];
            tensor<string, []> var_54255_equation_0 = const()[name = tensor<string, []>("op_54255_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54255_cast_fp16 = einsum(equation = var_54255_equation_0, values = (var_54075_cast_fp16, var_54234_cast_fp16))[name = tensor<string, []>("op_54255_cast_fp16")];
            tensor<string, []> var_54257_equation_0 = const()[name = tensor<string, []>("op_54257_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54257_cast_fp16 = einsum(equation = var_54257_equation_0, values = (var_54079_cast_fp16, var_54235_cast_fp16))[name = tensor<string, []>("op_54257_cast_fp16")];
            tensor<string, []> var_54259_equation_0 = const()[name = tensor<string, []>("op_54259_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54259_cast_fp16 = einsum(equation = var_54259_equation_0, values = (var_54083_cast_fp16, var_54236_cast_fp16))[name = tensor<string, []>("op_54259_cast_fp16")];
            tensor<string, []> var_54261_equation_0 = const()[name = tensor<string, []>("op_54261_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54261_cast_fp16 = einsum(equation = var_54261_equation_0, values = (var_54087_cast_fp16, var_54237_cast_fp16))[name = tensor<string, []>("op_54261_cast_fp16")];
            tensor<string, []> var_54263_equation_0 = const()[name = tensor<string, []>("op_54263_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54263_cast_fp16 = einsum(equation = var_54263_equation_0, values = (var_54091_cast_fp16, var_54238_cast_fp16))[name = tensor<string, []>("op_54263_cast_fp16")];
            tensor<string, []> var_54265_equation_0 = const()[name = tensor<string, []>("op_54265_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54265_cast_fp16 = einsum(equation = var_54265_equation_0, values = (var_54095_cast_fp16, var_54239_cast_fp16))[name = tensor<string, []>("op_54265_cast_fp16")];
            tensor<string, []> var_54267_equation_0 = const()[name = tensor<string, []>("op_54267_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54267_cast_fp16 = einsum(equation = var_54267_equation_0, values = (var_54099_cast_fp16, var_54240_cast_fp16))[name = tensor<string, []>("op_54267_cast_fp16")];
            tensor<string, []> var_54269_equation_0 = const()[name = tensor<string, []>("op_54269_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54269_cast_fp16 = einsum(equation = var_54269_equation_0, values = (var_54103_cast_fp16, var_54241_cast_fp16))[name = tensor<string, []>("op_54269_cast_fp16")];
            tensor<string, []> var_54271_equation_0 = const()[name = tensor<string, []>("op_54271_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54271_cast_fp16 = einsum(equation = var_54271_equation_0, values = (var_54107_cast_fp16, var_54242_cast_fp16))[name = tensor<string, []>("op_54271_cast_fp16")];
            tensor<string, []> var_54273_equation_0 = const()[name = tensor<string, []>("op_54273_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54273_cast_fp16 = einsum(equation = var_54273_equation_0, values = (var_54111_cast_fp16, var_54243_cast_fp16))[name = tensor<string, []>("op_54273_cast_fp16")];
            tensor<string, []> var_54275_equation_0 = const()[name = tensor<string, []>("op_54275_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54275_cast_fp16 = einsum(equation = var_54275_equation_0, values = (var_54115_cast_fp16, var_54244_cast_fp16))[name = tensor<string, []>("op_54275_cast_fp16")];
            tensor<string, []> var_54277_equation_0 = const()[name = tensor<string, []>("op_54277_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54277_cast_fp16 = einsum(equation = var_54277_equation_0, values = (var_54119_cast_fp16, var_54245_cast_fp16))[name = tensor<string, []>("op_54277_cast_fp16")];
            tensor<string, []> var_54279_equation_0 = const()[name = tensor<string, []>("op_54279_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54279_cast_fp16 = einsum(equation = var_54279_equation_0, values = (var_54123_cast_fp16, var_54246_cast_fp16))[name = tensor<string, []>("op_54279_cast_fp16")];
            tensor<string, []> var_54281_equation_0 = const()[name = tensor<string, []>("op_54281_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54281_cast_fp16 = einsum(equation = var_54281_equation_0, values = (var_54127_cast_fp16, var_54247_cast_fp16))[name = tensor<string, []>("op_54281_cast_fp16")];
            tensor<string, []> var_54283_equation_0 = const()[name = tensor<string, []>("op_54283_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54283_cast_fp16 = einsum(equation = var_54283_equation_0, values = (var_54131_cast_fp16, var_54248_cast_fp16))[name = tensor<string, []>("op_54283_cast_fp16")];
            tensor<string, []> var_54285_equation_0 = const()[name = tensor<string, []>("op_54285_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54285_cast_fp16 = einsum(equation = var_54285_equation_0, values = (var_54135_cast_fp16, var_54249_cast_fp16))[name = tensor<string, []>("op_54285_cast_fp16")];
            tensor<string, []> var_54287_equation_0 = const()[name = tensor<string, []>("op_54287_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54287_cast_fp16 = einsum(equation = var_54287_equation_0, values = (var_54139_cast_fp16, var_54250_cast_fp16))[name = tensor<string, []>("op_54287_cast_fp16")];
            tensor<string, []> var_54289_equation_0 = const()[name = tensor<string, []>("op_54289_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54289_cast_fp16 = einsum(equation = var_54289_equation_0, values = (var_54143_cast_fp16, var_54251_cast_fp16))[name = tensor<string, []>("op_54289_cast_fp16")];
            tensor<string, []> var_54291_equation_0 = const()[name = tensor<string, []>("op_54291_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54291_cast_fp16 = einsum(equation = var_54291_equation_0, values = (var_54147_cast_fp16, var_54252_cast_fp16))[name = tensor<string, []>("op_54291_cast_fp16")];
            tensor<string, []> var_54293_equation_0 = const()[name = tensor<string, []>("op_54293_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54293_cast_fp16 = einsum(equation = var_54293_equation_0, values = (var_54151_cast_fp16, var_54253_cast_fp16))[name = tensor<string, []>("op_54293_cast_fp16")];
            tensor<bool, []> input_677_interleave_0 = const()[name = tensor<string, []>("input_677_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_677_cast_fp16 = concat(axis = var_30355, interleave = input_677_interleave_0, values = (var_54255_cast_fp16, var_54257_cast_fp16, var_54259_cast_fp16, var_54261_cast_fp16, var_54263_cast_fp16, var_54265_cast_fp16, var_54267_cast_fp16, var_54269_cast_fp16, var_54271_cast_fp16, var_54273_cast_fp16, var_54275_cast_fp16, var_54277_cast_fp16, var_54279_cast_fp16, var_54281_cast_fp16, var_54283_cast_fp16, var_54285_cast_fp16, var_54287_cast_fp16, var_54289_cast_fp16, var_54291_cast_fp16, var_54293_cast_fp16))[name = tensor<string, []>("input_677_cast_fp16")];
            tensor<string, []> var_54303_pad_type_0 = const()[name = tensor<string, []>("op_54303_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_54303_strides_0 = const()[name = tensor<string, []>("op_54303_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_54303_pad_0 = const()[name = tensor<string, []>("op_54303_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_54303_dilations_0 = const()[name = tensor<string, []>("op_54303_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_54303_groups_0 = const()[name = tensor<string, []>("op_54303_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_5_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(738820992))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(740049856))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_5_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_5_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_5_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(740050048)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_54303_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_5_attn2_to_out_0_bias_to_fp16, dilations = var_54303_dilations_0, groups = var_54303_groups_0, pad = var_54303_pad_0, pad_type = var_54303_pad_type_0, strides = var_54303_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_5_attn2_to_out_0_weight_to_fp16_palettized, x = input_677_cast_fp16)[name = tensor<string, []>("op_54303_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_359_cast_fp16 = add(x = var_54303_cast_fp16, y = inputs_357_cast_fp16)[name = tensor<string, []>("inputs_359_cast_fp16")];
            tensor<int32, [1]> input_679_axes_0 = const()[name = tensor<string, []>("input_679_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> input_679_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_679_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(740052672)))];
            tensor<fp16, [1280]> input_679_beta_0_to_fp16 = const()[name = tensor<string, []>("input_679_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(740055296)))];
            tensor<fp16, []> var_54313_to_fp16 = const()[name = tensor<string, []>("op_54313_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> input_679_cast_fp16 = layer_norm(axes = input_679_axes_0, beta = input_679_beta_0_to_fp16, epsilon = var_54313_to_fp16, gamma = input_679_gamma_0_to_fp16, x = inputs_359_cast_fp16)[name = tensor<string, []>("input_679_cast_fp16")];
            tensor<string, []> var_54333_pad_type_0 = const()[name = tensor<string, []>("op_54333_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_54333_strides_0 = const()[name = tensor<string, []>("op_54333_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_54333_pad_0 = const()[name = tensor<string, []>("op_54333_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_54333_dilations_0 = const()[name = tensor<string, []>("op_54333_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_54333_groups_0 = const()[name = tensor<string, []>("op_54333_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_5_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [9830400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(740057920))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(749888384))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_5_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([10240, 1280, 1, 1])];
            tensor<fp16, [10240]> up_blocks_0_attentions_2_transformer_blocks_5_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_5_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(749888576)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_54333_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_5_ff_net_0_proj_bias_to_fp16, dilations = var_54333_dilations_0, groups = var_54333_groups_0, pad = var_54333_pad_0, pad_type = var_54333_pad_type_0, strides = var_54333_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_5_ff_net_0_proj_weight_to_fp16_palettized, x = input_679_cast_fp16)[name = tensor<string, []>("op_54333_cast_fp16")];
            tensor<int32, [2]> var_54334_split_sizes_0 = const()[name = tensor<string, []>("op_54334_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_54334_axis_0 = const()[name = tensor<string, []>("op_54334_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_54334_cast_fp16_0, tensor<fp16, [2, 5120, 1, 1024]> var_54334_cast_fp16_1 = split(axis = var_54334_axis_0, split_sizes = var_54334_split_sizes_0, x = var_54333_cast_fp16)[name = tensor<string, []>("op_54334_cast_fp16")];
            tensor<string, []> var_54336_mode_0 = const()[name = tensor<string, []>("op_54336_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_54336_cast_fp16 = gelu(mode = var_54336_mode_0, x = var_54334_cast_fp16_1)[name = tensor<string, []>("op_54336_cast_fp16")];
            tensor<fp16, [2, 5120, 1, 1024]> input_681_cast_fp16 = mul(x = var_54334_cast_fp16_0, y = var_54336_cast_fp16)[name = tensor<string, []>("input_681_cast_fp16")];
            tensor<string, []> var_54344_pad_type_0 = const()[name = tensor<string, []>("op_54344_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_54344_strides_0 = const()[name = tensor<string, []>("op_54344_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_54344_pad_0 = const()[name = tensor<string, []>("op_54344_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_54344_dilations_0 = const()[name = tensor<string, []>("op_54344_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_54344_groups_0 = const()[name = tensor<string, []>("op_54344_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_5_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [4915200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(749909120))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(754824384))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_5_ff_net_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 5120, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_5_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_5_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(754824576)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_54344_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_5_ff_net_2_bias_to_fp16, dilations = var_54344_dilations_0, groups = var_54344_groups_0, pad = var_54344_pad_0, pad_type = var_54344_pad_type_0, strides = var_54344_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_5_ff_net_2_weight_to_fp16_palettized, x = input_681_cast_fp16)[name = tensor<string, []>("op_54344_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_361_cast_fp16 = add(x = var_54344_cast_fp16, y = inputs_359_cast_fp16)[name = tensor<string, []>("inputs_361_cast_fp16")];
            tensor<int32, [1]> hidden_states_471_axes_0 = const()[name = tensor<string, []>("hidden_states_471_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_471_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_471_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(754827200)))];
            tensor<fp16, [1280]> hidden_states_471_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_471_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(754829824)))];
            tensor<fp16, []> var_54360_to_fp16 = const()[name = tensor<string, []>("op_54360_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_471_cast_fp16 = layer_norm(axes = hidden_states_471_axes_0, beta = hidden_states_471_beta_0_to_fp16, epsilon = var_54360_to_fp16, gamma = hidden_states_471_gamma_0_to_fp16, x = inputs_361_cast_fp16)[name = tensor<string, []>("hidden_states_471_cast_fp16")];
            tensor<string, []> q_241_pad_type_0 = const()[name = tensor<string, []>("q_241_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_241_strides_0 = const()[name = tensor<string, []>("q_241_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_241_pad_0 = const()[name = tensor<string, []>("q_241_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_241_dilations_0 = const()[name = tensor<string, []>("q_241_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_241_groups_0 = const()[name = tensor<string, []>("q_241_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_6_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(754832448))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(756061312))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_6_attn1_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_241_cast_fp16 = conv(dilations = q_241_dilations_0, groups = q_241_groups_0, pad = q_241_pad_0, pad_type = q_241_pad_type_0, strides = q_241_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_6_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_471_cast_fp16)[name = tensor<string, []>("q_241_cast_fp16")];
            tensor<string, []> k_481_pad_type_0 = const()[name = tensor<string, []>("k_481_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_481_strides_0 = const()[name = tensor<string, []>("k_481_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_481_pad_0 = const()[name = tensor<string, []>("k_481_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_481_dilations_0 = const()[name = tensor<string, []>("k_481_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_481_groups_0 = const()[name = tensor<string, []>("k_481_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_6_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(756061504))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(757290368))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_6_attn1_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> k_481_cast_fp16 = conv(dilations = k_481_dilations_0, groups = k_481_groups_0, pad = k_481_pad_0, pad_type = k_481_pad_type_0, strides = k_481_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_6_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_471_cast_fp16)[name = tensor<string, []>("k_481_cast_fp16")];
            tensor<string, []> v_241_pad_type_0 = const()[name = tensor<string, []>("v_241_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_241_strides_0 = const()[name = tensor<string, []>("v_241_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_241_pad_0 = const()[name = tensor<string, []>("v_241_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_241_dilations_0 = const()[name = tensor<string, []>("v_241_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_241_groups_0 = const()[name = tensor<string, []>("v_241_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_6_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(757290560))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(758519424))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_6_attn1_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> v_241_cast_fp16 = conv(dilations = v_241_dilations_0, groups = v_241_groups_0, pad = v_241_pad_0, pad_type = v_241_pad_type_0, strides = v_241_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_6_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_471_cast_fp16)[name = tensor<string, []>("v_241_cast_fp16")];
            tensor<int32, [4]> var_54393_begin_0 = const()[name = tensor<string, []>("op_54393_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_54393_end_0 = const()[name = tensor<string, []>("op_54393_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_54393_end_mask_0 = const()[name = tensor<string, []>("op_54393_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54393_cast_fp16 = slice_by_index(begin = var_54393_begin_0, end = var_54393_end_0, end_mask = var_54393_end_mask_0, x = q_241_cast_fp16)[name = tensor<string, []>("op_54393_cast_fp16")];
            tensor<int32, [4]> var_54397_begin_0 = const()[name = tensor<string, []>("op_54397_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_54397_end_0 = const()[name = tensor<string, []>("op_54397_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_54397_end_mask_0 = const()[name = tensor<string, []>("op_54397_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54397_cast_fp16 = slice_by_index(begin = var_54397_begin_0, end = var_54397_end_0, end_mask = var_54397_end_mask_0, x = q_241_cast_fp16)[name = tensor<string, []>("op_54397_cast_fp16")];
            tensor<int32, [4]> var_54401_begin_0 = const()[name = tensor<string, []>("op_54401_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_54401_end_0 = const()[name = tensor<string, []>("op_54401_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_54401_end_mask_0 = const()[name = tensor<string, []>("op_54401_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54401_cast_fp16 = slice_by_index(begin = var_54401_begin_0, end = var_54401_end_0, end_mask = var_54401_end_mask_0, x = q_241_cast_fp16)[name = tensor<string, []>("op_54401_cast_fp16")];
            tensor<int32, [4]> var_54405_begin_0 = const()[name = tensor<string, []>("op_54405_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_54405_end_0 = const()[name = tensor<string, []>("op_54405_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_54405_end_mask_0 = const()[name = tensor<string, []>("op_54405_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54405_cast_fp16 = slice_by_index(begin = var_54405_begin_0, end = var_54405_end_0, end_mask = var_54405_end_mask_0, x = q_241_cast_fp16)[name = tensor<string, []>("op_54405_cast_fp16")];
            tensor<int32, [4]> var_54409_begin_0 = const()[name = tensor<string, []>("op_54409_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_54409_end_0 = const()[name = tensor<string, []>("op_54409_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_54409_end_mask_0 = const()[name = tensor<string, []>("op_54409_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54409_cast_fp16 = slice_by_index(begin = var_54409_begin_0, end = var_54409_end_0, end_mask = var_54409_end_mask_0, x = q_241_cast_fp16)[name = tensor<string, []>("op_54409_cast_fp16")];
            tensor<int32, [4]> var_54413_begin_0 = const()[name = tensor<string, []>("op_54413_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_54413_end_0 = const()[name = tensor<string, []>("op_54413_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_54413_end_mask_0 = const()[name = tensor<string, []>("op_54413_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54413_cast_fp16 = slice_by_index(begin = var_54413_begin_0, end = var_54413_end_0, end_mask = var_54413_end_mask_0, x = q_241_cast_fp16)[name = tensor<string, []>("op_54413_cast_fp16")];
            tensor<int32, [4]> var_54417_begin_0 = const()[name = tensor<string, []>("op_54417_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_54417_end_0 = const()[name = tensor<string, []>("op_54417_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_54417_end_mask_0 = const()[name = tensor<string, []>("op_54417_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54417_cast_fp16 = slice_by_index(begin = var_54417_begin_0, end = var_54417_end_0, end_mask = var_54417_end_mask_0, x = q_241_cast_fp16)[name = tensor<string, []>("op_54417_cast_fp16")];
            tensor<int32, [4]> var_54421_begin_0 = const()[name = tensor<string, []>("op_54421_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_54421_end_0 = const()[name = tensor<string, []>("op_54421_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_54421_end_mask_0 = const()[name = tensor<string, []>("op_54421_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54421_cast_fp16 = slice_by_index(begin = var_54421_begin_0, end = var_54421_end_0, end_mask = var_54421_end_mask_0, x = q_241_cast_fp16)[name = tensor<string, []>("op_54421_cast_fp16")];
            tensor<int32, [4]> var_54425_begin_0 = const()[name = tensor<string, []>("op_54425_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_54425_end_0 = const()[name = tensor<string, []>("op_54425_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_54425_end_mask_0 = const()[name = tensor<string, []>("op_54425_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54425_cast_fp16 = slice_by_index(begin = var_54425_begin_0, end = var_54425_end_0, end_mask = var_54425_end_mask_0, x = q_241_cast_fp16)[name = tensor<string, []>("op_54425_cast_fp16")];
            tensor<int32, [4]> var_54429_begin_0 = const()[name = tensor<string, []>("op_54429_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_54429_end_0 = const()[name = tensor<string, []>("op_54429_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_54429_end_mask_0 = const()[name = tensor<string, []>("op_54429_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54429_cast_fp16 = slice_by_index(begin = var_54429_begin_0, end = var_54429_end_0, end_mask = var_54429_end_mask_0, x = q_241_cast_fp16)[name = tensor<string, []>("op_54429_cast_fp16")];
            tensor<int32, [4]> var_54433_begin_0 = const()[name = tensor<string, []>("op_54433_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_54433_end_0 = const()[name = tensor<string, []>("op_54433_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_54433_end_mask_0 = const()[name = tensor<string, []>("op_54433_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54433_cast_fp16 = slice_by_index(begin = var_54433_begin_0, end = var_54433_end_0, end_mask = var_54433_end_mask_0, x = q_241_cast_fp16)[name = tensor<string, []>("op_54433_cast_fp16")];
            tensor<int32, [4]> var_54437_begin_0 = const()[name = tensor<string, []>("op_54437_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_54437_end_0 = const()[name = tensor<string, []>("op_54437_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_54437_end_mask_0 = const()[name = tensor<string, []>("op_54437_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54437_cast_fp16 = slice_by_index(begin = var_54437_begin_0, end = var_54437_end_0, end_mask = var_54437_end_mask_0, x = q_241_cast_fp16)[name = tensor<string, []>("op_54437_cast_fp16")];
            tensor<int32, [4]> var_54441_begin_0 = const()[name = tensor<string, []>("op_54441_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_54441_end_0 = const()[name = tensor<string, []>("op_54441_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_54441_end_mask_0 = const()[name = tensor<string, []>("op_54441_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54441_cast_fp16 = slice_by_index(begin = var_54441_begin_0, end = var_54441_end_0, end_mask = var_54441_end_mask_0, x = q_241_cast_fp16)[name = tensor<string, []>("op_54441_cast_fp16")];
            tensor<int32, [4]> var_54445_begin_0 = const()[name = tensor<string, []>("op_54445_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_54445_end_0 = const()[name = tensor<string, []>("op_54445_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_54445_end_mask_0 = const()[name = tensor<string, []>("op_54445_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54445_cast_fp16 = slice_by_index(begin = var_54445_begin_0, end = var_54445_end_0, end_mask = var_54445_end_mask_0, x = q_241_cast_fp16)[name = tensor<string, []>("op_54445_cast_fp16")];
            tensor<int32, [4]> var_54449_begin_0 = const()[name = tensor<string, []>("op_54449_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_54449_end_0 = const()[name = tensor<string, []>("op_54449_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_54449_end_mask_0 = const()[name = tensor<string, []>("op_54449_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54449_cast_fp16 = slice_by_index(begin = var_54449_begin_0, end = var_54449_end_0, end_mask = var_54449_end_mask_0, x = q_241_cast_fp16)[name = tensor<string, []>("op_54449_cast_fp16")];
            tensor<int32, [4]> var_54453_begin_0 = const()[name = tensor<string, []>("op_54453_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_54453_end_0 = const()[name = tensor<string, []>("op_54453_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_54453_end_mask_0 = const()[name = tensor<string, []>("op_54453_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54453_cast_fp16 = slice_by_index(begin = var_54453_begin_0, end = var_54453_end_0, end_mask = var_54453_end_mask_0, x = q_241_cast_fp16)[name = tensor<string, []>("op_54453_cast_fp16")];
            tensor<int32, [4]> var_54457_begin_0 = const()[name = tensor<string, []>("op_54457_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_54457_end_0 = const()[name = tensor<string, []>("op_54457_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_54457_end_mask_0 = const()[name = tensor<string, []>("op_54457_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54457_cast_fp16 = slice_by_index(begin = var_54457_begin_0, end = var_54457_end_0, end_mask = var_54457_end_mask_0, x = q_241_cast_fp16)[name = tensor<string, []>("op_54457_cast_fp16")];
            tensor<int32, [4]> var_54461_begin_0 = const()[name = tensor<string, []>("op_54461_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_54461_end_0 = const()[name = tensor<string, []>("op_54461_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_54461_end_mask_0 = const()[name = tensor<string, []>("op_54461_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54461_cast_fp16 = slice_by_index(begin = var_54461_begin_0, end = var_54461_end_0, end_mask = var_54461_end_mask_0, x = q_241_cast_fp16)[name = tensor<string, []>("op_54461_cast_fp16")];
            tensor<int32, [4]> var_54465_begin_0 = const()[name = tensor<string, []>("op_54465_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_54465_end_0 = const()[name = tensor<string, []>("op_54465_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_54465_end_mask_0 = const()[name = tensor<string, []>("op_54465_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54465_cast_fp16 = slice_by_index(begin = var_54465_begin_0, end = var_54465_end_0, end_mask = var_54465_end_mask_0, x = q_241_cast_fp16)[name = tensor<string, []>("op_54465_cast_fp16")];
            tensor<int32, [4]> var_54469_begin_0 = const()[name = tensor<string, []>("op_54469_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_54469_end_0 = const()[name = tensor<string, []>("op_54469_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_54469_end_mask_0 = const()[name = tensor<string, []>("op_54469_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54469_cast_fp16 = slice_by_index(begin = var_54469_begin_0, end = var_54469_end_0, end_mask = var_54469_end_mask_0, x = q_241_cast_fp16)[name = tensor<string, []>("op_54469_cast_fp16")];
            tensor<int32, [4]> k_483_perm_0 = const()[name = tensor<string, []>("k_483_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_54476_begin_0 = const()[name = tensor<string, []>("op_54476_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_54476_end_0 = const()[name = tensor<string, []>("op_54476_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_54476_end_mask_0 = const()[name = tensor<string, []>("op_54476_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> k_483_cast_fp16 = transpose(perm = k_483_perm_0, x = k_481_cast_fp16)[name = tensor<string, []>("transpose_19")];
            tensor<fp16, [2, 1024, 1, 64]> var_54476_cast_fp16 = slice_by_index(begin = var_54476_begin_0, end = var_54476_end_0, end_mask = var_54476_end_mask_0, x = k_483_cast_fp16)[name = tensor<string, []>("op_54476_cast_fp16")];
            tensor<int32, [4]> var_54480_begin_0 = const()[name = tensor<string, []>("op_54480_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_54480_end_0 = const()[name = tensor<string, []>("op_54480_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_54480_end_mask_0 = const()[name = tensor<string, []>("op_54480_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_54480_cast_fp16 = slice_by_index(begin = var_54480_begin_0, end = var_54480_end_0, end_mask = var_54480_end_mask_0, x = k_483_cast_fp16)[name = tensor<string, []>("op_54480_cast_fp16")];
            tensor<int32, [4]> var_54484_begin_0 = const()[name = tensor<string, []>("op_54484_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_54484_end_0 = const()[name = tensor<string, []>("op_54484_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_54484_end_mask_0 = const()[name = tensor<string, []>("op_54484_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_54484_cast_fp16 = slice_by_index(begin = var_54484_begin_0, end = var_54484_end_0, end_mask = var_54484_end_mask_0, x = k_483_cast_fp16)[name = tensor<string, []>("op_54484_cast_fp16")];
            tensor<int32, [4]> var_54488_begin_0 = const()[name = tensor<string, []>("op_54488_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_54488_end_0 = const()[name = tensor<string, []>("op_54488_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_54488_end_mask_0 = const()[name = tensor<string, []>("op_54488_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_54488_cast_fp16 = slice_by_index(begin = var_54488_begin_0, end = var_54488_end_0, end_mask = var_54488_end_mask_0, x = k_483_cast_fp16)[name = tensor<string, []>("op_54488_cast_fp16")];
            tensor<int32, [4]> var_54492_begin_0 = const()[name = tensor<string, []>("op_54492_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_54492_end_0 = const()[name = tensor<string, []>("op_54492_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_54492_end_mask_0 = const()[name = tensor<string, []>("op_54492_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_54492_cast_fp16 = slice_by_index(begin = var_54492_begin_0, end = var_54492_end_0, end_mask = var_54492_end_mask_0, x = k_483_cast_fp16)[name = tensor<string, []>("op_54492_cast_fp16")];
            tensor<int32, [4]> var_54496_begin_0 = const()[name = tensor<string, []>("op_54496_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_54496_end_0 = const()[name = tensor<string, []>("op_54496_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_54496_end_mask_0 = const()[name = tensor<string, []>("op_54496_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_54496_cast_fp16 = slice_by_index(begin = var_54496_begin_0, end = var_54496_end_0, end_mask = var_54496_end_mask_0, x = k_483_cast_fp16)[name = tensor<string, []>("op_54496_cast_fp16")];
            tensor<int32, [4]> var_54500_begin_0 = const()[name = tensor<string, []>("op_54500_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_54500_end_0 = const()[name = tensor<string, []>("op_54500_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_54500_end_mask_0 = const()[name = tensor<string, []>("op_54500_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_54500_cast_fp16 = slice_by_index(begin = var_54500_begin_0, end = var_54500_end_0, end_mask = var_54500_end_mask_0, x = k_483_cast_fp16)[name = tensor<string, []>("op_54500_cast_fp16")];
            tensor<int32, [4]> var_54504_begin_0 = const()[name = tensor<string, []>("op_54504_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_54504_end_0 = const()[name = tensor<string, []>("op_54504_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_54504_end_mask_0 = const()[name = tensor<string, []>("op_54504_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_54504_cast_fp16 = slice_by_index(begin = var_54504_begin_0, end = var_54504_end_0, end_mask = var_54504_end_mask_0, x = k_483_cast_fp16)[name = tensor<string, []>("op_54504_cast_fp16")];
            tensor<int32, [4]> var_54508_begin_0 = const()[name = tensor<string, []>("op_54508_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_54508_end_0 = const()[name = tensor<string, []>("op_54508_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_54508_end_mask_0 = const()[name = tensor<string, []>("op_54508_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_54508_cast_fp16 = slice_by_index(begin = var_54508_begin_0, end = var_54508_end_0, end_mask = var_54508_end_mask_0, x = k_483_cast_fp16)[name = tensor<string, []>("op_54508_cast_fp16")];
            tensor<int32, [4]> var_54512_begin_0 = const()[name = tensor<string, []>("op_54512_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_54512_end_0 = const()[name = tensor<string, []>("op_54512_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_54512_end_mask_0 = const()[name = tensor<string, []>("op_54512_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_54512_cast_fp16 = slice_by_index(begin = var_54512_begin_0, end = var_54512_end_0, end_mask = var_54512_end_mask_0, x = k_483_cast_fp16)[name = tensor<string, []>("op_54512_cast_fp16")];
            tensor<int32, [4]> var_54516_begin_0 = const()[name = tensor<string, []>("op_54516_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_54516_end_0 = const()[name = tensor<string, []>("op_54516_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_54516_end_mask_0 = const()[name = tensor<string, []>("op_54516_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_54516_cast_fp16 = slice_by_index(begin = var_54516_begin_0, end = var_54516_end_0, end_mask = var_54516_end_mask_0, x = k_483_cast_fp16)[name = tensor<string, []>("op_54516_cast_fp16")];
            tensor<int32, [4]> var_54520_begin_0 = const()[name = tensor<string, []>("op_54520_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_54520_end_0 = const()[name = tensor<string, []>("op_54520_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_54520_end_mask_0 = const()[name = tensor<string, []>("op_54520_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_54520_cast_fp16 = slice_by_index(begin = var_54520_begin_0, end = var_54520_end_0, end_mask = var_54520_end_mask_0, x = k_483_cast_fp16)[name = tensor<string, []>("op_54520_cast_fp16")];
            tensor<int32, [4]> var_54524_begin_0 = const()[name = tensor<string, []>("op_54524_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_54524_end_0 = const()[name = tensor<string, []>("op_54524_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_54524_end_mask_0 = const()[name = tensor<string, []>("op_54524_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_54524_cast_fp16 = slice_by_index(begin = var_54524_begin_0, end = var_54524_end_0, end_mask = var_54524_end_mask_0, x = k_483_cast_fp16)[name = tensor<string, []>("op_54524_cast_fp16")];
            tensor<int32, [4]> var_54528_begin_0 = const()[name = tensor<string, []>("op_54528_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_54528_end_0 = const()[name = tensor<string, []>("op_54528_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_54528_end_mask_0 = const()[name = tensor<string, []>("op_54528_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_54528_cast_fp16 = slice_by_index(begin = var_54528_begin_0, end = var_54528_end_0, end_mask = var_54528_end_mask_0, x = k_483_cast_fp16)[name = tensor<string, []>("op_54528_cast_fp16")];
            tensor<int32, [4]> var_54532_begin_0 = const()[name = tensor<string, []>("op_54532_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_54532_end_0 = const()[name = tensor<string, []>("op_54532_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_54532_end_mask_0 = const()[name = tensor<string, []>("op_54532_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_54532_cast_fp16 = slice_by_index(begin = var_54532_begin_0, end = var_54532_end_0, end_mask = var_54532_end_mask_0, x = k_483_cast_fp16)[name = tensor<string, []>("op_54532_cast_fp16")];
            tensor<int32, [4]> var_54536_begin_0 = const()[name = tensor<string, []>("op_54536_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_54536_end_0 = const()[name = tensor<string, []>("op_54536_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_54536_end_mask_0 = const()[name = tensor<string, []>("op_54536_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_54536_cast_fp16 = slice_by_index(begin = var_54536_begin_0, end = var_54536_end_0, end_mask = var_54536_end_mask_0, x = k_483_cast_fp16)[name = tensor<string, []>("op_54536_cast_fp16")];
            tensor<int32, [4]> var_54540_begin_0 = const()[name = tensor<string, []>("op_54540_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_54540_end_0 = const()[name = tensor<string, []>("op_54540_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_54540_end_mask_0 = const()[name = tensor<string, []>("op_54540_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_54540_cast_fp16 = slice_by_index(begin = var_54540_begin_0, end = var_54540_end_0, end_mask = var_54540_end_mask_0, x = k_483_cast_fp16)[name = tensor<string, []>("op_54540_cast_fp16")];
            tensor<int32, [4]> var_54544_begin_0 = const()[name = tensor<string, []>("op_54544_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_54544_end_0 = const()[name = tensor<string, []>("op_54544_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_54544_end_mask_0 = const()[name = tensor<string, []>("op_54544_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_54544_cast_fp16 = slice_by_index(begin = var_54544_begin_0, end = var_54544_end_0, end_mask = var_54544_end_mask_0, x = k_483_cast_fp16)[name = tensor<string, []>("op_54544_cast_fp16")];
            tensor<int32, [4]> var_54548_begin_0 = const()[name = tensor<string, []>("op_54548_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_54548_end_0 = const()[name = tensor<string, []>("op_54548_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_54548_end_mask_0 = const()[name = tensor<string, []>("op_54548_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_54548_cast_fp16 = slice_by_index(begin = var_54548_begin_0, end = var_54548_end_0, end_mask = var_54548_end_mask_0, x = k_483_cast_fp16)[name = tensor<string, []>("op_54548_cast_fp16")];
            tensor<int32, [4]> var_54552_begin_0 = const()[name = tensor<string, []>("op_54552_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_54552_end_0 = const()[name = tensor<string, []>("op_54552_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_54552_end_mask_0 = const()[name = tensor<string, []>("op_54552_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_54552_cast_fp16 = slice_by_index(begin = var_54552_begin_0, end = var_54552_end_0, end_mask = var_54552_end_mask_0, x = k_483_cast_fp16)[name = tensor<string, []>("op_54552_cast_fp16")];
            tensor<int32, [4]> var_54554_begin_0 = const()[name = tensor<string, []>("op_54554_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_54554_end_0 = const()[name = tensor<string, []>("op_54554_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_54554_end_mask_0 = const()[name = tensor<string, []>("op_54554_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54554_cast_fp16 = slice_by_index(begin = var_54554_begin_0, end = var_54554_end_0, end_mask = var_54554_end_mask_0, x = v_241_cast_fp16)[name = tensor<string, []>("op_54554_cast_fp16")];
            tensor<int32, [4]> var_54558_begin_0 = const()[name = tensor<string, []>("op_54558_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_54558_end_0 = const()[name = tensor<string, []>("op_54558_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_54558_end_mask_0 = const()[name = tensor<string, []>("op_54558_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54558_cast_fp16 = slice_by_index(begin = var_54558_begin_0, end = var_54558_end_0, end_mask = var_54558_end_mask_0, x = v_241_cast_fp16)[name = tensor<string, []>("op_54558_cast_fp16")];
            tensor<int32, [4]> var_54562_begin_0 = const()[name = tensor<string, []>("op_54562_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_54562_end_0 = const()[name = tensor<string, []>("op_54562_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_54562_end_mask_0 = const()[name = tensor<string, []>("op_54562_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54562_cast_fp16 = slice_by_index(begin = var_54562_begin_0, end = var_54562_end_0, end_mask = var_54562_end_mask_0, x = v_241_cast_fp16)[name = tensor<string, []>("op_54562_cast_fp16")];
            tensor<int32, [4]> var_54566_begin_0 = const()[name = tensor<string, []>("op_54566_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_54566_end_0 = const()[name = tensor<string, []>("op_54566_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_54566_end_mask_0 = const()[name = tensor<string, []>("op_54566_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54566_cast_fp16 = slice_by_index(begin = var_54566_begin_0, end = var_54566_end_0, end_mask = var_54566_end_mask_0, x = v_241_cast_fp16)[name = tensor<string, []>("op_54566_cast_fp16")];
            tensor<int32, [4]> var_54570_begin_0 = const()[name = tensor<string, []>("op_54570_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_54570_end_0 = const()[name = tensor<string, []>("op_54570_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_54570_end_mask_0 = const()[name = tensor<string, []>("op_54570_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54570_cast_fp16 = slice_by_index(begin = var_54570_begin_0, end = var_54570_end_0, end_mask = var_54570_end_mask_0, x = v_241_cast_fp16)[name = tensor<string, []>("op_54570_cast_fp16")];
            tensor<int32, [4]> var_54574_begin_0 = const()[name = tensor<string, []>("op_54574_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_54574_end_0 = const()[name = tensor<string, []>("op_54574_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_54574_end_mask_0 = const()[name = tensor<string, []>("op_54574_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54574_cast_fp16 = slice_by_index(begin = var_54574_begin_0, end = var_54574_end_0, end_mask = var_54574_end_mask_0, x = v_241_cast_fp16)[name = tensor<string, []>("op_54574_cast_fp16")];
            tensor<int32, [4]> var_54578_begin_0 = const()[name = tensor<string, []>("op_54578_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_54578_end_0 = const()[name = tensor<string, []>("op_54578_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_54578_end_mask_0 = const()[name = tensor<string, []>("op_54578_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54578_cast_fp16 = slice_by_index(begin = var_54578_begin_0, end = var_54578_end_0, end_mask = var_54578_end_mask_0, x = v_241_cast_fp16)[name = tensor<string, []>("op_54578_cast_fp16")];
            tensor<int32, [4]> var_54582_begin_0 = const()[name = tensor<string, []>("op_54582_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_54582_end_0 = const()[name = tensor<string, []>("op_54582_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_54582_end_mask_0 = const()[name = tensor<string, []>("op_54582_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54582_cast_fp16 = slice_by_index(begin = var_54582_begin_0, end = var_54582_end_0, end_mask = var_54582_end_mask_0, x = v_241_cast_fp16)[name = tensor<string, []>("op_54582_cast_fp16")];
            tensor<int32, [4]> var_54586_begin_0 = const()[name = tensor<string, []>("op_54586_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_54586_end_0 = const()[name = tensor<string, []>("op_54586_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_54586_end_mask_0 = const()[name = tensor<string, []>("op_54586_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54586_cast_fp16 = slice_by_index(begin = var_54586_begin_0, end = var_54586_end_0, end_mask = var_54586_end_mask_0, x = v_241_cast_fp16)[name = tensor<string, []>("op_54586_cast_fp16")];
            tensor<int32, [4]> var_54590_begin_0 = const()[name = tensor<string, []>("op_54590_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_54590_end_0 = const()[name = tensor<string, []>("op_54590_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_54590_end_mask_0 = const()[name = tensor<string, []>("op_54590_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54590_cast_fp16 = slice_by_index(begin = var_54590_begin_0, end = var_54590_end_0, end_mask = var_54590_end_mask_0, x = v_241_cast_fp16)[name = tensor<string, []>("op_54590_cast_fp16")];
            tensor<int32, [4]> var_54594_begin_0 = const()[name = tensor<string, []>("op_54594_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_54594_end_0 = const()[name = tensor<string, []>("op_54594_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_54594_end_mask_0 = const()[name = tensor<string, []>("op_54594_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54594_cast_fp16 = slice_by_index(begin = var_54594_begin_0, end = var_54594_end_0, end_mask = var_54594_end_mask_0, x = v_241_cast_fp16)[name = tensor<string, []>("op_54594_cast_fp16")];
            tensor<int32, [4]> var_54598_begin_0 = const()[name = tensor<string, []>("op_54598_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_54598_end_0 = const()[name = tensor<string, []>("op_54598_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_54598_end_mask_0 = const()[name = tensor<string, []>("op_54598_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54598_cast_fp16 = slice_by_index(begin = var_54598_begin_0, end = var_54598_end_0, end_mask = var_54598_end_mask_0, x = v_241_cast_fp16)[name = tensor<string, []>("op_54598_cast_fp16")];
            tensor<int32, [4]> var_54602_begin_0 = const()[name = tensor<string, []>("op_54602_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_54602_end_0 = const()[name = tensor<string, []>("op_54602_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_54602_end_mask_0 = const()[name = tensor<string, []>("op_54602_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54602_cast_fp16 = slice_by_index(begin = var_54602_begin_0, end = var_54602_end_0, end_mask = var_54602_end_mask_0, x = v_241_cast_fp16)[name = tensor<string, []>("op_54602_cast_fp16")];
            tensor<int32, [4]> var_54606_begin_0 = const()[name = tensor<string, []>("op_54606_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_54606_end_0 = const()[name = tensor<string, []>("op_54606_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_54606_end_mask_0 = const()[name = tensor<string, []>("op_54606_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54606_cast_fp16 = slice_by_index(begin = var_54606_begin_0, end = var_54606_end_0, end_mask = var_54606_end_mask_0, x = v_241_cast_fp16)[name = tensor<string, []>("op_54606_cast_fp16")];
            tensor<int32, [4]> var_54610_begin_0 = const()[name = tensor<string, []>("op_54610_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_54610_end_0 = const()[name = tensor<string, []>("op_54610_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_54610_end_mask_0 = const()[name = tensor<string, []>("op_54610_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54610_cast_fp16 = slice_by_index(begin = var_54610_begin_0, end = var_54610_end_0, end_mask = var_54610_end_mask_0, x = v_241_cast_fp16)[name = tensor<string, []>("op_54610_cast_fp16")];
            tensor<int32, [4]> var_54614_begin_0 = const()[name = tensor<string, []>("op_54614_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_54614_end_0 = const()[name = tensor<string, []>("op_54614_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_54614_end_mask_0 = const()[name = tensor<string, []>("op_54614_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54614_cast_fp16 = slice_by_index(begin = var_54614_begin_0, end = var_54614_end_0, end_mask = var_54614_end_mask_0, x = v_241_cast_fp16)[name = tensor<string, []>("op_54614_cast_fp16")];
            tensor<int32, [4]> var_54618_begin_0 = const()[name = tensor<string, []>("op_54618_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_54618_end_0 = const()[name = tensor<string, []>("op_54618_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_54618_end_mask_0 = const()[name = tensor<string, []>("op_54618_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54618_cast_fp16 = slice_by_index(begin = var_54618_begin_0, end = var_54618_end_0, end_mask = var_54618_end_mask_0, x = v_241_cast_fp16)[name = tensor<string, []>("op_54618_cast_fp16")];
            tensor<int32, [4]> var_54622_begin_0 = const()[name = tensor<string, []>("op_54622_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_54622_end_0 = const()[name = tensor<string, []>("op_54622_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_54622_end_mask_0 = const()[name = tensor<string, []>("op_54622_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54622_cast_fp16 = slice_by_index(begin = var_54622_begin_0, end = var_54622_end_0, end_mask = var_54622_end_mask_0, x = v_241_cast_fp16)[name = tensor<string, []>("op_54622_cast_fp16")];
            tensor<int32, [4]> var_54626_begin_0 = const()[name = tensor<string, []>("op_54626_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_54626_end_0 = const()[name = tensor<string, []>("op_54626_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_54626_end_mask_0 = const()[name = tensor<string, []>("op_54626_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54626_cast_fp16 = slice_by_index(begin = var_54626_begin_0, end = var_54626_end_0, end_mask = var_54626_end_mask_0, x = v_241_cast_fp16)[name = tensor<string, []>("op_54626_cast_fp16")];
            tensor<int32, [4]> var_54630_begin_0 = const()[name = tensor<string, []>("op_54630_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_54630_end_0 = const()[name = tensor<string, []>("op_54630_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_54630_end_mask_0 = const()[name = tensor<string, []>("op_54630_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54630_cast_fp16 = slice_by_index(begin = var_54630_begin_0, end = var_54630_end_0, end_mask = var_54630_end_mask_0, x = v_241_cast_fp16)[name = tensor<string, []>("op_54630_cast_fp16")];
            tensor<string, []> var_54634_equation_0 = const()[name = tensor<string, []>("op_54634_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54634_cast_fp16 = einsum(equation = var_54634_equation_0, values = (var_54476_cast_fp16, var_54393_cast_fp16))[name = tensor<string, []>("op_54634_cast_fp16")];
            tensor<fp16, []> var_54635_to_fp16 = const()[name = tensor<string, []>("op_54635_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4641_cast_fp16 = mul(x = var_54634_cast_fp16, y = var_54635_to_fp16)[name = tensor<string, []>("aw_4641_cast_fp16")];
            tensor<string, []> var_54638_equation_0 = const()[name = tensor<string, []>("op_54638_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54638_cast_fp16 = einsum(equation = var_54638_equation_0, values = (var_54480_cast_fp16, var_54397_cast_fp16))[name = tensor<string, []>("op_54638_cast_fp16")];
            tensor<fp16, []> var_54639_to_fp16 = const()[name = tensor<string, []>("op_54639_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4643_cast_fp16 = mul(x = var_54638_cast_fp16, y = var_54639_to_fp16)[name = tensor<string, []>("aw_4643_cast_fp16")];
            tensor<string, []> var_54642_equation_0 = const()[name = tensor<string, []>("op_54642_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54642_cast_fp16 = einsum(equation = var_54642_equation_0, values = (var_54484_cast_fp16, var_54401_cast_fp16))[name = tensor<string, []>("op_54642_cast_fp16")];
            tensor<fp16, []> var_54643_to_fp16 = const()[name = tensor<string, []>("op_54643_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4645_cast_fp16 = mul(x = var_54642_cast_fp16, y = var_54643_to_fp16)[name = tensor<string, []>("aw_4645_cast_fp16")];
            tensor<string, []> var_54646_equation_0 = const()[name = tensor<string, []>("op_54646_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54646_cast_fp16 = einsum(equation = var_54646_equation_0, values = (var_54488_cast_fp16, var_54405_cast_fp16))[name = tensor<string, []>("op_54646_cast_fp16")];
            tensor<fp16, []> var_54647_to_fp16 = const()[name = tensor<string, []>("op_54647_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4647_cast_fp16 = mul(x = var_54646_cast_fp16, y = var_54647_to_fp16)[name = tensor<string, []>("aw_4647_cast_fp16")];
            tensor<string, []> var_54650_equation_0 = const()[name = tensor<string, []>("op_54650_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54650_cast_fp16 = einsum(equation = var_54650_equation_0, values = (var_54492_cast_fp16, var_54409_cast_fp16))[name = tensor<string, []>("op_54650_cast_fp16")];
            tensor<fp16, []> var_54651_to_fp16 = const()[name = tensor<string, []>("op_54651_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4649_cast_fp16 = mul(x = var_54650_cast_fp16, y = var_54651_to_fp16)[name = tensor<string, []>("aw_4649_cast_fp16")];
            tensor<string, []> var_54654_equation_0 = const()[name = tensor<string, []>("op_54654_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54654_cast_fp16 = einsum(equation = var_54654_equation_0, values = (var_54496_cast_fp16, var_54413_cast_fp16))[name = tensor<string, []>("op_54654_cast_fp16")];
            tensor<fp16, []> var_54655_to_fp16 = const()[name = tensor<string, []>("op_54655_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4651_cast_fp16 = mul(x = var_54654_cast_fp16, y = var_54655_to_fp16)[name = tensor<string, []>("aw_4651_cast_fp16")];
            tensor<string, []> var_54658_equation_0 = const()[name = tensor<string, []>("op_54658_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54658_cast_fp16 = einsum(equation = var_54658_equation_0, values = (var_54500_cast_fp16, var_54417_cast_fp16))[name = tensor<string, []>("op_54658_cast_fp16")];
            tensor<fp16, []> var_54659_to_fp16 = const()[name = tensor<string, []>("op_54659_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4653_cast_fp16 = mul(x = var_54658_cast_fp16, y = var_54659_to_fp16)[name = tensor<string, []>("aw_4653_cast_fp16")];
            tensor<string, []> var_54662_equation_0 = const()[name = tensor<string, []>("op_54662_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54662_cast_fp16 = einsum(equation = var_54662_equation_0, values = (var_54504_cast_fp16, var_54421_cast_fp16))[name = tensor<string, []>("op_54662_cast_fp16")];
            tensor<fp16, []> var_54663_to_fp16 = const()[name = tensor<string, []>("op_54663_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4655_cast_fp16 = mul(x = var_54662_cast_fp16, y = var_54663_to_fp16)[name = tensor<string, []>("aw_4655_cast_fp16")];
            tensor<string, []> var_54666_equation_0 = const()[name = tensor<string, []>("op_54666_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54666_cast_fp16 = einsum(equation = var_54666_equation_0, values = (var_54508_cast_fp16, var_54425_cast_fp16))[name = tensor<string, []>("op_54666_cast_fp16")];
            tensor<fp16, []> var_54667_to_fp16 = const()[name = tensor<string, []>("op_54667_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4657_cast_fp16 = mul(x = var_54666_cast_fp16, y = var_54667_to_fp16)[name = tensor<string, []>("aw_4657_cast_fp16")];
            tensor<string, []> var_54670_equation_0 = const()[name = tensor<string, []>("op_54670_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54670_cast_fp16 = einsum(equation = var_54670_equation_0, values = (var_54512_cast_fp16, var_54429_cast_fp16))[name = tensor<string, []>("op_54670_cast_fp16")];
            tensor<fp16, []> var_54671_to_fp16 = const()[name = tensor<string, []>("op_54671_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4659_cast_fp16 = mul(x = var_54670_cast_fp16, y = var_54671_to_fp16)[name = tensor<string, []>("aw_4659_cast_fp16")];
            tensor<string, []> var_54674_equation_0 = const()[name = tensor<string, []>("op_54674_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54674_cast_fp16 = einsum(equation = var_54674_equation_0, values = (var_54516_cast_fp16, var_54433_cast_fp16))[name = tensor<string, []>("op_54674_cast_fp16")];
            tensor<fp16, []> var_54675_to_fp16 = const()[name = tensor<string, []>("op_54675_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4661_cast_fp16 = mul(x = var_54674_cast_fp16, y = var_54675_to_fp16)[name = tensor<string, []>("aw_4661_cast_fp16")];
            tensor<string, []> var_54678_equation_0 = const()[name = tensor<string, []>("op_54678_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54678_cast_fp16 = einsum(equation = var_54678_equation_0, values = (var_54520_cast_fp16, var_54437_cast_fp16))[name = tensor<string, []>("op_54678_cast_fp16")];
            tensor<fp16, []> var_54679_to_fp16 = const()[name = tensor<string, []>("op_54679_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4663_cast_fp16 = mul(x = var_54678_cast_fp16, y = var_54679_to_fp16)[name = tensor<string, []>("aw_4663_cast_fp16")];
            tensor<string, []> var_54682_equation_0 = const()[name = tensor<string, []>("op_54682_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54682_cast_fp16 = einsum(equation = var_54682_equation_0, values = (var_54524_cast_fp16, var_54441_cast_fp16))[name = tensor<string, []>("op_54682_cast_fp16")];
            tensor<fp16, []> var_54683_to_fp16 = const()[name = tensor<string, []>("op_54683_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4665_cast_fp16 = mul(x = var_54682_cast_fp16, y = var_54683_to_fp16)[name = tensor<string, []>("aw_4665_cast_fp16")];
            tensor<string, []> var_54686_equation_0 = const()[name = tensor<string, []>("op_54686_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54686_cast_fp16 = einsum(equation = var_54686_equation_0, values = (var_54528_cast_fp16, var_54445_cast_fp16))[name = tensor<string, []>("op_54686_cast_fp16")];
            tensor<fp16, []> var_54687_to_fp16 = const()[name = tensor<string, []>("op_54687_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4667_cast_fp16 = mul(x = var_54686_cast_fp16, y = var_54687_to_fp16)[name = tensor<string, []>("aw_4667_cast_fp16")];
            tensor<string, []> var_54690_equation_0 = const()[name = tensor<string, []>("op_54690_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54690_cast_fp16 = einsum(equation = var_54690_equation_0, values = (var_54532_cast_fp16, var_54449_cast_fp16))[name = tensor<string, []>("op_54690_cast_fp16")];
            tensor<fp16, []> var_54691_to_fp16 = const()[name = tensor<string, []>("op_54691_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4669_cast_fp16 = mul(x = var_54690_cast_fp16, y = var_54691_to_fp16)[name = tensor<string, []>("aw_4669_cast_fp16")];
            tensor<string, []> var_54694_equation_0 = const()[name = tensor<string, []>("op_54694_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54694_cast_fp16 = einsum(equation = var_54694_equation_0, values = (var_54536_cast_fp16, var_54453_cast_fp16))[name = tensor<string, []>("op_54694_cast_fp16")];
            tensor<fp16, []> var_54695_to_fp16 = const()[name = tensor<string, []>("op_54695_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4671_cast_fp16 = mul(x = var_54694_cast_fp16, y = var_54695_to_fp16)[name = tensor<string, []>("aw_4671_cast_fp16")];
            tensor<string, []> var_54698_equation_0 = const()[name = tensor<string, []>("op_54698_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54698_cast_fp16 = einsum(equation = var_54698_equation_0, values = (var_54540_cast_fp16, var_54457_cast_fp16))[name = tensor<string, []>("op_54698_cast_fp16")];
            tensor<fp16, []> var_54699_to_fp16 = const()[name = tensor<string, []>("op_54699_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4673_cast_fp16 = mul(x = var_54698_cast_fp16, y = var_54699_to_fp16)[name = tensor<string, []>("aw_4673_cast_fp16")];
            tensor<string, []> var_54702_equation_0 = const()[name = tensor<string, []>("op_54702_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54702_cast_fp16 = einsum(equation = var_54702_equation_0, values = (var_54544_cast_fp16, var_54461_cast_fp16))[name = tensor<string, []>("op_54702_cast_fp16")];
            tensor<fp16, []> var_54703_to_fp16 = const()[name = tensor<string, []>("op_54703_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4675_cast_fp16 = mul(x = var_54702_cast_fp16, y = var_54703_to_fp16)[name = tensor<string, []>("aw_4675_cast_fp16")];
            tensor<string, []> var_54706_equation_0 = const()[name = tensor<string, []>("op_54706_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54706_cast_fp16 = einsum(equation = var_54706_equation_0, values = (var_54548_cast_fp16, var_54465_cast_fp16))[name = tensor<string, []>("op_54706_cast_fp16")];
            tensor<fp16, []> var_54707_to_fp16 = const()[name = tensor<string, []>("op_54707_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4677_cast_fp16 = mul(x = var_54706_cast_fp16, y = var_54707_to_fp16)[name = tensor<string, []>("aw_4677_cast_fp16")];
            tensor<string, []> var_54710_equation_0 = const()[name = tensor<string, []>("op_54710_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54710_cast_fp16 = einsum(equation = var_54710_equation_0, values = (var_54552_cast_fp16, var_54469_cast_fp16))[name = tensor<string, []>("op_54710_cast_fp16")];
            tensor<fp16, []> var_54711_to_fp16 = const()[name = tensor<string, []>("op_54711_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4679_cast_fp16 = mul(x = var_54710_cast_fp16, y = var_54711_to_fp16)[name = tensor<string, []>("aw_4679_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54713_cast_fp16 = softmax(axis = var_30355, x = aw_4641_cast_fp16)[name = tensor<string, []>("op_54713_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54714_cast_fp16 = softmax(axis = var_30355, x = aw_4643_cast_fp16)[name = tensor<string, []>("op_54714_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54715_cast_fp16 = softmax(axis = var_30355, x = aw_4645_cast_fp16)[name = tensor<string, []>("op_54715_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54716_cast_fp16 = softmax(axis = var_30355, x = aw_4647_cast_fp16)[name = tensor<string, []>("op_54716_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54717_cast_fp16 = softmax(axis = var_30355, x = aw_4649_cast_fp16)[name = tensor<string, []>("op_54717_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54718_cast_fp16 = softmax(axis = var_30355, x = aw_4651_cast_fp16)[name = tensor<string, []>("op_54718_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54719_cast_fp16 = softmax(axis = var_30355, x = aw_4653_cast_fp16)[name = tensor<string, []>("op_54719_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54720_cast_fp16 = softmax(axis = var_30355, x = aw_4655_cast_fp16)[name = tensor<string, []>("op_54720_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54721_cast_fp16 = softmax(axis = var_30355, x = aw_4657_cast_fp16)[name = tensor<string, []>("op_54721_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54722_cast_fp16 = softmax(axis = var_30355, x = aw_4659_cast_fp16)[name = tensor<string, []>("op_54722_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54723_cast_fp16 = softmax(axis = var_30355, x = aw_4661_cast_fp16)[name = tensor<string, []>("op_54723_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54724_cast_fp16 = softmax(axis = var_30355, x = aw_4663_cast_fp16)[name = tensor<string, []>("op_54724_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54725_cast_fp16 = softmax(axis = var_30355, x = aw_4665_cast_fp16)[name = tensor<string, []>("op_54725_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54726_cast_fp16 = softmax(axis = var_30355, x = aw_4667_cast_fp16)[name = tensor<string, []>("op_54726_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54727_cast_fp16 = softmax(axis = var_30355, x = aw_4669_cast_fp16)[name = tensor<string, []>("op_54727_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54728_cast_fp16 = softmax(axis = var_30355, x = aw_4671_cast_fp16)[name = tensor<string, []>("op_54728_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54729_cast_fp16 = softmax(axis = var_30355, x = aw_4673_cast_fp16)[name = tensor<string, []>("op_54729_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54730_cast_fp16 = softmax(axis = var_30355, x = aw_4675_cast_fp16)[name = tensor<string, []>("op_54730_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54731_cast_fp16 = softmax(axis = var_30355, x = aw_4677_cast_fp16)[name = tensor<string, []>("op_54731_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54732_cast_fp16 = softmax(axis = var_30355, x = aw_4679_cast_fp16)[name = tensor<string, []>("op_54732_cast_fp16")];
            tensor<string, []> var_54734_equation_0 = const()[name = tensor<string, []>("op_54734_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54734_cast_fp16 = einsum(equation = var_54734_equation_0, values = (var_54554_cast_fp16, var_54713_cast_fp16))[name = tensor<string, []>("op_54734_cast_fp16")];
            tensor<string, []> var_54736_equation_0 = const()[name = tensor<string, []>("op_54736_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54736_cast_fp16 = einsum(equation = var_54736_equation_0, values = (var_54558_cast_fp16, var_54714_cast_fp16))[name = tensor<string, []>("op_54736_cast_fp16")];
            tensor<string, []> var_54738_equation_0 = const()[name = tensor<string, []>("op_54738_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54738_cast_fp16 = einsum(equation = var_54738_equation_0, values = (var_54562_cast_fp16, var_54715_cast_fp16))[name = tensor<string, []>("op_54738_cast_fp16")];
            tensor<string, []> var_54740_equation_0 = const()[name = tensor<string, []>("op_54740_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54740_cast_fp16 = einsum(equation = var_54740_equation_0, values = (var_54566_cast_fp16, var_54716_cast_fp16))[name = tensor<string, []>("op_54740_cast_fp16")];
            tensor<string, []> var_54742_equation_0 = const()[name = tensor<string, []>("op_54742_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54742_cast_fp16 = einsum(equation = var_54742_equation_0, values = (var_54570_cast_fp16, var_54717_cast_fp16))[name = tensor<string, []>("op_54742_cast_fp16")];
            tensor<string, []> var_54744_equation_0 = const()[name = tensor<string, []>("op_54744_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54744_cast_fp16 = einsum(equation = var_54744_equation_0, values = (var_54574_cast_fp16, var_54718_cast_fp16))[name = tensor<string, []>("op_54744_cast_fp16")];
            tensor<string, []> var_54746_equation_0 = const()[name = tensor<string, []>("op_54746_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54746_cast_fp16 = einsum(equation = var_54746_equation_0, values = (var_54578_cast_fp16, var_54719_cast_fp16))[name = tensor<string, []>("op_54746_cast_fp16")];
            tensor<string, []> var_54748_equation_0 = const()[name = tensor<string, []>("op_54748_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54748_cast_fp16 = einsum(equation = var_54748_equation_0, values = (var_54582_cast_fp16, var_54720_cast_fp16))[name = tensor<string, []>("op_54748_cast_fp16")];
            tensor<string, []> var_54750_equation_0 = const()[name = tensor<string, []>("op_54750_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54750_cast_fp16 = einsum(equation = var_54750_equation_0, values = (var_54586_cast_fp16, var_54721_cast_fp16))[name = tensor<string, []>("op_54750_cast_fp16")];
            tensor<string, []> var_54752_equation_0 = const()[name = tensor<string, []>("op_54752_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54752_cast_fp16 = einsum(equation = var_54752_equation_0, values = (var_54590_cast_fp16, var_54722_cast_fp16))[name = tensor<string, []>("op_54752_cast_fp16")];
            tensor<string, []> var_54754_equation_0 = const()[name = tensor<string, []>("op_54754_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54754_cast_fp16 = einsum(equation = var_54754_equation_0, values = (var_54594_cast_fp16, var_54723_cast_fp16))[name = tensor<string, []>("op_54754_cast_fp16")];
            tensor<string, []> var_54756_equation_0 = const()[name = tensor<string, []>("op_54756_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54756_cast_fp16 = einsum(equation = var_54756_equation_0, values = (var_54598_cast_fp16, var_54724_cast_fp16))[name = tensor<string, []>("op_54756_cast_fp16")];
            tensor<string, []> var_54758_equation_0 = const()[name = tensor<string, []>("op_54758_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54758_cast_fp16 = einsum(equation = var_54758_equation_0, values = (var_54602_cast_fp16, var_54725_cast_fp16))[name = tensor<string, []>("op_54758_cast_fp16")];
            tensor<string, []> var_54760_equation_0 = const()[name = tensor<string, []>("op_54760_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54760_cast_fp16 = einsum(equation = var_54760_equation_0, values = (var_54606_cast_fp16, var_54726_cast_fp16))[name = tensor<string, []>("op_54760_cast_fp16")];
            tensor<string, []> var_54762_equation_0 = const()[name = tensor<string, []>("op_54762_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54762_cast_fp16 = einsum(equation = var_54762_equation_0, values = (var_54610_cast_fp16, var_54727_cast_fp16))[name = tensor<string, []>("op_54762_cast_fp16")];
            tensor<string, []> var_54764_equation_0 = const()[name = tensor<string, []>("op_54764_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54764_cast_fp16 = einsum(equation = var_54764_equation_0, values = (var_54614_cast_fp16, var_54728_cast_fp16))[name = tensor<string, []>("op_54764_cast_fp16")];
            tensor<string, []> var_54766_equation_0 = const()[name = tensor<string, []>("op_54766_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54766_cast_fp16 = einsum(equation = var_54766_equation_0, values = (var_54618_cast_fp16, var_54729_cast_fp16))[name = tensor<string, []>("op_54766_cast_fp16")];
            tensor<string, []> var_54768_equation_0 = const()[name = tensor<string, []>("op_54768_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54768_cast_fp16 = einsum(equation = var_54768_equation_0, values = (var_54622_cast_fp16, var_54730_cast_fp16))[name = tensor<string, []>("op_54768_cast_fp16")];
            tensor<string, []> var_54770_equation_0 = const()[name = tensor<string, []>("op_54770_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54770_cast_fp16 = einsum(equation = var_54770_equation_0, values = (var_54626_cast_fp16, var_54731_cast_fp16))[name = tensor<string, []>("op_54770_cast_fp16")];
            tensor<string, []> var_54772_equation_0 = const()[name = tensor<string, []>("op_54772_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54772_cast_fp16 = einsum(equation = var_54772_equation_0, values = (var_54630_cast_fp16, var_54732_cast_fp16))[name = tensor<string, []>("op_54772_cast_fp16")];
            tensor<bool, []> input_683_interleave_0 = const()[name = tensor<string, []>("input_683_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_683_cast_fp16 = concat(axis = var_30355, interleave = input_683_interleave_0, values = (var_54734_cast_fp16, var_54736_cast_fp16, var_54738_cast_fp16, var_54740_cast_fp16, var_54742_cast_fp16, var_54744_cast_fp16, var_54746_cast_fp16, var_54748_cast_fp16, var_54750_cast_fp16, var_54752_cast_fp16, var_54754_cast_fp16, var_54756_cast_fp16, var_54758_cast_fp16, var_54760_cast_fp16, var_54762_cast_fp16, var_54764_cast_fp16, var_54766_cast_fp16, var_54768_cast_fp16, var_54770_cast_fp16, var_54772_cast_fp16))[name = tensor<string, []>("input_683_cast_fp16")];
            tensor<string, []> var_54782_pad_type_0 = const()[name = tensor<string, []>("op_54782_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_54782_strides_0 = const()[name = tensor<string, []>("op_54782_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_54782_pad_0 = const()[name = tensor<string, []>("op_54782_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_54782_dilations_0 = const()[name = tensor<string, []>("op_54782_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_54782_groups_0 = const()[name = tensor<string, []>("op_54782_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_6_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(758519616))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(759748480))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_6_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_6_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_6_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(759748672)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_54782_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_6_attn1_to_out_0_bias_to_fp16, dilations = var_54782_dilations_0, groups = var_54782_groups_0, pad = var_54782_pad_0, pad_type = var_54782_pad_type_0, strides = var_54782_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_6_attn1_to_out_0_weight_to_fp16_palettized, x = input_683_cast_fp16)[name = tensor<string, []>("op_54782_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_363_cast_fp16 = add(x = var_54782_cast_fp16, y = inputs_361_cast_fp16)[name = tensor<string, []>("inputs_363_cast_fp16")];
            tensor<int32, [1]> hidden_states_473_axes_0 = const()[name = tensor<string, []>("hidden_states_473_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_473_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_473_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(759751296)))];
            tensor<fp16, [1280]> hidden_states_473_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_473_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(759753920)))];
            tensor<fp16, []> var_54792_to_fp16 = const()[name = tensor<string, []>("op_54792_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_473_cast_fp16 = layer_norm(axes = hidden_states_473_axes_0, beta = hidden_states_473_beta_0_to_fp16, epsilon = var_54792_to_fp16, gamma = hidden_states_473_gamma_0_to_fp16, x = inputs_363_cast_fp16)[name = tensor<string, []>("hidden_states_473_cast_fp16")];
            tensor<string, []> q_243_pad_type_0 = const()[name = tensor<string, []>("q_243_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_243_strides_0 = const()[name = tensor<string, []>("q_243_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_243_pad_0 = const()[name = tensor<string, []>("q_243_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_243_dilations_0 = const()[name = tensor<string, []>("q_243_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_243_groups_0 = const()[name = tensor<string, []>("q_243_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_6_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(759756544))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(760985408))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_6_attn2_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_243_cast_fp16 = conv(dilations = q_243_dilations_0, groups = q_243_groups_0, pad = q_243_pad_0, pad_type = q_243_pad_type_0, strides = q_243_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_6_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_473_cast_fp16)[name = tensor<string, []>("q_243_cast_fp16")];
            tensor<string, []> k_485_pad_type_0 = const()[name = tensor<string, []>("k_485_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_485_strides_0 = const()[name = tensor<string, []>("k_485_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_485_pad_0 = const()[name = tensor<string, []>("k_485_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_485_dilations_0 = const()[name = tensor<string, []>("k_485_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_485_groups_0 = const()[name = tensor<string, []>("k_485_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_6_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(760985600))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(762951744))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_6_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> k_485_cast_fp16 = conv(dilations = k_485_dilations_0, groups = k_485_groups_0, pad = k_485_pad_0, pad_type = k_485_pad_type_0, strides = k_485_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_6_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("k_485_cast_fp16")];
            tensor<string, []> v_243_pad_type_0 = const()[name = tensor<string, []>("v_243_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_243_strides_0 = const()[name = tensor<string, []>("v_243_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_243_pad_0 = const()[name = tensor<string, []>("v_243_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_243_dilations_0 = const()[name = tensor<string, []>("v_243_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_243_groups_0 = const()[name = tensor<string, []>("v_243_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_6_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(762951936))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(764918080))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_6_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> v_243_cast_fp16 = conv(dilations = v_243_dilations_0, groups = v_243_groups_0, pad = v_243_pad_0, pad_type = v_243_pad_type_0, strides = v_243_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_6_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("v_243_cast_fp16")];
            tensor<int32, [4]> var_54825_begin_0 = const()[name = tensor<string, []>("op_54825_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_54825_end_0 = const()[name = tensor<string, []>("op_54825_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_54825_end_mask_0 = const()[name = tensor<string, []>("op_54825_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54825_cast_fp16 = slice_by_index(begin = var_54825_begin_0, end = var_54825_end_0, end_mask = var_54825_end_mask_0, x = q_243_cast_fp16)[name = tensor<string, []>("op_54825_cast_fp16")];
            tensor<int32, [4]> var_54829_begin_0 = const()[name = tensor<string, []>("op_54829_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_54829_end_0 = const()[name = tensor<string, []>("op_54829_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_54829_end_mask_0 = const()[name = tensor<string, []>("op_54829_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54829_cast_fp16 = slice_by_index(begin = var_54829_begin_0, end = var_54829_end_0, end_mask = var_54829_end_mask_0, x = q_243_cast_fp16)[name = tensor<string, []>("op_54829_cast_fp16")];
            tensor<int32, [4]> var_54833_begin_0 = const()[name = tensor<string, []>("op_54833_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_54833_end_0 = const()[name = tensor<string, []>("op_54833_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_54833_end_mask_0 = const()[name = tensor<string, []>("op_54833_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54833_cast_fp16 = slice_by_index(begin = var_54833_begin_0, end = var_54833_end_0, end_mask = var_54833_end_mask_0, x = q_243_cast_fp16)[name = tensor<string, []>("op_54833_cast_fp16")];
            tensor<int32, [4]> var_54837_begin_0 = const()[name = tensor<string, []>("op_54837_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_54837_end_0 = const()[name = tensor<string, []>("op_54837_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_54837_end_mask_0 = const()[name = tensor<string, []>("op_54837_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54837_cast_fp16 = slice_by_index(begin = var_54837_begin_0, end = var_54837_end_0, end_mask = var_54837_end_mask_0, x = q_243_cast_fp16)[name = tensor<string, []>("op_54837_cast_fp16")];
            tensor<int32, [4]> var_54841_begin_0 = const()[name = tensor<string, []>("op_54841_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_54841_end_0 = const()[name = tensor<string, []>("op_54841_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_54841_end_mask_0 = const()[name = tensor<string, []>("op_54841_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54841_cast_fp16 = slice_by_index(begin = var_54841_begin_0, end = var_54841_end_0, end_mask = var_54841_end_mask_0, x = q_243_cast_fp16)[name = tensor<string, []>("op_54841_cast_fp16")];
            tensor<int32, [4]> var_54845_begin_0 = const()[name = tensor<string, []>("op_54845_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_54845_end_0 = const()[name = tensor<string, []>("op_54845_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_54845_end_mask_0 = const()[name = tensor<string, []>("op_54845_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54845_cast_fp16 = slice_by_index(begin = var_54845_begin_0, end = var_54845_end_0, end_mask = var_54845_end_mask_0, x = q_243_cast_fp16)[name = tensor<string, []>("op_54845_cast_fp16")];
            tensor<int32, [4]> var_54849_begin_0 = const()[name = tensor<string, []>("op_54849_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_54849_end_0 = const()[name = tensor<string, []>("op_54849_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_54849_end_mask_0 = const()[name = tensor<string, []>("op_54849_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54849_cast_fp16 = slice_by_index(begin = var_54849_begin_0, end = var_54849_end_0, end_mask = var_54849_end_mask_0, x = q_243_cast_fp16)[name = tensor<string, []>("op_54849_cast_fp16")];
            tensor<int32, [4]> var_54853_begin_0 = const()[name = tensor<string, []>("op_54853_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_54853_end_0 = const()[name = tensor<string, []>("op_54853_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_54853_end_mask_0 = const()[name = tensor<string, []>("op_54853_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54853_cast_fp16 = slice_by_index(begin = var_54853_begin_0, end = var_54853_end_0, end_mask = var_54853_end_mask_0, x = q_243_cast_fp16)[name = tensor<string, []>("op_54853_cast_fp16")];
            tensor<int32, [4]> var_54857_begin_0 = const()[name = tensor<string, []>("op_54857_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_54857_end_0 = const()[name = tensor<string, []>("op_54857_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_54857_end_mask_0 = const()[name = tensor<string, []>("op_54857_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54857_cast_fp16 = slice_by_index(begin = var_54857_begin_0, end = var_54857_end_0, end_mask = var_54857_end_mask_0, x = q_243_cast_fp16)[name = tensor<string, []>("op_54857_cast_fp16")];
            tensor<int32, [4]> var_54861_begin_0 = const()[name = tensor<string, []>("op_54861_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_54861_end_0 = const()[name = tensor<string, []>("op_54861_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_54861_end_mask_0 = const()[name = tensor<string, []>("op_54861_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54861_cast_fp16 = slice_by_index(begin = var_54861_begin_0, end = var_54861_end_0, end_mask = var_54861_end_mask_0, x = q_243_cast_fp16)[name = tensor<string, []>("op_54861_cast_fp16")];
            tensor<int32, [4]> var_54865_begin_0 = const()[name = tensor<string, []>("op_54865_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_54865_end_0 = const()[name = tensor<string, []>("op_54865_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_54865_end_mask_0 = const()[name = tensor<string, []>("op_54865_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54865_cast_fp16 = slice_by_index(begin = var_54865_begin_0, end = var_54865_end_0, end_mask = var_54865_end_mask_0, x = q_243_cast_fp16)[name = tensor<string, []>("op_54865_cast_fp16")];
            tensor<int32, [4]> var_54869_begin_0 = const()[name = tensor<string, []>("op_54869_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_54869_end_0 = const()[name = tensor<string, []>("op_54869_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_54869_end_mask_0 = const()[name = tensor<string, []>("op_54869_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54869_cast_fp16 = slice_by_index(begin = var_54869_begin_0, end = var_54869_end_0, end_mask = var_54869_end_mask_0, x = q_243_cast_fp16)[name = tensor<string, []>("op_54869_cast_fp16")];
            tensor<int32, [4]> var_54873_begin_0 = const()[name = tensor<string, []>("op_54873_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_54873_end_0 = const()[name = tensor<string, []>("op_54873_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_54873_end_mask_0 = const()[name = tensor<string, []>("op_54873_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54873_cast_fp16 = slice_by_index(begin = var_54873_begin_0, end = var_54873_end_0, end_mask = var_54873_end_mask_0, x = q_243_cast_fp16)[name = tensor<string, []>("op_54873_cast_fp16")];
            tensor<int32, [4]> var_54877_begin_0 = const()[name = tensor<string, []>("op_54877_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_54877_end_0 = const()[name = tensor<string, []>("op_54877_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_54877_end_mask_0 = const()[name = tensor<string, []>("op_54877_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54877_cast_fp16 = slice_by_index(begin = var_54877_begin_0, end = var_54877_end_0, end_mask = var_54877_end_mask_0, x = q_243_cast_fp16)[name = tensor<string, []>("op_54877_cast_fp16")];
            tensor<int32, [4]> var_54881_begin_0 = const()[name = tensor<string, []>("op_54881_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_54881_end_0 = const()[name = tensor<string, []>("op_54881_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_54881_end_mask_0 = const()[name = tensor<string, []>("op_54881_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54881_cast_fp16 = slice_by_index(begin = var_54881_begin_0, end = var_54881_end_0, end_mask = var_54881_end_mask_0, x = q_243_cast_fp16)[name = tensor<string, []>("op_54881_cast_fp16")];
            tensor<int32, [4]> var_54885_begin_0 = const()[name = tensor<string, []>("op_54885_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_54885_end_0 = const()[name = tensor<string, []>("op_54885_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_54885_end_mask_0 = const()[name = tensor<string, []>("op_54885_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54885_cast_fp16 = slice_by_index(begin = var_54885_begin_0, end = var_54885_end_0, end_mask = var_54885_end_mask_0, x = q_243_cast_fp16)[name = tensor<string, []>("op_54885_cast_fp16")];
            tensor<int32, [4]> var_54889_begin_0 = const()[name = tensor<string, []>("op_54889_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_54889_end_0 = const()[name = tensor<string, []>("op_54889_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_54889_end_mask_0 = const()[name = tensor<string, []>("op_54889_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54889_cast_fp16 = slice_by_index(begin = var_54889_begin_0, end = var_54889_end_0, end_mask = var_54889_end_mask_0, x = q_243_cast_fp16)[name = tensor<string, []>("op_54889_cast_fp16")];
            tensor<int32, [4]> var_54893_begin_0 = const()[name = tensor<string, []>("op_54893_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_54893_end_0 = const()[name = tensor<string, []>("op_54893_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_54893_end_mask_0 = const()[name = tensor<string, []>("op_54893_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54893_cast_fp16 = slice_by_index(begin = var_54893_begin_0, end = var_54893_end_0, end_mask = var_54893_end_mask_0, x = q_243_cast_fp16)[name = tensor<string, []>("op_54893_cast_fp16")];
            tensor<int32, [4]> var_54897_begin_0 = const()[name = tensor<string, []>("op_54897_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_54897_end_0 = const()[name = tensor<string, []>("op_54897_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_54897_end_mask_0 = const()[name = tensor<string, []>("op_54897_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54897_cast_fp16 = slice_by_index(begin = var_54897_begin_0, end = var_54897_end_0, end_mask = var_54897_end_mask_0, x = q_243_cast_fp16)[name = tensor<string, []>("op_54897_cast_fp16")];
            tensor<int32, [4]> var_54901_begin_0 = const()[name = tensor<string, []>("op_54901_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_54901_end_0 = const()[name = tensor<string, []>("op_54901_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_54901_end_mask_0 = const()[name = tensor<string, []>("op_54901_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54901_cast_fp16 = slice_by_index(begin = var_54901_begin_0, end = var_54901_end_0, end_mask = var_54901_end_mask_0, x = q_243_cast_fp16)[name = tensor<string, []>("op_54901_cast_fp16")];
            tensor<int32, [4]> k_487_perm_0 = const()[name = tensor<string, []>("k_487_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_54908_begin_0 = const()[name = tensor<string, []>("op_54908_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_54908_end_0 = const()[name = tensor<string, []>("op_54908_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_54908_end_mask_0 = const()[name = tensor<string, []>("op_54908_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> k_487_cast_fp16 = transpose(perm = k_487_perm_0, x = k_485_cast_fp16)[name = tensor<string, []>("transpose_18")];
            tensor<fp16, [2, 77, 1, 64]> var_54908_cast_fp16 = slice_by_index(begin = var_54908_begin_0, end = var_54908_end_0, end_mask = var_54908_end_mask_0, x = k_487_cast_fp16)[name = tensor<string, []>("op_54908_cast_fp16")];
            tensor<int32, [4]> var_54912_begin_0 = const()[name = tensor<string, []>("op_54912_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_54912_end_0 = const()[name = tensor<string, []>("op_54912_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_54912_end_mask_0 = const()[name = tensor<string, []>("op_54912_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54912_cast_fp16 = slice_by_index(begin = var_54912_begin_0, end = var_54912_end_0, end_mask = var_54912_end_mask_0, x = k_487_cast_fp16)[name = tensor<string, []>("op_54912_cast_fp16")];
            tensor<int32, [4]> var_54916_begin_0 = const()[name = tensor<string, []>("op_54916_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_54916_end_0 = const()[name = tensor<string, []>("op_54916_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_54916_end_mask_0 = const()[name = tensor<string, []>("op_54916_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54916_cast_fp16 = slice_by_index(begin = var_54916_begin_0, end = var_54916_end_0, end_mask = var_54916_end_mask_0, x = k_487_cast_fp16)[name = tensor<string, []>("op_54916_cast_fp16")];
            tensor<int32, [4]> var_54920_begin_0 = const()[name = tensor<string, []>("op_54920_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_54920_end_0 = const()[name = tensor<string, []>("op_54920_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_54920_end_mask_0 = const()[name = tensor<string, []>("op_54920_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54920_cast_fp16 = slice_by_index(begin = var_54920_begin_0, end = var_54920_end_0, end_mask = var_54920_end_mask_0, x = k_487_cast_fp16)[name = tensor<string, []>("op_54920_cast_fp16")];
            tensor<int32, [4]> var_54924_begin_0 = const()[name = tensor<string, []>("op_54924_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_54924_end_0 = const()[name = tensor<string, []>("op_54924_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_54924_end_mask_0 = const()[name = tensor<string, []>("op_54924_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54924_cast_fp16 = slice_by_index(begin = var_54924_begin_0, end = var_54924_end_0, end_mask = var_54924_end_mask_0, x = k_487_cast_fp16)[name = tensor<string, []>("op_54924_cast_fp16")];
            tensor<int32, [4]> var_54928_begin_0 = const()[name = tensor<string, []>("op_54928_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_54928_end_0 = const()[name = tensor<string, []>("op_54928_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_54928_end_mask_0 = const()[name = tensor<string, []>("op_54928_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54928_cast_fp16 = slice_by_index(begin = var_54928_begin_0, end = var_54928_end_0, end_mask = var_54928_end_mask_0, x = k_487_cast_fp16)[name = tensor<string, []>("op_54928_cast_fp16")];
            tensor<int32, [4]> var_54932_begin_0 = const()[name = tensor<string, []>("op_54932_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_54932_end_0 = const()[name = tensor<string, []>("op_54932_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_54932_end_mask_0 = const()[name = tensor<string, []>("op_54932_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54932_cast_fp16 = slice_by_index(begin = var_54932_begin_0, end = var_54932_end_0, end_mask = var_54932_end_mask_0, x = k_487_cast_fp16)[name = tensor<string, []>("op_54932_cast_fp16")];
            tensor<int32, [4]> var_54936_begin_0 = const()[name = tensor<string, []>("op_54936_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_54936_end_0 = const()[name = tensor<string, []>("op_54936_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_54936_end_mask_0 = const()[name = tensor<string, []>("op_54936_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54936_cast_fp16 = slice_by_index(begin = var_54936_begin_0, end = var_54936_end_0, end_mask = var_54936_end_mask_0, x = k_487_cast_fp16)[name = tensor<string, []>("op_54936_cast_fp16")];
            tensor<int32, [4]> var_54940_begin_0 = const()[name = tensor<string, []>("op_54940_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_54940_end_0 = const()[name = tensor<string, []>("op_54940_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_54940_end_mask_0 = const()[name = tensor<string, []>("op_54940_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54940_cast_fp16 = slice_by_index(begin = var_54940_begin_0, end = var_54940_end_0, end_mask = var_54940_end_mask_0, x = k_487_cast_fp16)[name = tensor<string, []>("op_54940_cast_fp16")];
            tensor<int32, [4]> var_54944_begin_0 = const()[name = tensor<string, []>("op_54944_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_54944_end_0 = const()[name = tensor<string, []>("op_54944_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_54944_end_mask_0 = const()[name = tensor<string, []>("op_54944_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54944_cast_fp16 = slice_by_index(begin = var_54944_begin_0, end = var_54944_end_0, end_mask = var_54944_end_mask_0, x = k_487_cast_fp16)[name = tensor<string, []>("op_54944_cast_fp16")];
            tensor<int32, [4]> var_54948_begin_0 = const()[name = tensor<string, []>("op_54948_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_54948_end_0 = const()[name = tensor<string, []>("op_54948_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_54948_end_mask_0 = const()[name = tensor<string, []>("op_54948_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54948_cast_fp16 = slice_by_index(begin = var_54948_begin_0, end = var_54948_end_0, end_mask = var_54948_end_mask_0, x = k_487_cast_fp16)[name = tensor<string, []>("op_54948_cast_fp16")];
            tensor<int32, [4]> var_54952_begin_0 = const()[name = tensor<string, []>("op_54952_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_54952_end_0 = const()[name = tensor<string, []>("op_54952_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_54952_end_mask_0 = const()[name = tensor<string, []>("op_54952_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54952_cast_fp16 = slice_by_index(begin = var_54952_begin_0, end = var_54952_end_0, end_mask = var_54952_end_mask_0, x = k_487_cast_fp16)[name = tensor<string, []>("op_54952_cast_fp16")];
            tensor<int32, [4]> var_54956_begin_0 = const()[name = tensor<string, []>("op_54956_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_54956_end_0 = const()[name = tensor<string, []>("op_54956_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_54956_end_mask_0 = const()[name = tensor<string, []>("op_54956_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54956_cast_fp16 = slice_by_index(begin = var_54956_begin_0, end = var_54956_end_0, end_mask = var_54956_end_mask_0, x = k_487_cast_fp16)[name = tensor<string, []>("op_54956_cast_fp16")];
            tensor<int32, [4]> var_54960_begin_0 = const()[name = tensor<string, []>("op_54960_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_54960_end_0 = const()[name = tensor<string, []>("op_54960_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_54960_end_mask_0 = const()[name = tensor<string, []>("op_54960_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54960_cast_fp16 = slice_by_index(begin = var_54960_begin_0, end = var_54960_end_0, end_mask = var_54960_end_mask_0, x = k_487_cast_fp16)[name = tensor<string, []>("op_54960_cast_fp16")];
            tensor<int32, [4]> var_54964_begin_0 = const()[name = tensor<string, []>("op_54964_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_54964_end_0 = const()[name = tensor<string, []>("op_54964_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_54964_end_mask_0 = const()[name = tensor<string, []>("op_54964_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54964_cast_fp16 = slice_by_index(begin = var_54964_begin_0, end = var_54964_end_0, end_mask = var_54964_end_mask_0, x = k_487_cast_fp16)[name = tensor<string, []>("op_54964_cast_fp16")];
            tensor<int32, [4]> var_54968_begin_0 = const()[name = tensor<string, []>("op_54968_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_54968_end_0 = const()[name = tensor<string, []>("op_54968_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_54968_end_mask_0 = const()[name = tensor<string, []>("op_54968_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54968_cast_fp16 = slice_by_index(begin = var_54968_begin_0, end = var_54968_end_0, end_mask = var_54968_end_mask_0, x = k_487_cast_fp16)[name = tensor<string, []>("op_54968_cast_fp16")];
            tensor<int32, [4]> var_54972_begin_0 = const()[name = tensor<string, []>("op_54972_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_54972_end_0 = const()[name = tensor<string, []>("op_54972_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_54972_end_mask_0 = const()[name = tensor<string, []>("op_54972_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54972_cast_fp16 = slice_by_index(begin = var_54972_begin_0, end = var_54972_end_0, end_mask = var_54972_end_mask_0, x = k_487_cast_fp16)[name = tensor<string, []>("op_54972_cast_fp16")];
            tensor<int32, [4]> var_54976_begin_0 = const()[name = tensor<string, []>("op_54976_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_54976_end_0 = const()[name = tensor<string, []>("op_54976_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_54976_end_mask_0 = const()[name = tensor<string, []>("op_54976_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54976_cast_fp16 = slice_by_index(begin = var_54976_begin_0, end = var_54976_end_0, end_mask = var_54976_end_mask_0, x = k_487_cast_fp16)[name = tensor<string, []>("op_54976_cast_fp16")];
            tensor<int32, [4]> var_54980_begin_0 = const()[name = tensor<string, []>("op_54980_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_54980_end_0 = const()[name = tensor<string, []>("op_54980_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_54980_end_mask_0 = const()[name = tensor<string, []>("op_54980_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54980_cast_fp16 = slice_by_index(begin = var_54980_begin_0, end = var_54980_end_0, end_mask = var_54980_end_mask_0, x = k_487_cast_fp16)[name = tensor<string, []>("op_54980_cast_fp16")];
            tensor<int32, [4]> var_54984_begin_0 = const()[name = tensor<string, []>("op_54984_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_54984_end_0 = const()[name = tensor<string, []>("op_54984_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_54984_end_mask_0 = const()[name = tensor<string, []>("op_54984_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54984_cast_fp16 = slice_by_index(begin = var_54984_begin_0, end = var_54984_end_0, end_mask = var_54984_end_mask_0, x = k_487_cast_fp16)[name = tensor<string, []>("op_54984_cast_fp16")];
            tensor<int32, [4]> var_54986_begin_0 = const()[name = tensor<string, []>("op_54986_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_54986_end_0 = const()[name = tensor<string, []>("op_54986_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_54986_end_mask_0 = const()[name = tensor<string, []>("op_54986_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_54986_cast_fp16 = slice_by_index(begin = var_54986_begin_0, end = var_54986_end_0, end_mask = var_54986_end_mask_0, x = v_243_cast_fp16)[name = tensor<string, []>("op_54986_cast_fp16")];
            tensor<int32, [4]> var_54990_begin_0 = const()[name = tensor<string, []>("op_54990_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_54990_end_0 = const()[name = tensor<string, []>("op_54990_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_54990_end_mask_0 = const()[name = tensor<string, []>("op_54990_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_54990_cast_fp16 = slice_by_index(begin = var_54990_begin_0, end = var_54990_end_0, end_mask = var_54990_end_mask_0, x = v_243_cast_fp16)[name = tensor<string, []>("op_54990_cast_fp16")];
            tensor<int32, [4]> var_54994_begin_0 = const()[name = tensor<string, []>("op_54994_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_54994_end_0 = const()[name = tensor<string, []>("op_54994_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_54994_end_mask_0 = const()[name = tensor<string, []>("op_54994_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_54994_cast_fp16 = slice_by_index(begin = var_54994_begin_0, end = var_54994_end_0, end_mask = var_54994_end_mask_0, x = v_243_cast_fp16)[name = tensor<string, []>("op_54994_cast_fp16")];
            tensor<int32, [4]> var_54998_begin_0 = const()[name = tensor<string, []>("op_54998_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_54998_end_0 = const()[name = tensor<string, []>("op_54998_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_54998_end_mask_0 = const()[name = tensor<string, []>("op_54998_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_54998_cast_fp16 = slice_by_index(begin = var_54998_begin_0, end = var_54998_end_0, end_mask = var_54998_end_mask_0, x = v_243_cast_fp16)[name = tensor<string, []>("op_54998_cast_fp16")];
            tensor<int32, [4]> var_55002_begin_0 = const()[name = tensor<string, []>("op_55002_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_55002_end_0 = const()[name = tensor<string, []>("op_55002_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_55002_end_mask_0 = const()[name = tensor<string, []>("op_55002_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55002_cast_fp16 = slice_by_index(begin = var_55002_begin_0, end = var_55002_end_0, end_mask = var_55002_end_mask_0, x = v_243_cast_fp16)[name = tensor<string, []>("op_55002_cast_fp16")];
            tensor<int32, [4]> var_55006_begin_0 = const()[name = tensor<string, []>("op_55006_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_55006_end_0 = const()[name = tensor<string, []>("op_55006_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_55006_end_mask_0 = const()[name = tensor<string, []>("op_55006_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55006_cast_fp16 = slice_by_index(begin = var_55006_begin_0, end = var_55006_end_0, end_mask = var_55006_end_mask_0, x = v_243_cast_fp16)[name = tensor<string, []>("op_55006_cast_fp16")];
            tensor<int32, [4]> var_55010_begin_0 = const()[name = tensor<string, []>("op_55010_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_55010_end_0 = const()[name = tensor<string, []>("op_55010_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_55010_end_mask_0 = const()[name = tensor<string, []>("op_55010_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55010_cast_fp16 = slice_by_index(begin = var_55010_begin_0, end = var_55010_end_0, end_mask = var_55010_end_mask_0, x = v_243_cast_fp16)[name = tensor<string, []>("op_55010_cast_fp16")];
            tensor<int32, [4]> var_55014_begin_0 = const()[name = tensor<string, []>("op_55014_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_55014_end_0 = const()[name = tensor<string, []>("op_55014_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_55014_end_mask_0 = const()[name = tensor<string, []>("op_55014_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55014_cast_fp16 = slice_by_index(begin = var_55014_begin_0, end = var_55014_end_0, end_mask = var_55014_end_mask_0, x = v_243_cast_fp16)[name = tensor<string, []>("op_55014_cast_fp16")];
            tensor<int32, [4]> var_55018_begin_0 = const()[name = tensor<string, []>("op_55018_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_55018_end_0 = const()[name = tensor<string, []>("op_55018_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_55018_end_mask_0 = const()[name = tensor<string, []>("op_55018_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55018_cast_fp16 = slice_by_index(begin = var_55018_begin_0, end = var_55018_end_0, end_mask = var_55018_end_mask_0, x = v_243_cast_fp16)[name = tensor<string, []>("op_55018_cast_fp16")];
            tensor<int32, [4]> var_55022_begin_0 = const()[name = tensor<string, []>("op_55022_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_55022_end_0 = const()[name = tensor<string, []>("op_55022_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_55022_end_mask_0 = const()[name = tensor<string, []>("op_55022_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55022_cast_fp16 = slice_by_index(begin = var_55022_begin_0, end = var_55022_end_0, end_mask = var_55022_end_mask_0, x = v_243_cast_fp16)[name = tensor<string, []>("op_55022_cast_fp16")];
            tensor<int32, [4]> var_55026_begin_0 = const()[name = tensor<string, []>("op_55026_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_55026_end_0 = const()[name = tensor<string, []>("op_55026_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_55026_end_mask_0 = const()[name = tensor<string, []>("op_55026_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55026_cast_fp16 = slice_by_index(begin = var_55026_begin_0, end = var_55026_end_0, end_mask = var_55026_end_mask_0, x = v_243_cast_fp16)[name = tensor<string, []>("op_55026_cast_fp16")];
            tensor<int32, [4]> var_55030_begin_0 = const()[name = tensor<string, []>("op_55030_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_55030_end_0 = const()[name = tensor<string, []>("op_55030_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_55030_end_mask_0 = const()[name = tensor<string, []>("op_55030_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55030_cast_fp16 = slice_by_index(begin = var_55030_begin_0, end = var_55030_end_0, end_mask = var_55030_end_mask_0, x = v_243_cast_fp16)[name = tensor<string, []>("op_55030_cast_fp16")];
            tensor<int32, [4]> var_55034_begin_0 = const()[name = tensor<string, []>("op_55034_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_55034_end_0 = const()[name = tensor<string, []>("op_55034_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_55034_end_mask_0 = const()[name = tensor<string, []>("op_55034_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55034_cast_fp16 = slice_by_index(begin = var_55034_begin_0, end = var_55034_end_0, end_mask = var_55034_end_mask_0, x = v_243_cast_fp16)[name = tensor<string, []>("op_55034_cast_fp16")];
            tensor<int32, [4]> var_55038_begin_0 = const()[name = tensor<string, []>("op_55038_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_55038_end_0 = const()[name = tensor<string, []>("op_55038_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_55038_end_mask_0 = const()[name = tensor<string, []>("op_55038_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55038_cast_fp16 = slice_by_index(begin = var_55038_begin_0, end = var_55038_end_0, end_mask = var_55038_end_mask_0, x = v_243_cast_fp16)[name = tensor<string, []>("op_55038_cast_fp16")];
            tensor<int32, [4]> var_55042_begin_0 = const()[name = tensor<string, []>("op_55042_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_55042_end_0 = const()[name = tensor<string, []>("op_55042_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_55042_end_mask_0 = const()[name = tensor<string, []>("op_55042_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55042_cast_fp16 = slice_by_index(begin = var_55042_begin_0, end = var_55042_end_0, end_mask = var_55042_end_mask_0, x = v_243_cast_fp16)[name = tensor<string, []>("op_55042_cast_fp16")];
            tensor<int32, [4]> var_55046_begin_0 = const()[name = tensor<string, []>("op_55046_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_55046_end_0 = const()[name = tensor<string, []>("op_55046_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_55046_end_mask_0 = const()[name = tensor<string, []>("op_55046_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55046_cast_fp16 = slice_by_index(begin = var_55046_begin_0, end = var_55046_end_0, end_mask = var_55046_end_mask_0, x = v_243_cast_fp16)[name = tensor<string, []>("op_55046_cast_fp16")];
            tensor<int32, [4]> var_55050_begin_0 = const()[name = tensor<string, []>("op_55050_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_55050_end_0 = const()[name = tensor<string, []>("op_55050_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_55050_end_mask_0 = const()[name = tensor<string, []>("op_55050_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55050_cast_fp16 = slice_by_index(begin = var_55050_begin_0, end = var_55050_end_0, end_mask = var_55050_end_mask_0, x = v_243_cast_fp16)[name = tensor<string, []>("op_55050_cast_fp16")];
            tensor<int32, [4]> var_55054_begin_0 = const()[name = tensor<string, []>("op_55054_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_55054_end_0 = const()[name = tensor<string, []>("op_55054_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_55054_end_mask_0 = const()[name = tensor<string, []>("op_55054_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55054_cast_fp16 = slice_by_index(begin = var_55054_begin_0, end = var_55054_end_0, end_mask = var_55054_end_mask_0, x = v_243_cast_fp16)[name = tensor<string, []>("op_55054_cast_fp16")];
            tensor<int32, [4]> var_55058_begin_0 = const()[name = tensor<string, []>("op_55058_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_55058_end_0 = const()[name = tensor<string, []>("op_55058_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_55058_end_mask_0 = const()[name = tensor<string, []>("op_55058_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55058_cast_fp16 = slice_by_index(begin = var_55058_begin_0, end = var_55058_end_0, end_mask = var_55058_end_mask_0, x = v_243_cast_fp16)[name = tensor<string, []>("op_55058_cast_fp16")];
            tensor<int32, [4]> var_55062_begin_0 = const()[name = tensor<string, []>("op_55062_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_55062_end_0 = const()[name = tensor<string, []>("op_55062_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_55062_end_mask_0 = const()[name = tensor<string, []>("op_55062_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55062_cast_fp16 = slice_by_index(begin = var_55062_begin_0, end = var_55062_end_0, end_mask = var_55062_end_mask_0, x = v_243_cast_fp16)[name = tensor<string, []>("op_55062_cast_fp16")];
            tensor<string, []> var_55066_equation_0 = const()[name = tensor<string, []>("op_55066_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_55066_cast_fp16 = einsum(equation = var_55066_equation_0, values = (var_54908_cast_fp16, var_54825_cast_fp16))[name = tensor<string, []>("op_55066_cast_fp16")];
            tensor<fp16, []> var_55067_to_fp16 = const()[name = tensor<string, []>("op_55067_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4681_cast_fp16 = mul(x = var_55066_cast_fp16, y = var_55067_to_fp16)[name = tensor<string, []>("aw_4681_cast_fp16")];
            tensor<string, []> var_55070_equation_0 = const()[name = tensor<string, []>("op_55070_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_55070_cast_fp16 = einsum(equation = var_55070_equation_0, values = (var_54912_cast_fp16, var_54829_cast_fp16))[name = tensor<string, []>("op_55070_cast_fp16")];
            tensor<fp16, []> var_55071_to_fp16 = const()[name = tensor<string, []>("op_55071_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4683_cast_fp16 = mul(x = var_55070_cast_fp16, y = var_55071_to_fp16)[name = tensor<string, []>("aw_4683_cast_fp16")];
            tensor<string, []> var_55074_equation_0 = const()[name = tensor<string, []>("op_55074_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_55074_cast_fp16 = einsum(equation = var_55074_equation_0, values = (var_54916_cast_fp16, var_54833_cast_fp16))[name = tensor<string, []>("op_55074_cast_fp16")];
            tensor<fp16, []> var_55075_to_fp16 = const()[name = tensor<string, []>("op_55075_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4685_cast_fp16 = mul(x = var_55074_cast_fp16, y = var_55075_to_fp16)[name = tensor<string, []>("aw_4685_cast_fp16")];
            tensor<string, []> var_55078_equation_0 = const()[name = tensor<string, []>("op_55078_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_55078_cast_fp16 = einsum(equation = var_55078_equation_0, values = (var_54920_cast_fp16, var_54837_cast_fp16))[name = tensor<string, []>("op_55078_cast_fp16")];
            tensor<fp16, []> var_55079_to_fp16 = const()[name = tensor<string, []>("op_55079_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4687_cast_fp16 = mul(x = var_55078_cast_fp16, y = var_55079_to_fp16)[name = tensor<string, []>("aw_4687_cast_fp16")];
            tensor<string, []> var_55082_equation_0 = const()[name = tensor<string, []>("op_55082_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_55082_cast_fp16 = einsum(equation = var_55082_equation_0, values = (var_54924_cast_fp16, var_54841_cast_fp16))[name = tensor<string, []>("op_55082_cast_fp16")];
            tensor<fp16, []> var_55083_to_fp16 = const()[name = tensor<string, []>("op_55083_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4689_cast_fp16 = mul(x = var_55082_cast_fp16, y = var_55083_to_fp16)[name = tensor<string, []>("aw_4689_cast_fp16")];
            tensor<string, []> var_55086_equation_0 = const()[name = tensor<string, []>("op_55086_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_55086_cast_fp16 = einsum(equation = var_55086_equation_0, values = (var_54928_cast_fp16, var_54845_cast_fp16))[name = tensor<string, []>("op_55086_cast_fp16")];
            tensor<fp16, []> var_55087_to_fp16 = const()[name = tensor<string, []>("op_55087_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4691_cast_fp16 = mul(x = var_55086_cast_fp16, y = var_55087_to_fp16)[name = tensor<string, []>("aw_4691_cast_fp16")];
            tensor<string, []> var_55090_equation_0 = const()[name = tensor<string, []>("op_55090_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_55090_cast_fp16 = einsum(equation = var_55090_equation_0, values = (var_54932_cast_fp16, var_54849_cast_fp16))[name = tensor<string, []>("op_55090_cast_fp16")];
            tensor<fp16, []> var_55091_to_fp16 = const()[name = tensor<string, []>("op_55091_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4693_cast_fp16 = mul(x = var_55090_cast_fp16, y = var_55091_to_fp16)[name = tensor<string, []>("aw_4693_cast_fp16")];
            tensor<string, []> var_55094_equation_0 = const()[name = tensor<string, []>("op_55094_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_55094_cast_fp16 = einsum(equation = var_55094_equation_0, values = (var_54936_cast_fp16, var_54853_cast_fp16))[name = tensor<string, []>("op_55094_cast_fp16")];
            tensor<fp16, []> var_55095_to_fp16 = const()[name = tensor<string, []>("op_55095_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4695_cast_fp16 = mul(x = var_55094_cast_fp16, y = var_55095_to_fp16)[name = tensor<string, []>("aw_4695_cast_fp16")];
            tensor<string, []> var_55098_equation_0 = const()[name = tensor<string, []>("op_55098_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_55098_cast_fp16 = einsum(equation = var_55098_equation_0, values = (var_54940_cast_fp16, var_54857_cast_fp16))[name = tensor<string, []>("op_55098_cast_fp16")];
            tensor<fp16, []> var_55099_to_fp16 = const()[name = tensor<string, []>("op_55099_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4697_cast_fp16 = mul(x = var_55098_cast_fp16, y = var_55099_to_fp16)[name = tensor<string, []>("aw_4697_cast_fp16")];
            tensor<string, []> var_55102_equation_0 = const()[name = tensor<string, []>("op_55102_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_55102_cast_fp16 = einsum(equation = var_55102_equation_0, values = (var_54944_cast_fp16, var_54861_cast_fp16))[name = tensor<string, []>("op_55102_cast_fp16")];
            tensor<fp16, []> var_55103_to_fp16 = const()[name = tensor<string, []>("op_55103_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4699_cast_fp16 = mul(x = var_55102_cast_fp16, y = var_55103_to_fp16)[name = tensor<string, []>("aw_4699_cast_fp16")];
            tensor<string, []> var_55106_equation_0 = const()[name = tensor<string, []>("op_55106_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_55106_cast_fp16 = einsum(equation = var_55106_equation_0, values = (var_54948_cast_fp16, var_54865_cast_fp16))[name = tensor<string, []>("op_55106_cast_fp16")];
            tensor<fp16, []> var_55107_to_fp16 = const()[name = tensor<string, []>("op_55107_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4701_cast_fp16 = mul(x = var_55106_cast_fp16, y = var_55107_to_fp16)[name = tensor<string, []>("aw_4701_cast_fp16")];
            tensor<string, []> var_55110_equation_0 = const()[name = tensor<string, []>("op_55110_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_55110_cast_fp16 = einsum(equation = var_55110_equation_0, values = (var_54952_cast_fp16, var_54869_cast_fp16))[name = tensor<string, []>("op_55110_cast_fp16")];
            tensor<fp16, []> var_55111_to_fp16 = const()[name = tensor<string, []>("op_55111_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4703_cast_fp16 = mul(x = var_55110_cast_fp16, y = var_55111_to_fp16)[name = tensor<string, []>("aw_4703_cast_fp16")];
            tensor<string, []> var_55114_equation_0 = const()[name = tensor<string, []>("op_55114_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_55114_cast_fp16 = einsum(equation = var_55114_equation_0, values = (var_54956_cast_fp16, var_54873_cast_fp16))[name = tensor<string, []>("op_55114_cast_fp16")];
            tensor<fp16, []> var_55115_to_fp16 = const()[name = tensor<string, []>("op_55115_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4705_cast_fp16 = mul(x = var_55114_cast_fp16, y = var_55115_to_fp16)[name = tensor<string, []>("aw_4705_cast_fp16")];
            tensor<string, []> var_55118_equation_0 = const()[name = tensor<string, []>("op_55118_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_55118_cast_fp16 = einsum(equation = var_55118_equation_0, values = (var_54960_cast_fp16, var_54877_cast_fp16))[name = tensor<string, []>("op_55118_cast_fp16")];
            tensor<fp16, []> var_55119_to_fp16 = const()[name = tensor<string, []>("op_55119_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4707_cast_fp16 = mul(x = var_55118_cast_fp16, y = var_55119_to_fp16)[name = tensor<string, []>("aw_4707_cast_fp16")];
            tensor<string, []> var_55122_equation_0 = const()[name = tensor<string, []>("op_55122_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_55122_cast_fp16 = einsum(equation = var_55122_equation_0, values = (var_54964_cast_fp16, var_54881_cast_fp16))[name = tensor<string, []>("op_55122_cast_fp16")];
            tensor<fp16, []> var_55123_to_fp16 = const()[name = tensor<string, []>("op_55123_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4709_cast_fp16 = mul(x = var_55122_cast_fp16, y = var_55123_to_fp16)[name = tensor<string, []>("aw_4709_cast_fp16")];
            tensor<string, []> var_55126_equation_0 = const()[name = tensor<string, []>("op_55126_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_55126_cast_fp16 = einsum(equation = var_55126_equation_0, values = (var_54968_cast_fp16, var_54885_cast_fp16))[name = tensor<string, []>("op_55126_cast_fp16")];
            tensor<fp16, []> var_55127_to_fp16 = const()[name = tensor<string, []>("op_55127_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4711_cast_fp16 = mul(x = var_55126_cast_fp16, y = var_55127_to_fp16)[name = tensor<string, []>("aw_4711_cast_fp16")];
            tensor<string, []> var_55130_equation_0 = const()[name = tensor<string, []>("op_55130_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_55130_cast_fp16 = einsum(equation = var_55130_equation_0, values = (var_54972_cast_fp16, var_54889_cast_fp16))[name = tensor<string, []>("op_55130_cast_fp16")];
            tensor<fp16, []> var_55131_to_fp16 = const()[name = tensor<string, []>("op_55131_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4713_cast_fp16 = mul(x = var_55130_cast_fp16, y = var_55131_to_fp16)[name = tensor<string, []>("aw_4713_cast_fp16")];
            tensor<string, []> var_55134_equation_0 = const()[name = tensor<string, []>("op_55134_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_55134_cast_fp16 = einsum(equation = var_55134_equation_0, values = (var_54976_cast_fp16, var_54893_cast_fp16))[name = tensor<string, []>("op_55134_cast_fp16")];
            tensor<fp16, []> var_55135_to_fp16 = const()[name = tensor<string, []>("op_55135_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4715_cast_fp16 = mul(x = var_55134_cast_fp16, y = var_55135_to_fp16)[name = tensor<string, []>("aw_4715_cast_fp16")];
            tensor<string, []> var_55138_equation_0 = const()[name = tensor<string, []>("op_55138_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_55138_cast_fp16 = einsum(equation = var_55138_equation_0, values = (var_54980_cast_fp16, var_54897_cast_fp16))[name = tensor<string, []>("op_55138_cast_fp16")];
            tensor<fp16, []> var_55139_to_fp16 = const()[name = tensor<string, []>("op_55139_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4717_cast_fp16 = mul(x = var_55138_cast_fp16, y = var_55139_to_fp16)[name = tensor<string, []>("aw_4717_cast_fp16")];
            tensor<string, []> var_55142_equation_0 = const()[name = tensor<string, []>("op_55142_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_55142_cast_fp16 = einsum(equation = var_55142_equation_0, values = (var_54984_cast_fp16, var_54901_cast_fp16))[name = tensor<string, []>("op_55142_cast_fp16")];
            tensor<fp16, []> var_55143_to_fp16 = const()[name = tensor<string, []>("op_55143_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4719_cast_fp16 = mul(x = var_55142_cast_fp16, y = var_55143_to_fp16)[name = tensor<string, []>("aw_4719_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_55145_cast_fp16 = softmax(axis = var_30355, x = aw_4681_cast_fp16)[name = tensor<string, []>("op_55145_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_55146_cast_fp16 = softmax(axis = var_30355, x = aw_4683_cast_fp16)[name = tensor<string, []>("op_55146_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_55147_cast_fp16 = softmax(axis = var_30355, x = aw_4685_cast_fp16)[name = tensor<string, []>("op_55147_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_55148_cast_fp16 = softmax(axis = var_30355, x = aw_4687_cast_fp16)[name = tensor<string, []>("op_55148_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_55149_cast_fp16 = softmax(axis = var_30355, x = aw_4689_cast_fp16)[name = tensor<string, []>("op_55149_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_55150_cast_fp16 = softmax(axis = var_30355, x = aw_4691_cast_fp16)[name = tensor<string, []>("op_55150_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_55151_cast_fp16 = softmax(axis = var_30355, x = aw_4693_cast_fp16)[name = tensor<string, []>("op_55151_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_55152_cast_fp16 = softmax(axis = var_30355, x = aw_4695_cast_fp16)[name = tensor<string, []>("op_55152_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_55153_cast_fp16 = softmax(axis = var_30355, x = aw_4697_cast_fp16)[name = tensor<string, []>("op_55153_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_55154_cast_fp16 = softmax(axis = var_30355, x = aw_4699_cast_fp16)[name = tensor<string, []>("op_55154_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_55155_cast_fp16 = softmax(axis = var_30355, x = aw_4701_cast_fp16)[name = tensor<string, []>("op_55155_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_55156_cast_fp16 = softmax(axis = var_30355, x = aw_4703_cast_fp16)[name = tensor<string, []>("op_55156_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_55157_cast_fp16 = softmax(axis = var_30355, x = aw_4705_cast_fp16)[name = tensor<string, []>("op_55157_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_55158_cast_fp16 = softmax(axis = var_30355, x = aw_4707_cast_fp16)[name = tensor<string, []>("op_55158_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_55159_cast_fp16 = softmax(axis = var_30355, x = aw_4709_cast_fp16)[name = tensor<string, []>("op_55159_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_55160_cast_fp16 = softmax(axis = var_30355, x = aw_4711_cast_fp16)[name = tensor<string, []>("op_55160_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_55161_cast_fp16 = softmax(axis = var_30355, x = aw_4713_cast_fp16)[name = tensor<string, []>("op_55161_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_55162_cast_fp16 = softmax(axis = var_30355, x = aw_4715_cast_fp16)[name = tensor<string, []>("op_55162_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_55163_cast_fp16 = softmax(axis = var_30355, x = aw_4717_cast_fp16)[name = tensor<string, []>("op_55163_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_55164_cast_fp16 = softmax(axis = var_30355, x = aw_4719_cast_fp16)[name = tensor<string, []>("op_55164_cast_fp16")];
            tensor<string, []> var_55166_equation_0 = const()[name = tensor<string, []>("op_55166_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55166_cast_fp16 = einsum(equation = var_55166_equation_0, values = (var_54986_cast_fp16, var_55145_cast_fp16))[name = tensor<string, []>("op_55166_cast_fp16")];
            tensor<string, []> var_55168_equation_0 = const()[name = tensor<string, []>("op_55168_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55168_cast_fp16 = einsum(equation = var_55168_equation_0, values = (var_54990_cast_fp16, var_55146_cast_fp16))[name = tensor<string, []>("op_55168_cast_fp16")];
            tensor<string, []> var_55170_equation_0 = const()[name = tensor<string, []>("op_55170_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55170_cast_fp16 = einsum(equation = var_55170_equation_0, values = (var_54994_cast_fp16, var_55147_cast_fp16))[name = tensor<string, []>("op_55170_cast_fp16")];
            tensor<string, []> var_55172_equation_0 = const()[name = tensor<string, []>("op_55172_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55172_cast_fp16 = einsum(equation = var_55172_equation_0, values = (var_54998_cast_fp16, var_55148_cast_fp16))[name = tensor<string, []>("op_55172_cast_fp16")];
            tensor<string, []> var_55174_equation_0 = const()[name = tensor<string, []>("op_55174_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55174_cast_fp16 = einsum(equation = var_55174_equation_0, values = (var_55002_cast_fp16, var_55149_cast_fp16))[name = tensor<string, []>("op_55174_cast_fp16")];
            tensor<string, []> var_55176_equation_0 = const()[name = tensor<string, []>("op_55176_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55176_cast_fp16 = einsum(equation = var_55176_equation_0, values = (var_55006_cast_fp16, var_55150_cast_fp16))[name = tensor<string, []>("op_55176_cast_fp16")];
            tensor<string, []> var_55178_equation_0 = const()[name = tensor<string, []>("op_55178_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55178_cast_fp16 = einsum(equation = var_55178_equation_0, values = (var_55010_cast_fp16, var_55151_cast_fp16))[name = tensor<string, []>("op_55178_cast_fp16")];
            tensor<string, []> var_55180_equation_0 = const()[name = tensor<string, []>("op_55180_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55180_cast_fp16 = einsum(equation = var_55180_equation_0, values = (var_55014_cast_fp16, var_55152_cast_fp16))[name = tensor<string, []>("op_55180_cast_fp16")];
            tensor<string, []> var_55182_equation_0 = const()[name = tensor<string, []>("op_55182_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55182_cast_fp16 = einsum(equation = var_55182_equation_0, values = (var_55018_cast_fp16, var_55153_cast_fp16))[name = tensor<string, []>("op_55182_cast_fp16")];
            tensor<string, []> var_55184_equation_0 = const()[name = tensor<string, []>("op_55184_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55184_cast_fp16 = einsum(equation = var_55184_equation_0, values = (var_55022_cast_fp16, var_55154_cast_fp16))[name = tensor<string, []>("op_55184_cast_fp16")];
            tensor<string, []> var_55186_equation_0 = const()[name = tensor<string, []>("op_55186_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55186_cast_fp16 = einsum(equation = var_55186_equation_0, values = (var_55026_cast_fp16, var_55155_cast_fp16))[name = tensor<string, []>("op_55186_cast_fp16")];
            tensor<string, []> var_55188_equation_0 = const()[name = tensor<string, []>("op_55188_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55188_cast_fp16 = einsum(equation = var_55188_equation_0, values = (var_55030_cast_fp16, var_55156_cast_fp16))[name = tensor<string, []>("op_55188_cast_fp16")];
            tensor<string, []> var_55190_equation_0 = const()[name = tensor<string, []>("op_55190_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55190_cast_fp16 = einsum(equation = var_55190_equation_0, values = (var_55034_cast_fp16, var_55157_cast_fp16))[name = tensor<string, []>("op_55190_cast_fp16")];
            tensor<string, []> var_55192_equation_0 = const()[name = tensor<string, []>("op_55192_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55192_cast_fp16 = einsum(equation = var_55192_equation_0, values = (var_55038_cast_fp16, var_55158_cast_fp16))[name = tensor<string, []>("op_55192_cast_fp16")];
            tensor<string, []> var_55194_equation_0 = const()[name = tensor<string, []>("op_55194_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55194_cast_fp16 = einsum(equation = var_55194_equation_0, values = (var_55042_cast_fp16, var_55159_cast_fp16))[name = tensor<string, []>("op_55194_cast_fp16")];
            tensor<string, []> var_55196_equation_0 = const()[name = tensor<string, []>("op_55196_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55196_cast_fp16 = einsum(equation = var_55196_equation_0, values = (var_55046_cast_fp16, var_55160_cast_fp16))[name = tensor<string, []>("op_55196_cast_fp16")];
            tensor<string, []> var_55198_equation_0 = const()[name = tensor<string, []>("op_55198_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55198_cast_fp16 = einsum(equation = var_55198_equation_0, values = (var_55050_cast_fp16, var_55161_cast_fp16))[name = tensor<string, []>("op_55198_cast_fp16")];
            tensor<string, []> var_55200_equation_0 = const()[name = tensor<string, []>("op_55200_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55200_cast_fp16 = einsum(equation = var_55200_equation_0, values = (var_55054_cast_fp16, var_55162_cast_fp16))[name = tensor<string, []>("op_55200_cast_fp16")];
            tensor<string, []> var_55202_equation_0 = const()[name = tensor<string, []>("op_55202_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55202_cast_fp16 = einsum(equation = var_55202_equation_0, values = (var_55058_cast_fp16, var_55163_cast_fp16))[name = tensor<string, []>("op_55202_cast_fp16")];
            tensor<string, []> var_55204_equation_0 = const()[name = tensor<string, []>("op_55204_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55204_cast_fp16 = einsum(equation = var_55204_equation_0, values = (var_55062_cast_fp16, var_55164_cast_fp16))[name = tensor<string, []>("op_55204_cast_fp16")];
            tensor<bool, []> input_685_interleave_0 = const()[name = tensor<string, []>("input_685_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_685_cast_fp16 = concat(axis = var_30355, interleave = input_685_interleave_0, values = (var_55166_cast_fp16, var_55168_cast_fp16, var_55170_cast_fp16, var_55172_cast_fp16, var_55174_cast_fp16, var_55176_cast_fp16, var_55178_cast_fp16, var_55180_cast_fp16, var_55182_cast_fp16, var_55184_cast_fp16, var_55186_cast_fp16, var_55188_cast_fp16, var_55190_cast_fp16, var_55192_cast_fp16, var_55194_cast_fp16, var_55196_cast_fp16, var_55198_cast_fp16, var_55200_cast_fp16, var_55202_cast_fp16, var_55204_cast_fp16))[name = tensor<string, []>("input_685_cast_fp16")];
            tensor<string, []> var_55214_pad_type_0 = const()[name = tensor<string, []>("op_55214_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_55214_strides_0 = const()[name = tensor<string, []>("op_55214_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_55214_pad_0 = const()[name = tensor<string, []>("op_55214_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_55214_dilations_0 = const()[name = tensor<string, []>("op_55214_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_55214_groups_0 = const()[name = tensor<string, []>("op_55214_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_6_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(764918272))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(766147136))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_6_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_6_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_6_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(766147328)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_55214_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_6_attn2_to_out_0_bias_to_fp16, dilations = var_55214_dilations_0, groups = var_55214_groups_0, pad = var_55214_pad_0, pad_type = var_55214_pad_type_0, strides = var_55214_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_6_attn2_to_out_0_weight_to_fp16_palettized, x = input_685_cast_fp16)[name = tensor<string, []>("op_55214_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_365_cast_fp16 = add(x = var_55214_cast_fp16, y = inputs_363_cast_fp16)[name = tensor<string, []>("inputs_365_cast_fp16")];
            tensor<int32, [1]> input_687_axes_0 = const()[name = tensor<string, []>("input_687_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> input_687_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_687_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(766149952)))];
            tensor<fp16, [1280]> input_687_beta_0_to_fp16 = const()[name = tensor<string, []>("input_687_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(766152576)))];
            tensor<fp16, []> var_55224_to_fp16 = const()[name = tensor<string, []>("op_55224_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> input_687_cast_fp16 = layer_norm(axes = input_687_axes_0, beta = input_687_beta_0_to_fp16, epsilon = var_55224_to_fp16, gamma = input_687_gamma_0_to_fp16, x = inputs_365_cast_fp16)[name = tensor<string, []>("input_687_cast_fp16")];
            tensor<string, []> var_55244_pad_type_0 = const()[name = tensor<string, []>("op_55244_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_55244_strides_0 = const()[name = tensor<string, []>("op_55244_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_55244_pad_0 = const()[name = tensor<string, []>("op_55244_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_55244_dilations_0 = const()[name = tensor<string, []>("op_55244_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_55244_groups_0 = const()[name = tensor<string, []>("op_55244_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_6_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [9830400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(766155200))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(775985664))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_6_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([10240, 1280, 1, 1])];
            tensor<fp16, [10240]> up_blocks_0_attentions_2_transformer_blocks_6_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_6_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(775985856)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_55244_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_6_ff_net_0_proj_bias_to_fp16, dilations = var_55244_dilations_0, groups = var_55244_groups_0, pad = var_55244_pad_0, pad_type = var_55244_pad_type_0, strides = var_55244_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_6_ff_net_0_proj_weight_to_fp16_palettized, x = input_687_cast_fp16)[name = tensor<string, []>("op_55244_cast_fp16")];
            tensor<int32, [2]> var_55245_split_sizes_0 = const()[name = tensor<string, []>("op_55245_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_55245_axis_0 = const()[name = tensor<string, []>("op_55245_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_55245_cast_fp16_0, tensor<fp16, [2, 5120, 1, 1024]> var_55245_cast_fp16_1 = split(axis = var_55245_axis_0, split_sizes = var_55245_split_sizes_0, x = var_55244_cast_fp16)[name = tensor<string, []>("op_55245_cast_fp16")];
            tensor<string, []> var_55247_mode_0 = const()[name = tensor<string, []>("op_55247_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_55247_cast_fp16 = gelu(mode = var_55247_mode_0, x = var_55245_cast_fp16_1)[name = tensor<string, []>("op_55247_cast_fp16")];
            tensor<fp16, [2, 5120, 1, 1024]> input_689_cast_fp16 = mul(x = var_55245_cast_fp16_0, y = var_55247_cast_fp16)[name = tensor<string, []>("input_689_cast_fp16")];
            tensor<string, []> var_55255_pad_type_0 = const()[name = tensor<string, []>("op_55255_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_55255_strides_0 = const()[name = tensor<string, []>("op_55255_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_55255_pad_0 = const()[name = tensor<string, []>("op_55255_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_55255_dilations_0 = const()[name = tensor<string, []>("op_55255_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_55255_groups_0 = const()[name = tensor<string, []>("op_55255_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_6_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [4915200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(776006400))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(780921664))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_6_ff_net_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 5120, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_6_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_6_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(780921856)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_55255_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_6_ff_net_2_bias_to_fp16, dilations = var_55255_dilations_0, groups = var_55255_groups_0, pad = var_55255_pad_0, pad_type = var_55255_pad_type_0, strides = var_55255_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_6_ff_net_2_weight_to_fp16_palettized, x = input_689_cast_fp16)[name = tensor<string, []>("op_55255_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_367_cast_fp16 = add(x = var_55255_cast_fp16, y = inputs_365_cast_fp16)[name = tensor<string, []>("inputs_367_cast_fp16")];
            tensor<int32, [1]> hidden_states_477_axes_0 = const()[name = tensor<string, []>("hidden_states_477_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_477_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_477_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(780924480)))];
            tensor<fp16, [1280]> hidden_states_477_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_477_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(780927104)))];
            tensor<fp16, []> var_55271_to_fp16 = const()[name = tensor<string, []>("op_55271_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_477_cast_fp16 = layer_norm(axes = hidden_states_477_axes_0, beta = hidden_states_477_beta_0_to_fp16, epsilon = var_55271_to_fp16, gamma = hidden_states_477_gamma_0_to_fp16, x = inputs_367_cast_fp16)[name = tensor<string, []>("hidden_states_477_cast_fp16")];
            tensor<string, []> q_245_pad_type_0 = const()[name = tensor<string, []>("q_245_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_245_strides_0 = const()[name = tensor<string, []>("q_245_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_245_pad_0 = const()[name = tensor<string, []>("q_245_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_245_dilations_0 = const()[name = tensor<string, []>("q_245_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_245_groups_0 = const()[name = tensor<string, []>("q_245_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_7_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(780929728))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(782158592))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_7_attn1_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_245_cast_fp16 = conv(dilations = q_245_dilations_0, groups = q_245_groups_0, pad = q_245_pad_0, pad_type = q_245_pad_type_0, strides = q_245_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_7_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_477_cast_fp16)[name = tensor<string, []>("q_245_cast_fp16")];
            tensor<string, []> k_489_pad_type_0 = const()[name = tensor<string, []>("k_489_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_489_strides_0 = const()[name = tensor<string, []>("k_489_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_489_pad_0 = const()[name = tensor<string, []>("k_489_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_489_dilations_0 = const()[name = tensor<string, []>("k_489_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_489_groups_0 = const()[name = tensor<string, []>("k_489_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_7_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(782158784))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(783387648))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_7_attn1_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> k_489_cast_fp16 = conv(dilations = k_489_dilations_0, groups = k_489_groups_0, pad = k_489_pad_0, pad_type = k_489_pad_type_0, strides = k_489_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_7_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_477_cast_fp16)[name = tensor<string, []>("k_489_cast_fp16")];
            tensor<string, []> v_245_pad_type_0 = const()[name = tensor<string, []>("v_245_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_245_strides_0 = const()[name = tensor<string, []>("v_245_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_245_pad_0 = const()[name = tensor<string, []>("v_245_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_245_dilations_0 = const()[name = tensor<string, []>("v_245_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_245_groups_0 = const()[name = tensor<string, []>("v_245_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_7_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(783387840))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(784616704))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_7_attn1_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> v_245_cast_fp16 = conv(dilations = v_245_dilations_0, groups = v_245_groups_0, pad = v_245_pad_0, pad_type = v_245_pad_type_0, strides = v_245_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_7_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_477_cast_fp16)[name = tensor<string, []>("v_245_cast_fp16")];
            tensor<int32, [4]> var_55304_begin_0 = const()[name = tensor<string, []>("op_55304_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_55304_end_0 = const()[name = tensor<string, []>("op_55304_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_55304_end_mask_0 = const()[name = tensor<string, []>("op_55304_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55304_cast_fp16 = slice_by_index(begin = var_55304_begin_0, end = var_55304_end_0, end_mask = var_55304_end_mask_0, x = q_245_cast_fp16)[name = tensor<string, []>("op_55304_cast_fp16")];
            tensor<int32, [4]> var_55308_begin_0 = const()[name = tensor<string, []>("op_55308_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_55308_end_0 = const()[name = tensor<string, []>("op_55308_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_55308_end_mask_0 = const()[name = tensor<string, []>("op_55308_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55308_cast_fp16 = slice_by_index(begin = var_55308_begin_0, end = var_55308_end_0, end_mask = var_55308_end_mask_0, x = q_245_cast_fp16)[name = tensor<string, []>("op_55308_cast_fp16")];
            tensor<int32, [4]> var_55312_begin_0 = const()[name = tensor<string, []>("op_55312_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_55312_end_0 = const()[name = tensor<string, []>("op_55312_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_55312_end_mask_0 = const()[name = tensor<string, []>("op_55312_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55312_cast_fp16 = slice_by_index(begin = var_55312_begin_0, end = var_55312_end_0, end_mask = var_55312_end_mask_0, x = q_245_cast_fp16)[name = tensor<string, []>("op_55312_cast_fp16")];
            tensor<int32, [4]> var_55316_begin_0 = const()[name = tensor<string, []>("op_55316_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_55316_end_0 = const()[name = tensor<string, []>("op_55316_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_55316_end_mask_0 = const()[name = tensor<string, []>("op_55316_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55316_cast_fp16 = slice_by_index(begin = var_55316_begin_0, end = var_55316_end_0, end_mask = var_55316_end_mask_0, x = q_245_cast_fp16)[name = tensor<string, []>("op_55316_cast_fp16")];
            tensor<int32, [4]> var_55320_begin_0 = const()[name = tensor<string, []>("op_55320_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_55320_end_0 = const()[name = tensor<string, []>("op_55320_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_55320_end_mask_0 = const()[name = tensor<string, []>("op_55320_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55320_cast_fp16 = slice_by_index(begin = var_55320_begin_0, end = var_55320_end_0, end_mask = var_55320_end_mask_0, x = q_245_cast_fp16)[name = tensor<string, []>("op_55320_cast_fp16")];
            tensor<int32, [4]> var_55324_begin_0 = const()[name = tensor<string, []>("op_55324_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_55324_end_0 = const()[name = tensor<string, []>("op_55324_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_55324_end_mask_0 = const()[name = tensor<string, []>("op_55324_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55324_cast_fp16 = slice_by_index(begin = var_55324_begin_0, end = var_55324_end_0, end_mask = var_55324_end_mask_0, x = q_245_cast_fp16)[name = tensor<string, []>("op_55324_cast_fp16")];
            tensor<int32, [4]> var_55328_begin_0 = const()[name = tensor<string, []>("op_55328_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_55328_end_0 = const()[name = tensor<string, []>("op_55328_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_55328_end_mask_0 = const()[name = tensor<string, []>("op_55328_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55328_cast_fp16 = slice_by_index(begin = var_55328_begin_0, end = var_55328_end_0, end_mask = var_55328_end_mask_0, x = q_245_cast_fp16)[name = tensor<string, []>("op_55328_cast_fp16")];
            tensor<int32, [4]> var_55332_begin_0 = const()[name = tensor<string, []>("op_55332_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_55332_end_0 = const()[name = tensor<string, []>("op_55332_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_55332_end_mask_0 = const()[name = tensor<string, []>("op_55332_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55332_cast_fp16 = slice_by_index(begin = var_55332_begin_0, end = var_55332_end_0, end_mask = var_55332_end_mask_0, x = q_245_cast_fp16)[name = tensor<string, []>("op_55332_cast_fp16")];
            tensor<int32, [4]> var_55336_begin_0 = const()[name = tensor<string, []>("op_55336_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_55336_end_0 = const()[name = tensor<string, []>("op_55336_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_55336_end_mask_0 = const()[name = tensor<string, []>("op_55336_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55336_cast_fp16 = slice_by_index(begin = var_55336_begin_0, end = var_55336_end_0, end_mask = var_55336_end_mask_0, x = q_245_cast_fp16)[name = tensor<string, []>("op_55336_cast_fp16")];
            tensor<int32, [4]> var_55340_begin_0 = const()[name = tensor<string, []>("op_55340_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_55340_end_0 = const()[name = tensor<string, []>("op_55340_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_55340_end_mask_0 = const()[name = tensor<string, []>("op_55340_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55340_cast_fp16 = slice_by_index(begin = var_55340_begin_0, end = var_55340_end_0, end_mask = var_55340_end_mask_0, x = q_245_cast_fp16)[name = tensor<string, []>("op_55340_cast_fp16")];
            tensor<int32, [4]> var_55344_begin_0 = const()[name = tensor<string, []>("op_55344_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_55344_end_0 = const()[name = tensor<string, []>("op_55344_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_55344_end_mask_0 = const()[name = tensor<string, []>("op_55344_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55344_cast_fp16 = slice_by_index(begin = var_55344_begin_0, end = var_55344_end_0, end_mask = var_55344_end_mask_0, x = q_245_cast_fp16)[name = tensor<string, []>("op_55344_cast_fp16")];
            tensor<int32, [4]> var_55348_begin_0 = const()[name = tensor<string, []>("op_55348_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_55348_end_0 = const()[name = tensor<string, []>("op_55348_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_55348_end_mask_0 = const()[name = tensor<string, []>("op_55348_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55348_cast_fp16 = slice_by_index(begin = var_55348_begin_0, end = var_55348_end_0, end_mask = var_55348_end_mask_0, x = q_245_cast_fp16)[name = tensor<string, []>("op_55348_cast_fp16")];
            tensor<int32, [4]> var_55352_begin_0 = const()[name = tensor<string, []>("op_55352_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_55352_end_0 = const()[name = tensor<string, []>("op_55352_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_55352_end_mask_0 = const()[name = tensor<string, []>("op_55352_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55352_cast_fp16 = slice_by_index(begin = var_55352_begin_0, end = var_55352_end_0, end_mask = var_55352_end_mask_0, x = q_245_cast_fp16)[name = tensor<string, []>("op_55352_cast_fp16")];
            tensor<int32, [4]> var_55356_begin_0 = const()[name = tensor<string, []>("op_55356_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_55356_end_0 = const()[name = tensor<string, []>("op_55356_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_55356_end_mask_0 = const()[name = tensor<string, []>("op_55356_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55356_cast_fp16 = slice_by_index(begin = var_55356_begin_0, end = var_55356_end_0, end_mask = var_55356_end_mask_0, x = q_245_cast_fp16)[name = tensor<string, []>("op_55356_cast_fp16")];
            tensor<int32, [4]> var_55360_begin_0 = const()[name = tensor<string, []>("op_55360_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_55360_end_0 = const()[name = tensor<string, []>("op_55360_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_55360_end_mask_0 = const()[name = tensor<string, []>("op_55360_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55360_cast_fp16 = slice_by_index(begin = var_55360_begin_0, end = var_55360_end_0, end_mask = var_55360_end_mask_0, x = q_245_cast_fp16)[name = tensor<string, []>("op_55360_cast_fp16")];
            tensor<int32, [4]> var_55364_begin_0 = const()[name = tensor<string, []>("op_55364_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_55364_end_0 = const()[name = tensor<string, []>("op_55364_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_55364_end_mask_0 = const()[name = tensor<string, []>("op_55364_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55364_cast_fp16 = slice_by_index(begin = var_55364_begin_0, end = var_55364_end_0, end_mask = var_55364_end_mask_0, x = q_245_cast_fp16)[name = tensor<string, []>("op_55364_cast_fp16")];
            tensor<int32, [4]> var_55368_begin_0 = const()[name = tensor<string, []>("op_55368_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_55368_end_0 = const()[name = tensor<string, []>("op_55368_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_55368_end_mask_0 = const()[name = tensor<string, []>("op_55368_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55368_cast_fp16 = slice_by_index(begin = var_55368_begin_0, end = var_55368_end_0, end_mask = var_55368_end_mask_0, x = q_245_cast_fp16)[name = tensor<string, []>("op_55368_cast_fp16")];
            tensor<int32, [4]> var_55372_begin_0 = const()[name = tensor<string, []>("op_55372_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_55372_end_0 = const()[name = tensor<string, []>("op_55372_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_55372_end_mask_0 = const()[name = tensor<string, []>("op_55372_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55372_cast_fp16 = slice_by_index(begin = var_55372_begin_0, end = var_55372_end_0, end_mask = var_55372_end_mask_0, x = q_245_cast_fp16)[name = tensor<string, []>("op_55372_cast_fp16")];
            tensor<int32, [4]> var_55376_begin_0 = const()[name = tensor<string, []>("op_55376_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_55376_end_0 = const()[name = tensor<string, []>("op_55376_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_55376_end_mask_0 = const()[name = tensor<string, []>("op_55376_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55376_cast_fp16 = slice_by_index(begin = var_55376_begin_0, end = var_55376_end_0, end_mask = var_55376_end_mask_0, x = q_245_cast_fp16)[name = tensor<string, []>("op_55376_cast_fp16")];
            tensor<int32, [4]> var_55380_begin_0 = const()[name = tensor<string, []>("op_55380_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_55380_end_0 = const()[name = tensor<string, []>("op_55380_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_55380_end_mask_0 = const()[name = tensor<string, []>("op_55380_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55380_cast_fp16 = slice_by_index(begin = var_55380_begin_0, end = var_55380_end_0, end_mask = var_55380_end_mask_0, x = q_245_cast_fp16)[name = tensor<string, []>("op_55380_cast_fp16")];
            tensor<int32, [4]> k_491_perm_0 = const()[name = tensor<string, []>("k_491_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_55387_begin_0 = const()[name = tensor<string, []>("op_55387_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_55387_end_0 = const()[name = tensor<string, []>("op_55387_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_55387_end_mask_0 = const()[name = tensor<string, []>("op_55387_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> k_491_cast_fp16 = transpose(perm = k_491_perm_0, x = k_489_cast_fp16)[name = tensor<string, []>("transpose_17")];
            tensor<fp16, [2, 1024, 1, 64]> var_55387_cast_fp16 = slice_by_index(begin = var_55387_begin_0, end = var_55387_end_0, end_mask = var_55387_end_mask_0, x = k_491_cast_fp16)[name = tensor<string, []>("op_55387_cast_fp16")];
            tensor<int32, [4]> var_55391_begin_0 = const()[name = tensor<string, []>("op_55391_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_55391_end_0 = const()[name = tensor<string, []>("op_55391_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_55391_end_mask_0 = const()[name = tensor<string, []>("op_55391_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_55391_cast_fp16 = slice_by_index(begin = var_55391_begin_0, end = var_55391_end_0, end_mask = var_55391_end_mask_0, x = k_491_cast_fp16)[name = tensor<string, []>("op_55391_cast_fp16")];
            tensor<int32, [4]> var_55395_begin_0 = const()[name = tensor<string, []>("op_55395_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_55395_end_0 = const()[name = tensor<string, []>("op_55395_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_55395_end_mask_0 = const()[name = tensor<string, []>("op_55395_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_55395_cast_fp16 = slice_by_index(begin = var_55395_begin_0, end = var_55395_end_0, end_mask = var_55395_end_mask_0, x = k_491_cast_fp16)[name = tensor<string, []>("op_55395_cast_fp16")];
            tensor<int32, [4]> var_55399_begin_0 = const()[name = tensor<string, []>("op_55399_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_55399_end_0 = const()[name = tensor<string, []>("op_55399_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_55399_end_mask_0 = const()[name = tensor<string, []>("op_55399_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_55399_cast_fp16 = slice_by_index(begin = var_55399_begin_0, end = var_55399_end_0, end_mask = var_55399_end_mask_0, x = k_491_cast_fp16)[name = tensor<string, []>("op_55399_cast_fp16")];
            tensor<int32, [4]> var_55403_begin_0 = const()[name = tensor<string, []>("op_55403_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_55403_end_0 = const()[name = tensor<string, []>("op_55403_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_55403_end_mask_0 = const()[name = tensor<string, []>("op_55403_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_55403_cast_fp16 = slice_by_index(begin = var_55403_begin_0, end = var_55403_end_0, end_mask = var_55403_end_mask_0, x = k_491_cast_fp16)[name = tensor<string, []>("op_55403_cast_fp16")];
            tensor<int32, [4]> var_55407_begin_0 = const()[name = tensor<string, []>("op_55407_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_55407_end_0 = const()[name = tensor<string, []>("op_55407_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_55407_end_mask_0 = const()[name = tensor<string, []>("op_55407_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_55407_cast_fp16 = slice_by_index(begin = var_55407_begin_0, end = var_55407_end_0, end_mask = var_55407_end_mask_0, x = k_491_cast_fp16)[name = tensor<string, []>("op_55407_cast_fp16")];
            tensor<int32, [4]> var_55411_begin_0 = const()[name = tensor<string, []>("op_55411_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_55411_end_0 = const()[name = tensor<string, []>("op_55411_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_55411_end_mask_0 = const()[name = tensor<string, []>("op_55411_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_55411_cast_fp16 = slice_by_index(begin = var_55411_begin_0, end = var_55411_end_0, end_mask = var_55411_end_mask_0, x = k_491_cast_fp16)[name = tensor<string, []>("op_55411_cast_fp16")];
            tensor<int32, [4]> var_55415_begin_0 = const()[name = tensor<string, []>("op_55415_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_55415_end_0 = const()[name = tensor<string, []>("op_55415_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_55415_end_mask_0 = const()[name = tensor<string, []>("op_55415_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_55415_cast_fp16 = slice_by_index(begin = var_55415_begin_0, end = var_55415_end_0, end_mask = var_55415_end_mask_0, x = k_491_cast_fp16)[name = tensor<string, []>("op_55415_cast_fp16")];
            tensor<int32, [4]> var_55419_begin_0 = const()[name = tensor<string, []>("op_55419_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_55419_end_0 = const()[name = tensor<string, []>("op_55419_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_55419_end_mask_0 = const()[name = tensor<string, []>("op_55419_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_55419_cast_fp16 = slice_by_index(begin = var_55419_begin_0, end = var_55419_end_0, end_mask = var_55419_end_mask_0, x = k_491_cast_fp16)[name = tensor<string, []>("op_55419_cast_fp16")];
            tensor<int32, [4]> var_55423_begin_0 = const()[name = tensor<string, []>("op_55423_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_55423_end_0 = const()[name = tensor<string, []>("op_55423_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_55423_end_mask_0 = const()[name = tensor<string, []>("op_55423_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_55423_cast_fp16 = slice_by_index(begin = var_55423_begin_0, end = var_55423_end_0, end_mask = var_55423_end_mask_0, x = k_491_cast_fp16)[name = tensor<string, []>("op_55423_cast_fp16")];
            tensor<int32, [4]> var_55427_begin_0 = const()[name = tensor<string, []>("op_55427_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_55427_end_0 = const()[name = tensor<string, []>("op_55427_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_55427_end_mask_0 = const()[name = tensor<string, []>("op_55427_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_55427_cast_fp16 = slice_by_index(begin = var_55427_begin_0, end = var_55427_end_0, end_mask = var_55427_end_mask_0, x = k_491_cast_fp16)[name = tensor<string, []>("op_55427_cast_fp16")];
            tensor<int32, [4]> var_55431_begin_0 = const()[name = tensor<string, []>("op_55431_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_55431_end_0 = const()[name = tensor<string, []>("op_55431_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_55431_end_mask_0 = const()[name = tensor<string, []>("op_55431_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_55431_cast_fp16 = slice_by_index(begin = var_55431_begin_0, end = var_55431_end_0, end_mask = var_55431_end_mask_0, x = k_491_cast_fp16)[name = tensor<string, []>("op_55431_cast_fp16")];
            tensor<int32, [4]> var_55435_begin_0 = const()[name = tensor<string, []>("op_55435_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_55435_end_0 = const()[name = tensor<string, []>("op_55435_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_55435_end_mask_0 = const()[name = tensor<string, []>("op_55435_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_55435_cast_fp16 = slice_by_index(begin = var_55435_begin_0, end = var_55435_end_0, end_mask = var_55435_end_mask_0, x = k_491_cast_fp16)[name = tensor<string, []>("op_55435_cast_fp16")];
            tensor<int32, [4]> var_55439_begin_0 = const()[name = tensor<string, []>("op_55439_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_55439_end_0 = const()[name = tensor<string, []>("op_55439_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_55439_end_mask_0 = const()[name = tensor<string, []>("op_55439_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_55439_cast_fp16 = slice_by_index(begin = var_55439_begin_0, end = var_55439_end_0, end_mask = var_55439_end_mask_0, x = k_491_cast_fp16)[name = tensor<string, []>("op_55439_cast_fp16")];
            tensor<int32, [4]> var_55443_begin_0 = const()[name = tensor<string, []>("op_55443_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_55443_end_0 = const()[name = tensor<string, []>("op_55443_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_55443_end_mask_0 = const()[name = tensor<string, []>("op_55443_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_55443_cast_fp16 = slice_by_index(begin = var_55443_begin_0, end = var_55443_end_0, end_mask = var_55443_end_mask_0, x = k_491_cast_fp16)[name = tensor<string, []>("op_55443_cast_fp16")];
            tensor<int32, [4]> var_55447_begin_0 = const()[name = tensor<string, []>("op_55447_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_55447_end_0 = const()[name = tensor<string, []>("op_55447_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_55447_end_mask_0 = const()[name = tensor<string, []>("op_55447_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_55447_cast_fp16 = slice_by_index(begin = var_55447_begin_0, end = var_55447_end_0, end_mask = var_55447_end_mask_0, x = k_491_cast_fp16)[name = tensor<string, []>("op_55447_cast_fp16")];
            tensor<int32, [4]> var_55451_begin_0 = const()[name = tensor<string, []>("op_55451_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_55451_end_0 = const()[name = tensor<string, []>("op_55451_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_55451_end_mask_0 = const()[name = tensor<string, []>("op_55451_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_55451_cast_fp16 = slice_by_index(begin = var_55451_begin_0, end = var_55451_end_0, end_mask = var_55451_end_mask_0, x = k_491_cast_fp16)[name = tensor<string, []>("op_55451_cast_fp16")];
            tensor<int32, [4]> var_55455_begin_0 = const()[name = tensor<string, []>("op_55455_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_55455_end_0 = const()[name = tensor<string, []>("op_55455_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_55455_end_mask_0 = const()[name = tensor<string, []>("op_55455_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_55455_cast_fp16 = slice_by_index(begin = var_55455_begin_0, end = var_55455_end_0, end_mask = var_55455_end_mask_0, x = k_491_cast_fp16)[name = tensor<string, []>("op_55455_cast_fp16")];
            tensor<int32, [4]> var_55459_begin_0 = const()[name = tensor<string, []>("op_55459_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_55459_end_0 = const()[name = tensor<string, []>("op_55459_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_55459_end_mask_0 = const()[name = tensor<string, []>("op_55459_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_55459_cast_fp16 = slice_by_index(begin = var_55459_begin_0, end = var_55459_end_0, end_mask = var_55459_end_mask_0, x = k_491_cast_fp16)[name = tensor<string, []>("op_55459_cast_fp16")];
            tensor<int32, [4]> var_55463_begin_0 = const()[name = tensor<string, []>("op_55463_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_55463_end_0 = const()[name = tensor<string, []>("op_55463_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_55463_end_mask_0 = const()[name = tensor<string, []>("op_55463_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_55463_cast_fp16 = slice_by_index(begin = var_55463_begin_0, end = var_55463_end_0, end_mask = var_55463_end_mask_0, x = k_491_cast_fp16)[name = tensor<string, []>("op_55463_cast_fp16")];
            tensor<int32, [4]> var_55465_begin_0 = const()[name = tensor<string, []>("op_55465_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_55465_end_0 = const()[name = tensor<string, []>("op_55465_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_55465_end_mask_0 = const()[name = tensor<string, []>("op_55465_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55465_cast_fp16 = slice_by_index(begin = var_55465_begin_0, end = var_55465_end_0, end_mask = var_55465_end_mask_0, x = v_245_cast_fp16)[name = tensor<string, []>("op_55465_cast_fp16")];
            tensor<int32, [4]> var_55469_begin_0 = const()[name = tensor<string, []>("op_55469_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_55469_end_0 = const()[name = tensor<string, []>("op_55469_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_55469_end_mask_0 = const()[name = tensor<string, []>("op_55469_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55469_cast_fp16 = slice_by_index(begin = var_55469_begin_0, end = var_55469_end_0, end_mask = var_55469_end_mask_0, x = v_245_cast_fp16)[name = tensor<string, []>("op_55469_cast_fp16")];
            tensor<int32, [4]> var_55473_begin_0 = const()[name = tensor<string, []>("op_55473_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_55473_end_0 = const()[name = tensor<string, []>("op_55473_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_55473_end_mask_0 = const()[name = tensor<string, []>("op_55473_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55473_cast_fp16 = slice_by_index(begin = var_55473_begin_0, end = var_55473_end_0, end_mask = var_55473_end_mask_0, x = v_245_cast_fp16)[name = tensor<string, []>("op_55473_cast_fp16")];
            tensor<int32, [4]> var_55477_begin_0 = const()[name = tensor<string, []>("op_55477_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_55477_end_0 = const()[name = tensor<string, []>("op_55477_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_55477_end_mask_0 = const()[name = tensor<string, []>("op_55477_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55477_cast_fp16 = slice_by_index(begin = var_55477_begin_0, end = var_55477_end_0, end_mask = var_55477_end_mask_0, x = v_245_cast_fp16)[name = tensor<string, []>("op_55477_cast_fp16")];
            tensor<int32, [4]> var_55481_begin_0 = const()[name = tensor<string, []>("op_55481_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_55481_end_0 = const()[name = tensor<string, []>("op_55481_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_55481_end_mask_0 = const()[name = tensor<string, []>("op_55481_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55481_cast_fp16 = slice_by_index(begin = var_55481_begin_0, end = var_55481_end_0, end_mask = var_55481_end_mask_0, x = v_245_cast_fp16)[name = tensor<string, []>("op_55481_cast_fp16")];
            tensor<int32, [4]> var_55485_begin_0 = const()[name = tensor<string, []>("op_55485_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_55485_end_0 = const()[name = tensor<string, []>("op_55485_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_55485_end_mask_0 = const()[name = tensor<string, []>("op_55485_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55485_cast_fp16 = slice_by_index(begin = var_55485_begin_0, end = var_55485_end_0, end_mask = var_55485_end_mask_0, x = v_245_cast_fp16)[name = tensor<string, []>("op_55485_cast_fp16")];
            tensor<int32, [4]> var_55489_begin_0 = const()[name = tensor<string, []>("op_55489_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_55489_end_0 = const()[name = tensor<string, []>("op_55489_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_55489_end_mask_0 = const()[name = tensor<string, []>("op_55489_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55489_cast_fp16 = slice_by_index(begin = var_55489_begin_0, end = var_55489_end_0, end_mask = var_55489_end_mask_0, x = v_245_cast_fp16)[name = tensor<string, []>("op_55489_cast_fp16")];
            tensor<int32, [4]> var_55493_begin_0 = const()[name = tensor<string, []>("op_55493_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_55493_end_0 = const()[name = tensor<string, []>("op_55493_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_55493_end_mask_0 = const()[name = tensor<string, []>("op_55493_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55493_cast_fp16 = slice_by_index(begin = var_55493_begin_0, end = var_55493_end_0, end_mask = var_55493_end_mask_0, x = v_245_cast_fp16)[name = tensor<string, []>("op_55493_cast_fp16")];
            tensor<int32, [4]> var_55497_begin_0 = const()[name = tensor<string, []>("op_55497_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_55497_end_0 = const()[name = tensor<string, []>("op_55497_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_55497_end_mask_0 = const()[name = tensor<string, []>("op_55497_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55497_cast_fp16 = slice_by_index(begin = var_55497_begin_0, end = var_55497_end_0, end_mask = var_55497_end_mask_0, x = v_245_cast_fp16)[name = tensor<string, []>("op_55497_cast_fp16")];
            tensor<int32, [4]> var_55501_begin_0 = const()[name = tensor<string, []>("op_55501_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_55501_end_0 = const()[name = tensor<string, []>("op_55501_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_55501_end_mask_0 = const()[name = tensor<string, []>("op_55501_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55501_cast_fp16 = slice_by_index(begin = var_55501_begin_0, end = var_55501_end_0, end_mask = var_55501_end_mask_0, x = v_245_cast_fp16)[name = tensor<string, []>("op_55501_cast_fp16")];
            tensor<int32, [4]> var_55505_begin_0 = const()[name = tensor<string, []>("op_55505_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_55505_end_0 = const()[name = tensor<string, []>("op_55505_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_55505_end_mask_0 = const()[name = tensor<string, []>("op_55505_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55505_cast_fp16 = slice_by_index(begin = var_55505_begin_0, end = var_55505_end_0, end_mask = var_55505_end_mask_0, x = v_245_cast_fp16)[name = tensor<string, []>("op_55505_cast_fp16")];
            tensor<int32, [4]> var_55509_begin_0 = const()[name = tensor<string, []>("op_55509_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_55509_end_0 = const()[name = tensor<string, []>("op_55509_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_55509_end_mask_0 = const()[name = tensor<string, []>("op_55509_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55509_cast_fp16 = slice_by_index(begin = var_55509_begin_0, end = var_55509_end_0, end_mask = var_55509_end_mask_0, x = v_245_cast_fp16)[name = tensor<string, []>("op_55509_cast_fp16")];
            tensor<int32, [4]> var_55513_begin_0 = const()[name = tensor<string, []>("op_55513_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_55513_end_0 = const()[name = tensor<string, []>("op_55513_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_55513_end_mask_0 = const()[name = tensor<string, []>("op_55513_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55513_cast_fp16 = slice_by_index(begin = var_55513_begin_0, end = var_55513_end_0, end_mask = var_55513_end_mask_0, x = v_245_cast_fp16)[name = tensor<string, []>("op_55513_cast_fp16")];
            tensor<int32, [4]> var_55517_begin_0 = const()[name = tensor<string, []>("op_55517_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_55517_end_0 = const()[name = tensor<string, []>("op_55517_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_55517_end_mask_0 = const()[name = tensor<string, []>("op_55517_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55517_cast_fp16 = slice_by_index(begin = var_55517_begin_0, end = var_55517_end_0, end_mask = var_55517_end_mask_0, x = v_245_cast_fp16)[name = tensor<string, []>("op_55517_cast_fp16")];
            tensor<int32, [4]> var_55521_begin_0 = const()[name = tensor<string, []>("op_55521_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_55521_end_0 = const()[name = tensor<string, []>("op_55521_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_55521_end_mask_0 = const()[name = tensor<string, []>("op_55521_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55521_cast_fp16 = slice_by_index(begin = var_55521_begin_0, end = var_55521_end_0, end_mask = var_55521_end_mask_0, x = v_245_cast_fp16)[name = tensor<string, []>("op_55521_cast_fp16")];
            tensor<int32, [4]> var_55525_begin_0 = const()[name = tensor<string, []>("op_55525_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_55525_end_0 = const()[name = tensor<string, []>("op_55525_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_55525_end_mask_0 = const()[name = tensor<string, []>("op_55525_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55525_cast_fp16 = slice_by_index(begin = var_55525_begin_0, end = var_55525_end_0, end_mask = var_55525_end_mask_0, x = v_245_cast_fp16)[name = tensor<string, []>("op_55525_cast_fp16")];
            tensor<int32, [4]> var_55529_begin_0 = const()[name = tensor<string, []>("op_55529_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_55529_end_0 = const()[name = tensor<string, []>("op_55529_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_55529_end_mask_0 = const()[name = tensor<string, []>("op_55529_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55529_cast_fp16 = slice_by_index(begin = var_55529_begin_0, end = var_55529_end_0, end_mask = var_55529_end_mask_0, x = v_245_cast_fp16)[name = tensor<string, []>("op_55529_cast_fp16")];
            tensor<int32, [4]> var_55533_begin_0 = const()[name = tensor<string, []>("op_55533_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_55533_end_0 = const()[name = tensor<string, []>("op_55533_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_55533_end_mask_0 = const()[name = tensor<string, []>("op_55533_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55533_cast_fp16 = slice_by_index(begin = var_55533_begin_0, end = var_55533_end_0, end_mask = var_55533_end_mask_0, x = v_245_cast_fp16)[name = tensor<string, []>("op_55533_cast_fp16")];
            tensor<int32, [4]> var_55537_begin_0 = const()[name = tensor<string, []>("op_55537_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_55537_end_0 = const()[name = tensor<string, []>("op_55537_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_55537_end_mask_0 = const()[name = tensor<string, []>("op_55537_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55537_cast_fp16 = slice_by_index(begin = var_55537_begin_0, end = var_55537_end_0, end_mask = var_55537_end_mask_0, x = v_245_cast_fp16)[name = tensor<string, []>("op_55537_cast_fp16")];
            tensor<int32, [4]> var_55541_begin_0 = const()[name = tensor<string, []>("op_55541_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_55541_end_0 = const()[name = tensor<string, []>("op_55541_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_55541_end_mask_0 = const()[name = tensor<string, []>("op_55541_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55541_cast_fp16 = slice_by_index(begin = var_55541_begin_0, end = var_55541_end_0, end_mask = var_55541_end_mask_0, x = v_245_cast_fp16)[name = tensor<string, []>("op_55541_cast_fp16")];
            tensor<string, []> var_55545_equation_0 = const()[name = tensor<string, []>("op_55545_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55545_cast_fp16 = einsum(equation = var_55545_equation_0, values = (var_55387_cast_fp16, var_55304_cast_fp16))[name = tensor<string, []>("op_55545_cast_fp16")];
            tensor<fp16, []> var_55546_to_fp16 = const()[name = tensor<string, []>("op_55546_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4721_cast_fp16 = mul(x = var_55545_cast_fp16, y = var_55546_to_fp16)[name = tensor<string, []>("aw_4721_cast_fp16")];
            tensor<string, []> var_55549_equation_0 = const()[name = tensor<string, []>("op_55549_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55549_cast_fp16 = einsum(equation = var_55549_equation_0, values = (var_55391_cast_fp16, var_55308_cast_fp16))[name = tensor<string, []>("op_55549_cast_fp16")];
            tensor<fp16, []> var_55550_to_fp16 = const()[name = tensor<string, []>("op_55550_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4723_cast_fp16 = mul(x = var_55549_cast_fp16, y = var_55550_to_fp16)[name = tensor<string, []>("aw_4723_cast_fp16")];
            tensor<string, []> var_55553_equation_0 = const()[name = tensor<string, []>("op_55553_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55553_cast_fp16 = einsum(equation = var_55553_equation_0, values = (var_55395_cast_fp16, var_55312_cast_fp16))[name = tensor<string, []>("op_55553_cast_fp16")];
            tensor<fp16, []> var_55554_to_fp16 = const()[name = tensor<string, []>("op_55554_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4725_cast_fp16 = mul(x = var_55553_cast_fp16, y = var_55554_to_fp16)[name = tensor<string, []>("aw_4725_cast_fp16")];
            tensor<string, []> var_55557_equation_0 = const()[name = tensor<string, []>("op_55557_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55557_cast_fp16 = einsum(equation = var_55557_equation_0, values = (var_55399_cast_fp16, var_55316_cast_fp16))[name = tensor<string, []>("op_55557_cast_fp16")];
            tensor<fp16, []> var_55558_to_fp16 = const()[name = tensor<string, []>("op_55558_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4727_cast_fp16 = mul(x = var_55557_cast_fp16, y = var_55558_to_fp16)[name = tensor<string, []>("aw_4727_cast_fp16")];
            tensor<string, []> var_55561_equation_0 = const()[name = tensor<string, []>("op_55561_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55561_cast_fp16 = einsum(equation = var_55561_equation_0, values = (var_55403_cast_fp16, var_55320_cast_fp16))[name = tensor<string, []>("op_55561_cast_fp16")];
            tensor<fp16, []> var_55562_to_fp16 = const()[name = tensor<string, []>("op_55562_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4729_cast_fp16 = mul(x = var_55561_cast_fp16, y = var_55562_to_fp16)[name = tensor<string, []>("aw_4729_cast_fp16")];
            tensor<string, []> var_55565_equation_0 = const()[name = tensor<string, []>("op_55565_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55565_cast_fp16 = einsum(equation = var_55565_equation_0, values = (var_55407_cast_fp16, var_55324_cast_fp16))[name = tensor<string, []>("op_55565_cast_fp16")];
            tensor<fp16, []> var_55566_to_fp16 = const()[name = tensor<string, []>("op_55566_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4731_cast_fp16 = mul(x = var_55565_cast_fp16, y = var_55566_to_fp16)[name = tensor<string, []>("aw_4731_cast_fp16")];
            tensor<string, []> var_55569_equation_0 = const()[name = tensor<string, []>("op_55569_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55569_cast_fp16 = einsum(equation = var_55569_equation_0, values = (var_55411_cast_fp16, var_55328_cast_fp16))[name = tensor<string, []>("op_55569_cast_fp16")];
            tensor<fp16, []> var_55570_to_fp16 = const()[name = tensor<string, []>("op_55570_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4733_cast_fp16 = mul(x = var_55569_cast_fp16, y = var_55570_to_fp16)[name = tensor<string, []>("aw_4733_cast_fp16")];
            tensor<string, []> var_55573_equation_0 = const()[name = tensor<string, []>("op_55573_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55573_cast_fp16 = einsum(equation = var_55573_equation_0, values = (var_55415_cast_fp16, var_55332_cast_fp16))[name = tensor<string, []>("op_55573_cast_fp16")];
            tensor<fp16, []> var_55574_to_fp16 = const()[name = tensor<string, []>("op_55574_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4735_cast_fp16 = mul(x = var_55573_cast_fp16, y = var_55574_to_fp16)[name = tensor<string, []>("aw_4735_cast_fp16")];
            tensor<string, []> var_55577_equation_0 = const()[name = tensor<string, []>("op_55577_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55577_cast_fp16 = einsum(equation = var_55577_equation_0, values = (var_55419_cast_fp16, var_55336_cast_fp16))[name = tensor<string, []>("op_55577_cast_fp16")];
            tensor<fp16, []> var_55578_to_fp16 = const()[name = tensor<string, []>("op_55578_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4737_cast_fp16 = mul(x = var_55577_cast_fp16, y = var_55578_to_fp16)[name = tensor<string, []>("aw_4737_cast_fp16")];
            tensor<string, []> var_55581_equation_0 = const()[name = tensor<string, []>("op_55581_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55581_cast_fp16 = einsum(equation = var_55581_equation_0, values = (var_55423_cast_fp16, var_55340_cast_fp16))[name = tensor<string, []>("op_55581_cast_fp16")];
            tensor<fp16, []> var_55582_to_fp16 = const()[name = tensor<string, []>("op_55582_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4739_cast_fp16 = mul(x = var_55581_cast_fp16, y = var_55582_to_fp16)[name = tensor<string, []>("aw_4739_cast_fp16")];
            tensor<string, []> var_55585_equation_0 = const()[name = tensor<string, []>("op_55585_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55585_cast_fp16 = einsum(equation = var_55585_equation_0, values = (var_55427_cast_fp16, var_55344_cast_fp16))[name = tensor<string, []>("op_55585_cast_fp16")];
            tensor<fp16, []> var_55586_to_fp16 = const()[name = tensor<string, []>("op_55586_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4741_cast_fp16 = mul(x = var_55585_cast_fp16, y = var_55586_to_fp16)[name = tensor<string, []>("aw_4741_cast_fp16")];
            tensor<string, []> var_55589_equation_0 = const()[name = tensor<string, []>("op_55589_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55589_cast_fp16 = einsum(equation = var_55589_equation_0, values = (var_55431_cast_fp16, var_55348_cast_fp16))[name = tensor<string, []>("op_55589_cast_fp16")];
            tensor<fp16, []> var_55590_to_fp16 = const()[name = tensor<string, []>("op_55590_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4743_cast_fp16 = mul(x = var_55589_cast_fp16, y = var_55590_to_fp16)[name = tensor<string, []>("aw_4743_cast_fp16")];
            tensor<string, []> var_55593_equation_0 = const()[name = tensor<string, []>("op_55593_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55593_cast_fp16 = einsum(equation = var_55593_equation_0, values = (var_55435_cast_fp16, var_55352_cast_fp16))[name = tensor<string, []>("op_55593_cast_fp16")];
            tensor<fp16, []> var_55594_to_fp16 = const()[name = tensor<string, []>("op_55594_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4745_cast_fp16 = mul(x = var_55593_cast_fp16, y = var_55594_to_fp16)[name = tensor<string, []>("aw_4745_cast_fp16")];
            tensor<string, []> var_55597_equation_0 = const()[name = tensor<string, []>("op_55597_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55597_cast_fp16 = einsum(equation = var_55597_equation_0, values = (var_55439_cast_fp16, var_55356_cast_fp16))[name = tensor<string, []>("op_55597_cast_fp16")];
            tensor<fp16, []> var_55598_to_fp16 = const()[name = tensor<string, []>("op_55598_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4747_cast_fp16 = mul(x = var_55597_cast_fp16, y = var_55598_to_fp16)[name = tensor<string, []>("aw_4747_cast_fp16")];
            tensor<string, []> var_55601_equation_0 = const()[name = tensor<string, []>("op_55601_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55601_cast_fp16 = einsum(equation = var_55601_equation_0, values = (var_55443_cast_fp16, var_55360_cast_fp16))[name = tensor<string, []>("op_55601_cast_fp16")];
            tensor<fp16, []> var_55602_to_fp16 = const()[name = tensor<string, []>("op_55602_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4749_cast_fp16 = mul(x = var_55601_cast_fp16, y = var_55602_to_fp16)[name = tensor<string, []>("aw_4749_cast_fp16")];
            tensor<string, []> var_55605_equation_0 = const()[name = tensor<string, []>("op_55605_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55605_cast_fp16 = einsum(equation = var_55605_equation_0, values = (var_55447_cast_fp16, var_55364_cast_fp16))[name = tensor<string, []>("op_55605_cast_fp16")];
            tensor<fp16, []> var_55606_to_fp16 = const()[name = tensor<string, []>("op_55606_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4751_cast_fp16 = mul(x = var_55605_cast_fp16, y = var_55606_to_fp16)[name = tensor<string, []>("aw_4751_cast_fp16")];
            tensor<string, []> var_55609_equation_0 = const()[name = tensor<string, []>("op_55609_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55609_cast_fp16 = einsum(equation = var_55609_equation_0, values = (var_55451_cast_fp16, var_55368_cast_fp16))[name = tensor<string, []>("op_55609_cast_fp16")];
            tensor<fp16, []> var_55610_to_fp16 = const()[name = tensor<string, []>("op_55610_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4753_cast_fp16 = mul(x = var_55609_cast_fp16, y = var_55610_to_fp16)[name = tensor<string, []>("aw_4753_cast_fp16")];
            tensor<string, []> var_55613_equation_0 = const()[name = tensor<string, []>("op_55613_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55613_cast_fp16 = einsum(equation = var_55613_equation_0, values = (var_55455_cast_fp16, var_55372_cast_fp16))[name = tensor<string, []>("op_55613_cast_fp16")];
            tensor<fp16, []> var_55614_to_fp16 = const()[name = tensor<string, []>("op_55614_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4755_cast_fp16 = mul(x = var_55613_cast_fp16, y = var_55614_to_fp16)[name = tensor<string, []>("aw_4755_cast_fp16")];
            tensor<string, []> var_55617_equation_0 = const()[name = tensor<string, []>("op_55617_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55617_cast_fp16 = einsum(equation = var_55617_equation_0, values = (var_55459_cast_fp16, var_55376_cast_fp16))[name = tensor<string, []>("op_55617_cast_fp16")];
            tensor<fp16, []> var_55618_to_fp16 = const()[name = tensor<string, []>("op_55618_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4757_cast_fp16 = mul(x = var_55617_cast_fp16, y = var_55618_to_fp16)[name = tensor<string, []>("aw_4757_cast_fp16")];
            tensor<string, []> var_55621_equation_0 = const()[name = tensor<string, []>("op_55621_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55621_cast_fp16 = einsum(equation = var_55621_equation_0, values = (var_55463_cast_fp16, var_55380_cast_fp16))[name = tensor<string, []>("op_55621_cast_fp16")];
            tensor<fp16, []> var_55622_to_fp16 = const()[name = tensor<string, []>("op_55622_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4759_cast_fp16 = mul(x = var_55621_cast_fp16, y = var_55622_to_fp16)[name = tensor<string, []>("aw_4759_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55624_cast_fp16 = softmax(axis = var_30355, x = aw_4721_cast_fp16)[name = tensor<string, []>("op_55624_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55625_cast_fp16 = softmax(axis = var_30355, x = aw_4723_cast_fp16)[name = tensor<string, []>("op_55625_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55626_cast_fp16 = softmax(axis = var_30355, x = aw_4725_cast_fp16)[name = tensor<string, []>("op_55626_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55627_cast_fp16 = softmax(axis = var_30355, x = aw_4727_cast_fp16)[name = tensor<string, []>("op_55627_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55628_cast_fp16 = softmax(axis = var_30355, x = aw_4729_cast_fp16)[name = tensor<string, []>("op_55628_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55629_cast_fp16 = softmax(axis = var_30355, x = aw_4731_cast_fp16)[name = tensor<string, []>("op_55629_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55630_cast_fp16 = softmax(axis = var_30355, x = aw_4733_cast_fp16)[name = tensor<string, []>("op_55630_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55631_cast_fp16 = softmax(axis = var_30355, x = aw_4735_cast_fp16)[name = tensor<string, []>("op_55631_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55632_cast_fp16 = softmax(axis = var_30355, x = aw_4737_cast_fp16)[name = tensor<string, []>("op_55632_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55633_cast_fp16 = softmax(axis = var_30355, x = aw_4739_cast_fp16)[name = tensor<string, []>("op_55633_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55634_cast_fp16 = softmax(axis = var_30355, x = aw_4741_cast_fp16)[name = tensor<string, []>("op_55634_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55635_cast_fp16 = softmax(axis = var_30355, x = aw_4743_cast_fp16)[name = tensor<string, []>("op_55635_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55636_cast_fp16 = softmax(axis = var_30355, x = aw_4745_cast_fp16)[name = tensor<string, []>("op_55636_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55637_cast_fp16 = softmax(axis = var_30355, x = aw_4747_cast_fp16)[name = tensor<string, []>("op_55637_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55638_cast_fp16 = softmax(axis = var_30355, x = aw_4749_cast_fp16)[name = tensor<string, []>("op_55638_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55639_cast_fp16 = softmax(axis = var_30355, x = aw_4751_cast_fp16)[name = tensor<string, []>("op_55639_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55640_cast_fp16 = softmax(axis = var_30355, x = aw_4753_cast_fp16)[name = tensor<string, []>("op_55640_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55641_cast_fp16 = softmax(axis = var_30355, x = aw_4755_cast_fp16)[name = tensor<string, []>("op_55641_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55642_cast_fp16 = softmax(axis = var_30355, x = aw_4757_cast_fp16)[name = tensor<string, []>("op_55642_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55643_cast_fp16 = softmax(axis = var_30355, x = aw_4759_cast_fp16)[name = tensor<string, []>("op_55643_cast_fp16")];
            tensor<string, []> var_55645_equation_0 = const()[name = tensor<string, []>("op_55645_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55645_cast_fp16 = einsum(equation = var_55645_equation_0, values = (var_55465_cast_fp16, var_55624_cast_fp16))[name = tensor<string, []>("op_55645_cast_fp16")];
            tensor<string, []> var_55647_equation_0 = const()[name = tensor<string, []>("op_55647_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55647_cast_fp16 = einsum(equation = var_55647_equation_0, values = (var_55469_cast_fp16, var_55625_cast_fp16))[name = tensor<string, []>("op_55647_cast_fp16")];
            tensor<string, []> var_55649_equation_0 = const()[name = tensor<string, []>("op_55649_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55649_cast_fp16 = einsum(equation = var_55649_equation_0, values = (var_55473_cast_fp16, var_55626_cast_fp16))[name = tensor<string, []>("op_55649_cast_fp16")];
            tensor<string, []> var_55651_equation_0 = const()[name = tensor<string, []>("op_55651_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55651_cast_fp16 = einsum(equation = var_55651_equation_0, values = (var_55477_cast_fp16, var_55627_cast_fp16))[name = tensor<string, []>("op_55651_cast_fp16")];
            tensor<string, []> var_55653_equation_0 = const()[name = tensor<string, []>("op_55653_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55653_cast_fp16 = einsum(equation = var_55653_equation_0, values = (var_55481_cast_fp16, var_55628_cast_fp16))[name = tensor<string, []>("op_55653_cast_fp16")];
            tensor<string, []> var_55655_equation_0 = const()[name = tensor<string, []>("op_55655_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55655_cast_fp16 = einsum(equation = var_55655_equation_0, values = (var_55485_cast_fp16, var_55629_cast_fp16))[name = tensor<string, []>("op_55655_cast_fp16")];
            tensor<string, []> var_55657_equation_0 = const()[name = tensor<string, []>("op_55657_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55657_cast_fp16 = einsum(equation = var_55657_equation_0, values = (var_55489_cast_fp16, var_55630_cast_fp16))[name = tensor<string, []>("op_55657_cast_fp16")];
            tensor<string, []> var_55659_equation_0 = const()[name = tensor<string, []>("op_55659_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55659_cast_fp16 = einsum(equation = var_55659_equation_0, values = (var_55493_cast_fp16, var_55631_cast_fp16))[name = tensor<string, []>("op_55659_cast_fp16")];
            tensor<string, []> var_55661_equation_0 = const()[name = tensor<string, []>("op_55661_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55661_cast_fp16 = einsum(equation = var_55661_equation_0, values = (var_55497_cast_fp16, var_55632_cast_fp16))[name = tensor<string, []>("op_55661_cast_fp16")];
            tensor<string, []> var_55663_equation_0 = const()[name = tensor<string, []>("op_55663_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55663_cast_fp16 = einsum(equation = var_55663_equation_0, values = (var_55501_cast_fp16, var_55633_cast_fp16))[name = tensor<string, []>("op_55663_cast_fp16")];
            tensor<string, []> var_55665_equation_0 = const()[name = tensor<string, []>("op_55665_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55665_cast_fp16 = einsum(equation = var_55665_equation_0, values = (var_55505_cast_fp16, var_55634_cast_fp16))[name = tensor<string, []>("op_55665_cast_fp16")];
            tensor<string, []> var_55667_equation_0 = const()[name = tensor<string, []>("op_55667_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55667_cast_fp16 = einsum(equation = var_55667_equation_0, values = (var_55509_cast_fp16, var_55635_cast_fp16))[name = tensor<string, []>("op_55667_cast_fp16")];
            tensor<string, []> var_55669_equation_0 = const()[name = tensor<string, []>("op_55669_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55669_cast_fp16 = einsum(equation = var_55669_equation_0, values = (var_55513_cast_fp16, var_55636_cast_fp16))[name = tensor<string, []>("op_55669_cast_fp16")];
            tensor<string, []> var_55671_equation_0 = const()[name = tensor<string, []>("op_55671_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55671_cast_fp16 = einsum(equation = var_55671_equation_0, values = (var_55517_cast_fp16, var_55637_cast_fp16))[name = tensor<string, []>("op_55671_cast_fp16")];
            tensor<string, []> var_55673_equation_0 = const()[name = tensor<string, []>("op_55673_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55673_cast_fp16 = einsum(equation = var_55673_equation_0, values = (var_55521_cast_fp16, var_55638_cast_fp16))[name = tensor<string, []>("op_55673_cast_fp16")];
            tensor<string, []> var_55675_equation_0 = const()[name = tensor<string, []>("op_55675_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55675_cast_fp16 = einsum(equation = var_55675_equation_0, values = (var_55525_cast_fp16, var_55639_cast_fp16))[name = tensor<string, []>("op_55675_cast_fp16")];
            tensor<string, []> var_55677_equation_0 = const()[name = tensor<string, []>("op_55677_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55677_cast_fp16 = einsum(equation = var_55677_equation_0, values = (var_55529_cast_fp16, var_55640_cast_fp16))[name = tensor<string, []>("op_55677_cast_fp16")];
            tensor<string, []> var_55679_equation_0 = const()[name = tensor<string, []>("op_55679_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55679_cast_fp16 = einsum(equation = var_55679_equation_0, values = (var_55533_cast_fp16, var_55641_cast_fp16))[name = tensor<string, []>("op_55679_cast_fp16")];
            tensor<string, []> var_55681_equation_0 = const()[name = tensor<string, []>("op_55681_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55681_cast_fp16 = einsum(equation = var_55681_equation_0, values = (var_55537_cast_fp16, var_55642_cast_fp16))[name = tensor<string, []>("op_55681_cast_fp16")];
            tensor<string, []> var_55683_equation_0 = const()[name = tensor<string, []>("op_55683_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55683_cast_fp16 = einsum(equation = var_55683_equation_0, values = (var_55541_cast_fp16, var_55643_cast_fp16))[name = tensor<string, []>("op_55683_cast_fp16")];
            tensor<bool, []> input_691_interleave_0 = const()[name = tensor<string, []>("input_691_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_691_cast_fp16 = concat(axis = var_30355, interleave = input_691_interleave_0, values = (var_55645_cast_fp16, var_55647_cast_fp16, var_55649_cast_fp16, var_55651_cast_fp16, var_55653_cast_fp16, var_55655_cast_fp16, var_55657_cast_fp16, var_55659_cast_fp16, var_55661_cast_fp16, var_55663_cast_fp16, var_55665_cast_fp16, var_55667_cast_fp16, var_55669_cast_fp16, var_55671_cast_fp16, var_55673_cast_fp16, var_55675_cast_fp16, var_55677_cast_fp16, var_55679_cast_fp16, var_55681_cast_fp16, var_55683_cast_fp16))[name = tensor<string, []>("input_691_cast_fp16")];
            tensor<string, []> var_55693_pad_type_0 = const()[name = tensor<string, []>("op_55693_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_55693_strides_0 = const()[name = tensor<string, []>("op_55693_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_55693_pad_0 = const()[name = tensor<string, []>("op_55693_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_55693_dilations_0 = const()[name = tensor<string, []>("op_55693_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_55693_groups_0 = const()[name = tensor<string, []>("op_55693_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_7_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(784616896))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(785845760))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_7_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_7_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_7_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(785845952)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_55693_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_7_attn1_to_out_0_bias_to_fp16, dilations = var_55693_dilations_0, groups = var_55693_groups_0, pad = var_55693_pad_0, pad_type = var_55693_pad_type_0, strides = var_55693_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_7_attn1_to_out_0_weight_to_fp16_palettized, x = input_691_cast_fp16)[name = tensor<string, []>("op_55693_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_369_cast_fp16 = add(x = var_55693_cast_fp16, y = inputs_367_cast_fp16)[name = tensor<string, []>("inputs_369_cast_fp16")];
            tensor<int32, [1]> hidden_states_479_axes_0 = const()[name = tensor<string, []>("hidden_states_479_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_479_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_479_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(785848576)))];
            tensor<fp16, [1280]> hidden_states_479_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_479_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(785851200)))];
            tensor<fp16, []> var_55703_to_fp16 = const()[name = tensor<string, []>("op_55703_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_479_cast_fp16 = layer_norm(axes = hidden_states_479_axes_0, beta = hidden_states_479_beta_0_to_fp16, epsilon = var_55703_to_fp16, gamma = hidden_states_479_gamma_0_to_fp16, x = inputs_369_cast_fp16)[name = tensor<string, []>("hidden_states_479_cast_fp16")];
            tensor<string, []> q_247_pad_type_0 = const()[name = tensor<string, []>("q_247_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_247_strides_0 = const()[name = tensor<string, []>("q_247_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_247_pad_0 = const()[name = tensor<string, []>("q_247_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_247_dilations_0 = const()[name = tensor<string, []>("q_247_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_247_groups_0 = const()[name = tensor<string, []>("q_247_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_7_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(785853824))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(787082688))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_7_attn2_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_247_cast_fp16 = conv(dilations = q_247_dilations_0, groups = q_247_groups_0, pad = q_247_pad_0, pad_type = q_247_pad_type_0, strides = q_247_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_7_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_479_cast_fp16)[name = tensor<string, []>("q_247_cast_fp16")];
            tensor<string, []> k_493_pad_type_0 = const()[name = tensor<string, []>("k_493_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_493_strides_0 = const()[name = tensor<string, []>("k_493_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_493_pad_0 = const()[name = tensor<string, []>("k_493_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_493_dilations_0 = const()[name = tensor<string, []>("k_493_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_493_groups_0 = const()[name = tensor<string, []>("k_493_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_7_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(787082880))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(789049024))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_7_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> k_493_cast_fp16 = conv(dilations = k_493_dilations_0, groups = k_493_groups_0, pad = k_493_pad_0, pad_type = k_493_pad_type_0, strides = k_493_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_7_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("k_493_cast_fp16")];
            tensor<string, []> v_247_pad_type_0 = const()[name = tensor<string, []>("v_247_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_247_strides_0 = const()[name = tensor<string, []>("v_247_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_247_pad_0 = const()[name = tensor<string, []>("v_247_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_247_dilations_0 = const()[name = tensor<string, []>("v_247_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_247_groups_0 = const()[name = tensor<string, []>("v_247_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_7_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(789049216))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(791015360))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_7_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> v_247_cast_fp16 = conv(dilations = v_247_dilations_0, groups = v_247_groups_0, pad = v_247_pad_0, pad_type = v_247_pad_type_0, strides = v_247_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_7_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("v_247_cast_fp16")];
            tensor<int32, [4]> var_55736_begin_0 = const()[name = tensor<string, []>("op_55736_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_55736_end_0 = const()[name = tensor<string, []>("op_55736_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_55736_end_mask_0 = const()[name = tensor<string, []>("op_55736_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55736_cast_fp16 = slice_by_index(begin = var_55736_begin_0, end = var_55736_end_0, end_mask = var_55736_end_mask_0, x = q_247_cast_fp16)[name = tensor<string, []>("op_55736_cast_fp16")];
            tensor<int32, [4]> var_55740_begin_0 = const()[name = tensor<string, []>("op_55740_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_55740_end_0 = const()[name = tensor<string, []>("op_55740_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_55740_end_mask_0 = const()[name = tensor<string, []>("op_55740_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55740_cast_fp16 = slice_by_index(begin = var_55740_begin_0, end = var_55740_end_0, end_mask = var_55740_end_mask_0, x = q_247_cast_fp16)[name = tensor<string, []>("op_55740_cast_fp16")];
            tensor<int32, [4]> var_55744_begin_0 = const()[name = tensor<string, []>("op_55744_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_55744_end_0 = const()[name = tensor<string, []>("op_55744_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_55744_end_mask_0 = const()[name = tensor<string, []>("op_55744_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55744_cast_fp16 = slice_by_index(begin = var_55744_begin_0, end = var_55744_end_0, end_mask = var_55744_end_mask_0, x = q_247_cast_fp16)[name = tensor<string, []>("op_55744_cast_fp16")];
            tensor<int32, [4]> var_55748_begin_0 = const()[name = tensor<string, []>("op_55748_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_55748_end_0 = const()[name = tensor<string, []>("op_55748_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_55748_end_mask_0 = const()[name = tensor<string, []>("op_55748_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55748_cast_fp16 = slice_by_index(begin = var_55748_begin_0, end = var_55748_end_0, end_mask = var_55748_end_mask_0, x = q_247_cast_fp16)[name = tensor<string, []>("op_55748_cast_fp16")];
            tensor<int32, [4]> var_55752_begin_0 = const()[name = tensor<string, []>("op_55752_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_55752_end_0 = const()[name = tensor<string, []>("op_55752_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_55752_end_mask_0 = const()[name = tensor<string, []>("op_55752_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55752_cast_fp16 = slice_by_index(begin = var_55752_begin_0, end = var_55752_end_0, end_mask = var_55752_end_mask_0, x = q_247_cast_fp16)[name = tensor<string, []>("op_55752_cast_fp16")];
            tensor<int32, [4]> var_55756_begin_0 = const()[name = tensor<string, []>("op_55756_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_55756_end_0 = const()[name = tensor<string, []>("op_55756_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_55756_end_mask_0 = const()[name = tensor<string, []>("op_55756_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55756_cast_fp16 = slice_by_index(begin = var_55756_begin_0, end = var_55756_end_0, end_mask = var_55756_end_mask_0, x = q_247_cast_fp16)[name = tensor<string, []>("op_55756_cast_fp16")];
            tensor<int32, [4]> var_55760_begin_0 = const()[name = tensor<string, []>("op_55760_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_55760_end_0 = const()[name = tensor<string, []>("op_55760_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_55760_end_mask_0 = const()[name = tensor<string, []>("op_55760_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55760_cast_fp16 = slice_by_index(begin = var_55760_begin_0, end = var_55760_end_0, end_mask = var_55760_end_mask_0, x = q_247_cast_fp16)[name = tensor<string, []>("op_55760_cast_fp16")];
            tensor<int32, [4]> var_55764_begin_0 = const()[name = tensor<string, []>("op_55764_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_55764_end_0 = const()[name = tensor<string, []>("op_55764_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_55764_end_mask_0 = const()[name = tensor<string, []>("op_55764_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55764_cast_fp16 = slice_by_index(begin = var_55764_begin_0, end = var_55764_end_0, end_mask = var_55764_end_mask_0, x = q_247_cast_fp16)[name = tensor<string, []>("op_55764_cast_fp16")];
            tensor<int32, [4]> var_55768_begin_0 = const()[name = tensor<string, []>("op_55768_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_55768_end_0 = const()[name = tensor<string, []>("op_55768_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_55768_end_mask_0 = const()[name = tensor<string, []>("op_55768_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55768_cast_fp16 = slice_by_index(begin = var_55768_begin_0, end = var_55768_end_0, end_mask = var_55768_end_mask_0, x = q_247_cast_fp16)[name = tensor<string, []>("op_55768_cast_fp16")];
            tensor<int32, [4]> var_55772_begin_0 = const()[name = tensor<string, []>("op_55772_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_55772_end_0 = const()[name = tensor<string, []>("op_55772_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_55772_end_mask_0 = const()[name = tensor<string, []>("op_55772_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55772_cast_fp16 = slice_by_index(begin = var_55772_begin_0, end = var_55772_end_0, end_mask = var_55772_end_mask_0, x = q_247_cast_fp16)[name = tensor<string, []>("op_55772_cast_fp16")];
            tensor<int32, [4]> var_55776_begin_0 = const()[name = tensor<string, []>("op_55776_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_55776_end_0 = const()[name = tensor<string, []>("op_55776_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_55776_end_mask_0 = const()[name = tensor<string, []>("op_55776_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55776_cast_fp16 = slice_by_index(begin = var_55776_begin_0, end = var_55776_end_0, end_mask = var_55776_end_mask_0, x = q_247_cast_fp16)[name = tensor<string, []>("op_55776_cast_fp16")];
            tensor<int32, [4]> var_55780_begin_0 = const()[name = tensor<string, []>("op_55780_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_55780_end_0 = const()[name = tensor<string, []>("op_55780_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_55780_end_mask_0 = const()[name = tensor<string, []>("op_55780_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55780_cast_fp16 = slice_by_index(begin = var_55780_begin_0, end = var_55780_end_0, end_mask = var_55780_end_mask_0, x = q_247_cast_fp16)[name = tensor<string, []>("op_55780_cast_fp16")];
            tensor<int32, [4]> var_55784_begin_0 = const()[name = tensor<string, []>("op_55784_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_55784_end_0 = const()[name = tensor<string, []>("op_55784_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_55784_end_mask_0 = const()[name = tensor<string, []>("op_55784_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55784_cast_fp16 = slice_by_index(begin = var_55784_begin_0, end = var_55784_end_0, end_mask = var_55784_end_mask_0, x = q_247_cast_fp16)[name = tensor<string, []>("op_55784_cast_fp16")];
            tensor<int32, [4]> var_55788_begin_0 = const()[name = tensor<string, []>("op_55788_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_55788_end_0 = const()[name = tensor<string, []>("op_55788_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_55788_end_mask_0 = const()[name = tensor<string, []>("op_55788_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55788_cast_fp16 = slice_by_index(begin = var_55788_begin_0, end = var_55788_end_0, end_mask = var_55788_end_mask_0, x = q_247_cast_fp16)[name = tensor<string, []>("op_55788_cast_fp16")];
            tensor<int32, [4]> var_55792_begin_0 = const()[name = tensor<string, []>("op_55792_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_55792_end_0 = const()[name = tensor<string, []>("op_55792_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_55792_end_mask_0 = const()[name = tensor<string, []>("op_55792_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55792_cast_fp16 = slice_by_index(begin = var_55792_begin_0, end = var_55792_end_0, end_mask = var_55792_end_mask_0, x = q_247_cast_fp16)[name = tensor<string, []>("op_55792_cast_fp16")];
            tensor<int32, [4]> var_55796_begin_0 = const()[name = tensor<string, []>("op_55796_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_55796_end_0 = const()[name = tensor<string, []>("op_55796_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_55796_end_mask_0 = const()[name = tensor<string, []>("op_55796_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55796_cast_fp16 = slice_by_index(begin = var_55796_begin_0, end = var_55796_end_0, end_mask = var_55796_end_mask_0, x = q_247_cast_fp16)[name = tensor<string, []>("op_55796_cast_fp16")];
            tensor<int32, [4]> var_55800_begin_0 = const()[name = tensor<string, []>("op_55800_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_55800_end_0 = const()[name = tensor<string, []>("op_55800_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_55800_end_mask_0 = const()[name = tensor<string, []>("op_55800_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55800_cast_fp16 = slice_by_index(begin = var_55800_begin_0, end = var_55800_end_0, end_mask = var_55800_end_mask_0, x = q_247_cast_fp16)[name = tensor<string, []>("op_55800_cast_fp16")];
            tensor<int32, [4]> var_55804_begin_0 = const()[name = tensor<string, []>("op_55804_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_55804_end_0 = const()[name = tensor<string, []>("op_55804_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_55804_end_mask_0 = const()[name = tensor<string, []>("op_55804_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55804_cast_fp16 = slice_by_index(begin = var_55804_begin_0, end = var_55804_end_0, end_mask = var_55804_end_mask_0, x = q_247_cast_fp16)[name = tensor<string, []>("op_55804_cast_fp16")];
            tensor<int32, [4]> var_55808_begin_0 = const()[name = tensor<string, []>("op_55808_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_55808_end_0 = const()[name = tensor<string, []>("op_55808_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_55808_end_mask_0 = const()[name = tensor<string, []>("op_55808_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55808_cast_fp16 = slice_by_index(begin = var_55808_begin_0, end = var_55808_end_0, end_mask = var_55808_end_mask_0, x = q_247_cast_fp16)[name = tensor<string, []>("op_55808_cast_fp16")];
            tensor<int32, [4]> var_55812_begin_0 = const()[name = tensor<string, []>("op_55812_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_55812_end_0 = const()[name = tensor<string, []>("op_55812_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_55812_end_mask_0 = const()[name = tensor<string, []>("op_55812_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55812_cast_fp16 = slice_by_index(begin = var_55812_begin_0, end = var_55812_end_0, end_mask = var_55812_end_mask_0, x = q_247_cast_fp16)[name = tensor<string, []>("op_55812_cast_fp16")];
            tensor<int32, [4]> k_495_perm_0 = const()[name = tensor<string, []>("k_495_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_55819_begin_0 = const()[name = tensor<string, []>("op_55819_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_55819_end_0 = const()[name = tensor<string, []>("op_55819_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_55819_end_mask_0 = const()[name = tensor<string, []>("op_55819_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> k_495_cast_fp16 = transpose(perm = k_495_perm_0, x = k_493_cast_fp16)[name = tensor<string, []>("transpose_16")];
            tensor<fp16, [2, 77, 1, 64]> var_55819_cast_fp16 = slice_by_index(begin = var_55819_begin_0, end = var_55819_end_0, end_mask = var_55819_end_mask_0, x = k_495_cast_fp16)[name = tensor<string, []>("op_55819_cast_fp16")];
            tensor<int32, [4]> var_55823_begin_0 = const()[name = tensor<string, []>("op_55823_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_55823_end_0 = const()[name = tensor<string, []>("op_55823_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_55823_end_mask_0 = const()[name = tensor<string, []>("op_55823_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_55823_cast_fp16 = slice_by_index(begin = var_55823_begin_0, end = var_55823_end_0, end_mask = var_55823_end_mask_0, x = k_495_cast_fp16)[name = tensor<string, []>("op_55823_cast_fp16")];
            tensor<int32, [4]> var_55827_begin_0 = const()[name = tensor<string, []>("op_55827_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_55827_end_0 = const()[name = tensor<string, []>("op_55827_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_55827_end_mask_0 = const()[name = tensor<string, []>("op_55827_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_55827_cast_fp16 = slice_by_index(begin = var_55827_begin_0, end = var_55827_end_0, end_mask = var_55827_end_mask_0, x = k_495_cast_fp16)[name = tensor<string, []>("op_55827_cast_fp16")];
            tensor<int32, [4]> var_55831_begin_0 = const()[name = tensor<string, []>("op_55831_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_55831_end_0 = const()[name = tensor<string, []>("op_55831_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_55831_end_mask_0 = const()[name = tensor<string, []>("op_55831_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_55831_cast_fp16 = slice_by_index(begin = var_55831_begin_0, end = var_55831_end_0, end_mask = var_55831_end_mask_0, x = k_495_cast_fp16)[name = tensor<string, []>("op_55831_cast_fp16")];
            tensor<int32, [4]> var_55835_begin_0 = const()[name = tensor<string, []>("op_55835_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_55835_end_0 = const()[name = tensor<string, []>("op_55835_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_55835_end_mask_0 = const()[name = tensor<string, []>("op_55835_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_55835_cast_fp16 = slice_by_index(begin = var_55835_begin_0, end = var_55835_end_0, end_mask = var_55835_end_mask_0, x = k_495_cast_fp16)[name = tensor<string, []>("op_55835_cast_fp16")];
            tensor<int32, [4]> var_55839_begin_0 = const()[name = tensor<string, []>("op_55839_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_55839_end_0 = const()[name = tensor<string, []>("op_55839_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_55839_end_mask_0 = const()[name = tensor<string, []>("op_55839_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_55839_cast_fp16 = slice_by_index(begin = var_55839_begin_0, end = var_55839_end_0, end_mask = var_55839_end_mask_0, x = k_495_cast_fp16)[name = tensor<string, []>("op_55839_cast_fp16")];
            tensor<int32, [4]> var_55843_begin_0 = const()[name = tensor<string, []>("op_55843_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_55843_end_0 = const()[name = tensor<string, []>("op_55843_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_55843_end_mask_0 = const()[name = tensor<string, []>("op_55843_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_55843_cast_fp16 = slice_by_index(begin = var_55843_begin_0, end = var_55843_end_0, end_mask = var_55843_end_mask_0, x = k_495_cast_fp16)[name = tensor<string, []>("op_55843_cast_fp16")];
            tensor<int32, [4]> var_55847_begin_0 = const()[name = tensor<string, []>("op_55847_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_55847_end_0 = const()[name = tensor<string, []>("op_55847_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_55847_end_mask_0 = const()[name = tensor<string, []>("op_55847_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_55847_cast_fp16 = slice_by_index(begin = var_55847_begin_0, end = var_55847_end_0, end_mask = var_55847_end_mask_0, x = k_495_cast_fp16)[name = tensor<string, []>("op_55847_cast_fp16")];
            tensor<int32, [4]> var_55851_begin_0 = const()[name = tensor<string, []>("op_55851_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_55851_end_0 = const()[name = tensor<string, []>("op_55851_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_55851_end_mask_0 = const()[name = tensor<string, []>("op_55851_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_55851_cast_fp16 = slice_by_index(begin = var_55851_begin_0, end = var_55851_end_0, end_mask = var_55851_end_mask_0, x = k_495_cast_fp16)[name = tensor<string, []>("op_55851_cast_fp16")];
            tensor<int32, [4]> var_55855_begin_0 = const()[name = tensor<string, []>("op_55855_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_55855_end_0 = const()[name = tensor<string, []>("op_55855_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_55855_end_mask_0 = const()[name = tensor<string, []>("op_55855_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_55855_cast_fp16 = slice_by_index(begin = var_55855_begin_0, end = var_55855_end_0, end_mask = var_55855_end_mask_0, x = k_495_cast_fp16)[name = tensor<string, []>("op_55855_cast_fp16")];
            tensor<int32, [4]> var_55859_begin_0 = const()[name = tensor<string, []>("op_55859_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_55859_end_0 = const()[name = tensor<string, []>("op_55859_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_55859_end_mask_0 = const()[name = tensor<string, []>("op_55859_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_55859_cast_fp16 = slice_by_index(begin = var_55859_begin_0, end = var_55859_end_0, end_mask = var_55859_end_mask_0, x = k_495_cast_fp16)[name = tensor<string, []>("op_55859_cast_fp16")];
            tensor<int32, [4]> var_55863_begin_0 = const()[name = tensor<string, []>("op_55863_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_55863_end_0 = const()[name = tensor<string, []>("op_55863_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_55863_end_mask_0 = const()[name = tensor<string, []>("op_55863_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_55863_cast_fp16 = slice_by_index(begin = var_55863_begin_0, end = var_55863_end_0, end_mask = var_55863_end_mask_0, x = k_495_cast_fp16)[name = tensor<string, []>("op_55863_cast_fp16")];
            tensor<int32, [4]> var_55867_begin_0 = const()[name = tensor<string, []>("op_55867_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_55867_end_0 = const()[name = tensor<string, []>("op_55867_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_55867_end_mask_0 = const()[name = tensor<string, []>("op_55867_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_55867_cast_fp16 = slice_by_index(begin = var_55867_begin_0, end = var_55867_end_0, end_mask = var_55867_end_mask_0, x = k_495_cast_fp16)[name = tensor<string, []>("op_55867_cast_fp16")];
            tensor<int32, [4]> var_55871_begin_0 = const()[name = tensor<string, []>("op_55871_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_55871_end_0 = const()[name = tensor<string, []>("op_55871_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_55871_end_mask_0 = const()[name = tensor<string, []>("op_55871_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_55871_cast_fp16 = slice_by_index(begin = var_55871_begin_0, end = var_55871_end_0, end_mask = var_55871_end_mask_0, x = k_495_cast_fp16)[name = tensor<string, []>("op_55871_cast_fp16")];
            tensor<int32, [4]> var_55875_begin_0 = const()[name = tensor<string, []>("op_55875_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_55875_end_0 = const()[name = tensor<string, []>("op_55875_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_55875_end_mask_0 = const()[name = tensor<string, []>("op_55875_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_55875_cast_fp16 = slice_by_index(begin = var_55875_begin_0, end = var_55875_end_0, end_mask = var_55875_end_mask_0, x = k_495_cast_fp16)[name = tensor<string, []>("op_55875_cast_fp16")];
            tensor<int32, [4]> var_55879_begin_0 = const()[name = tensor<string, []>("op_55879_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_55879_end_0 = const()[name = tensor<string, []>("op_55879_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_55879_end_mask_0 = const()[name = tensor<string, []>("op_55879_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_55879_cast_fp16 = slice_by_index(begin = var_55879_begin_0, end = var_55879_end_0, end_mask = var_55879_end_mask_0, x = k_495_cast_fp16)[name = tensor<string, []>("op_55879_cast_fp16")];
            tensor<int32, [4]> var_55883_begin_0 = const()[name = tensor<string, []>("op_55883_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_55883_end_0 = const()[name = tensor<string, []>("op_55883_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_55883_end_mask_0 = const()[name = tensor<string, []>("op_55883_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_55883_cast_fp16 = slice_by_index(begin = var_55883_begin_0, end = var_55883_end_0, end_mask = var_55883_end_mask_0, x = k_495_cast_fp16)[name = tensor<string, []>("op_55883_cast_fp16")];
            tensor<int32, [4]> var_55887_begin_0 = const()[name = tensor<string, []>("op_55887_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_55887_end_0 = const()[name = tensor<string, []>("op_55887_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_55887_end_mask_0 = const()[name = tensor<string, []>("op_55887_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_55887_cast_fp16 = slice_by_index(begin = var_55887_begin_0, end = var_55887_end_0, end_mask = var_55887_end_mask_0, x = k_495_cast_fp16)[name = tensor<string, []>("op_55887_cast_fp16")];
            tensor<int32, [4]> var_55891_begin_0 = const()[name = tensor<string, []>("op_55891_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_55891_end_0 = const()[name = tensor<string, []>("op_55891_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_55891_end_mask_0 = const()[name = tensor<string, []>("op_55891_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_55891_cast_fp16 = slice_by_index(begin = var_55891_begin_0, end = var_55891_end_0, end_mask = var_55891_end_mask_0, x = k_495_cast_fp16)[name = tensor<string, []>("op_55891_cast_fp16")];
            tensor<int32, [4]> var_55895_begin_0 = const()[name = tensor<string, []>("op_55895_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_55895_end_0 = const()[name = tensor<string, []>("op_55895_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_55895_end_mask_0 = const()[name = tensor<string, []>("op_55895_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_55895_cast_fp16 = slice_by_index(begin = var_55895_begin_0, end = var_55895_end_0, end_mask = var_55895_end_mask_0, x = k_495_cast_fp16)[name = tensor<string, []>("op_55895_cast_fp16")];
            tensor<int32, [4]> var_55897_begin_0 = const()[name = tensor<string, []>("op_55897_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_55897_end_0 = const()[name = tensor<string, []>("op_55897_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_55897_end_mask_0 = const()[name = tensor<string, []>("op_55897_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55897_cast_fp16 = slice_by_index(begin = var_55897_begin_0, end = var_55897_end_0, end_mask = var_55897_end_mask_0, x = v_247_cast_fp16)[name = tensor<string, []>("op_55897_cast_fp16")];
            tensor<int32, [4]> var_55901_begin_0 = const()[name = tensor<string, []>("op_55901_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_55901_end_0 = const()[name = tensor<string, []>("op_55901_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_55901_end_mask_0 = const()[name = tensor<string, []>("op_55901_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55901_cast_fp16 = slice_by_index(begin = var_55901_begin_0, end = var_55901_end_0, end_mask = var_55901_end_mask_0, x = v_247_cast_fp16)[name = tensor<string, []>("op_55901_cast_fp16")];
            tensor<int32, [4]> var_55905_begin_0 = const()[name = tensor<string, []>("op_55905_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_55905_end_0 = const()[name = tensor<string, []>("op_55905_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_55905_end_mask_0 = const()[name = tensor<string, []>("op_55905_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55905_cast_fp16 = slice_by_index(begin = var_55905_begin_0, end = var_55905_end_0, end_mask = var_55905_end_mask_0, x = v_247_cast_fp16)[name = tensor<string, []>("op_55905_cast_fp16")];
            tensor<int32, [4]> var_55909_begin_0 = const()[name = tensor<string, []>("op_55909_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_55909_end_0 = const()[name = tensor<string, []>("op_55909_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_55909_end_mask_0 = const()[name = tensor<string, []>("op_55909_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55909_cast_fp16 = slice_by_index(begin = var_55909_begin_0, end = var_55909_end_0, end_mask = var_55909_end_mask_0, x = v_247_cast_fp16)[name = tensor<string, []>("op_55909_cast_fp16")];
            tensor<int32, [4]> var_55913_begin_0 = const()[name = tensor<string, []>("op_55913_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_55913_end_0 = const()[name = tensor<string, []>("op_55913_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_55913_end_mask_0 = const()[name = tensor<string, []>("op_55913_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55913_cast_fp16 = slice_by_index(begin = var_55913_begin_0, end = var_55913_end_0, end_mask = var_55913_end_mask_0, x = v_247_cast_fp16)[name = tensor<string, []>("op_55913_cast_fp16")];
            tensor<int32, [4]> var_55917_begin_0 = const()[name = tensor<string, []>("op_55917_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_55917_end_0 = const()[name = tensor<string, []>("op_55917_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_55917_end_mask_0 = const()[name = tensor<string, []>("op_55917_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55917_cast_fp16 = slice_by_index(begin = var_55917_begin_0, end = var_55917_end_0, end_mask = var_55917_end_mask_0, x = v_247_cast_fp16)[name = tensor<string, []>("op_55917_cast_fp16")];
            tensor<int32, [4]> var_55921_begin_0 = const()[name = tensor<string, []>("op_55921_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_55921_end_0 = const()[name = tensor<string, []>("op_55921_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_55921_end_mask_0 = const()[name = tensor<string, []>("op_55921_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55921_cast_fp16 = slice_by_index(begin = var_55921_begin_0, end = var_55921_end_0, end_mask = var_55921_end_mask_0, x = v_247_cast_fp16)[name = tensor<string, []>("op_55921_cast_fp16")];
            tensor<int32, [4]> var_55925_begin_0 = const()[name = tensor<string, []>("op_55925_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_55925_end_0 = const()[name = tensor<string, []>("op_55925_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_55925_end_mask_0 = const()[name = tensor<string, []>("op_55925_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55925_cast_fp16 = slice_by_index(begin = var_55925_begin_0, end = var_55925_end_0, end_mask = var_55925_end_mask_0, x = v_247_cast_fp16)[name = tensor<string, []>("op_55925_cast_fp16")];
            tensor<int32, [4]> var_55929_begin_0 = const()[name = tensor<string, []>("op_55929_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_55929_end_0 = const()[name = tensor<string, []>("op_55929_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_55929_end_mask_0 = const()[name = tensor<string, []>("op_55929_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55929_cast_fp16 = slice_by_index(begin = var_55929_begin_0, end = var_55929_end_0, end_mask = var_55929_end_mask_0, x = v_247_cast_fp16)[name = tensor<string, []>("op_55929_cast_fp16")];
            tensor<int32, [4]> var_55933_begin_0 = const()[name = tensor<string, []>("op_55933_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_55933_end_0 = const()[name = tensor<string, []>("op_55933_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_55933_end_mask_0 = const()[name = tensor<string, []>("op_55933_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55933_cast_fp16 = slice_by_index(begin = var_55933_begin_0, end = var_55933_end_0, end_mask = var_55933_end_mask_0, x = v_247_cast_fp16)[name = tensor<string, []>("op_55933_cast_fp16")];
            tensor<int32, [4]> var_55937_begin_0 = const()[name = tensor<string, []>("op_55937_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_55937_end_0 = const()[name = tensor<string, []>("op_55937_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_55937_end_mask_0 = const()[name = tensor<string, []>("op_55937_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55937_cast_fp16 = slice_by_index(begin = var_55937_begin_0, end = var_55937_end_0, end_mask = var_55937_end_mask_0, x = v_247_cast_fp16)[name = tensor<string, []>("op_55937_cast_fp16")];
            tensor<int32, [4]> var_55941_begin_0 = const()[name = tensor<string, []>("op_55941_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_55941_end_0 = const()[name = tensor<string, []>("op_55941_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_55941_end_mask_0 = const()[name = tensor<string, []>("op_55941_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55941_cast_fp16 = slice_by_index(begin = var_55941_begin_0, end = var_55941_end_0, end_mask = var_55941_end_mask_0, x = v_247_cast_fp16)[name = tensor<string, []>("op_55941_cast_fp16")];
            tensor<int32, [4]> var_55945_begin_0 = const()[name = tensor<string, []>("op_55945_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_55945_end_0 = const()[name = tensor<string, []>("op_55945_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_55945_end_mask_0 = const()[name = tensor<string, []>("op_55945_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55945_cast_fp16 = slice_by_index(begin = var_55945_begin_0, end = var_55945_end_0, end_mask = var_55945_end_mask_0, x = v_247_cast_fp16)[name = tensor<string, []>("op_55945_cast_fp16")];
            tensor<int32, [4]> var_55949_begin_0 = const()[name = tensor<string, []>("op_55949_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_55949_end_0 = const()[name = tensor<string, []>("op_55949_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_55949_end_mask_0 = const()[name = tensor<string, []>("op_55949_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55949_cast_fp16 = slice_by_index(begin = var_55949_begin_0, end = var_55949_end_0, end_mask = var_55949_end_mask_0, x = v_247_cast_fp16)[name = tensor<string, []>("op_55949_cast_fp16")];
            tensor<int32, [4]> var_55953_begin_0 = const()[name = tensor<string, []>("op_55953_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_55953_end_0 = const()[name = tensor<string, []>("op_55953_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_55953_end_mask_0 = const()[name = tensor<string, []>("op_55953_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55953_cast_fp16 = slice_by_index(begin = var_55953_begin_0, end = var_55953_end_0, end_mask = var_55953_end_mask_0, x = v_247_cast_fp16)[name = tensor<string, []>("op_55953_cast_fp16")];
            tensor<int32, [4]> var_55957_begin_0 = const()[name = tensor<string, []>("op_55957_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_55957_end_0 = const()[name = tensor<string, []>("op_55957_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_55957_end_mask_0 = const()[name = tensor<string, []>("op_55957_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55957_cast_fp16 = slice_by_index(begin = var_55957_begin_0, end = var_55957_end_0, end_mask = var_55957_end_mask_0, x = v_247_cast_fp16)[name = tensor<string, []>("op_55957_cast_fp16")];
            tensor<int32, [4]> var_55961_begin_0 = const()[name = tensor<string, []>("op_55961_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_55961_end_0 = const()[name = tensor<string, []>("op_55961_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_55961_end_mask_0 = const()[name = tensor<string, []>("op_55961_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55961_cast_fp16 = slice_by_index(begin = var_55961_begin_0, end = var_55961_end_0, end_mask = var_55961_end_mask_0, x = v_247_cast_fp16)[name = tensor<string, []>("op_55961_cast_fp16")];
            tensor<int32, [4]> var_55965_begin_0 = const()[name = tensor<string, []>("op_55965_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_55965_end_0 = const()[name = tensor<string, []>("op_55965_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_55965_end_mask_0 = const()[name = tensor<string, []>("op_55965_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55965_cast_fp16 = slice_by_index(begin = var_55965_begin_0, end = var_55965_end_0, end_mask = var_55965_end_mask_0, x = v_247_cast_fp16)[name = tensor<string, []>("op_55965_cast_fp16")];
            tensor<int32, [4]> var_55969_begin_0 = const()[name = tensor<string, []>("op_55969_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_55969_end_0 = const()[name = tensor<string, []>("op_55969_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_55969_end_mask_0 = const()[name = tensor<string, []>("op_55969_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55969_cast_fp16 = slice_by_index(begin = var_55969_begin_0, end = var_55969_end_0, end_mask = var_55969_end_mask_0, x = v_247_cast_fp16)[name = tensor<string, []>("op_55969_cast_fp16")];
            tensor<int32, [4]> var_55973_begin_0 = const()[name = tensor<string, []>("op_55973_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_55973_end_0 = const()[name = tensor<string, []>("op_55973_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_55973_end_mask_0 = const()[name = tensor<string, []>("op_55973_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55973_cast_fp16 = slice_by_index(begin = var_55973_begin_0, end = var_55973_end_0, end_mask = var_55973_end_mask_0, x = v_247_cast_fp16)[name = tensor<string, []>("op_55973_cast_fp16")];
            tensor<string, []> var_55977_equation_0 = const()[name = tensor<string, []>("op_55977_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_55977_cast_fp16 = einsum(equation = var_55977_equation_0, values = (var_55819_cast_fp16, var_55736_cast_fp16))[name = tensor<string, []>("op_55977_cast_fp16")];
            tensor<fp16, []> var_55978_to_fp16 = const()[name = tensor<string, []>("op_55978_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4761_cast_fp16 = mul(x = var_55977_cast_fp16, y = var_55978_to_fp16)[name = tensor<string, []>("aw_4761_cast_fp16")];
            tensor<string, []> var_55981_equation_0 = const()[name = tensor<string, []>("op_55981_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_55981_cast_fp16 = einsum(equation = var_55981_equation_0, values = (var_55823_cast_fp16, var_55740_cast_fp16))[name = tensor<string, []>("op_55981_cast_fp16")];
            tensor<fp16, []> var_55982_to_fp16 = const()[name = tensor<string, []>("op_55982_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4763_cast_fp16 = mul(x = var_55981_cast_fp16, y = var_55982_to_fp16)[name = tensor<string, []>("aw_4763_cast_fp16")];
            tensor<string, []> var_55985_equation_0 = const()[name = tensor<string, []>("op_55985_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_55985_cast_fp16 = einsum(equation = var_55985_equation_0, values = (var_55827_cast_fp16, var_55744_cast_fp16))[name = tensor<string, []>("op_55985_cast_fp16")];
            tensor<fp16, []> var_55986_to_fp16 = const()[name = tensor<string, []>("op_55986_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4765_cast_fp16 = mul(x = var_55985_cast_fp16, y = var_55986_to_fp16)[name = tensor<string, []>("aw_4765_cast_fp16")];
            tensor<string, []> var_55989_equation_0 = const()[name = tensor<string, []>("op_55989_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_55989_cast_fp16 = einsum(equation = var_55989_equation_0, values = (var_55831_cast_fp16, var_55748_cast_fp16))[name = tensor<string, []>("op_55989_cast_fp16")];
            tensor<fp16, []> var_55990_to_fp16 = const()[name = tensor<string, []>("op_55990_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4767_cast_fp16 = mul(x = var_55989_cast_fp16, y = var_55990_to_fp16)[name = tensor<string, []>("aw_4767_cast_fp16")];
            tensor<string, []> var_55993_equation_0 = const()[name = tensor<string, []>("op_55993_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_55993_cast_fp16 = einsum(equation = var_55993_equation_0, values = (var_55835_cast_fp16, var_55752_cast_fp16))[name = tensor<string, []>("op_55993_cast_fp16")];
            tensor<fp16, []> var_55994_to_fp16 = const()[name = tensor<string, []>("op_55994_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4769_cast_fp16 = mul(x = var_55993_cast_fp16, y = var_55994_to_fp16)[name = tensor<string, []>("aw_4769_cast_fp16")];
            tensor<string, []> var_55997_equation_0 = const()[name = tensor<string, []>("op_55997_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_55997_cast_fp16 = einsum(equation = var_55997_equation_0, values = (var_55839_cast_fp16, var_55756_cast_fp16))[name = tensor<string, []>("op_55997_cast_fp16")];
            tensor<fp16, []> var_55998_to_fp16 = const()[name = tensor<string, []>("op_55998_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4771_cast_fp16 = mul(x = var_55997_cast_fp16, y = var_55998_to_fp16)[name = tensor<string, []>("aw_4771_cast_fp16")];
            tensor<string, []> var_56001_equation_0 = const()[name = tensor<string, []>("op_56001_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56001_cast_fp16 = einsum(equation = var_56001_equation_0, values = (var_55843_cast_fp16, var_55760_cast_fp16))[name = tensor<string, []>("op_56001_cast_fp16")];
            tensor<fp16, []> var_56002_to_fp16 = const()[name = tensor<string, []>("op_56002_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4773_cast_fp16 = mul(x = var_56001_cast_fp16, y = var_56002_to_fp16)[name = tensor<string, []>("aw_4773_cast_fp16")];
            tensor<string, []> var_56005_equation_0 = const()[name = tensor<string, []>("op_56005_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56005_cast_fp16 = einsum(equation = var_56005_equation_0, values = (var_55847_cast_fp16, var_55764_cast_fp16))[name = tensor<string, []>("op_56005_cast_fp16")];
            tensor<fp16, []> var_56006_to_fp16 = const()[name = tensor<string, []>("op_56006_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4775_cast_fp16 = mul(x = var_56005_cast_fp16, y = var_56006_to_fp16)[name = tensor<string, []>("aw_4775_cast_fp16")];
            tensor<string, []> var_56009_equation_0 = const()[name = tensor<string, []>("op_56009_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56009_cast_fp16 = einsum(equation = var_56009_equation_0, values = (var_55851_cast_fp16, var_55768_cast_fp16))[name = tensor<string, []>("op_56009_cast_fp16")];
            tensor<fp16, []> var_56010_to_fp16 = const()[name = tensor<string, []>("op_56010_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4777_cast_fp16 = mul(x = var_56009_cast_fp16, y = var_56010_to_fp16)[name = tensor<string, []>("aw_4777_cast_fp16")];
            tensor<string, []> var_56013_equation_0 = const()[name = tensor<string, []>("op_56013_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56013_cast_fp16 = einsum(equation = var_56013_equation_0, values = (var_55855_cast_fp16, var_55772_cast_fp16))[name = tensor<string, []>("op_56013_cast_fp16")];
            tensor<fp16, []> var_56014_to_fp16 = const()[name = tensor<string, []>("op_56014_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4779_cast_fp16 = mul(x = var_56013_cast_fp16, y = var_56014_to_fp16)[name = tensor<string, []>("aw_4779_cast_fp16")];
            tensor<string, []> var_56017_equation_0 = const()[name = tensor<string, []>("op_56017_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56017_cast_fp16 = einsum(equation = var_56017_equation_0, values = (var_55859_cast_fp16, var_55776_cast_fp16))[name = tensor<string, []>("op_56017_cast_fp16")];
            tensor<fp16, []> var_56018_to_fp16 = const()[name = tensor<string, []>("op_56018_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4781_cast_fp16 = mul(x = var_56017_cast_fp16, y = var_56018_to_fp16)[name = tensor<string, []>("aw_4781_cast_fp16")];
            tensor<string, []> var_56021_equation_0 = const()[name = tensor<string, []>("op_56021_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56021_cast_fp16 = einsum(equation = var_56021_equation_0, values = (var_55863_cast_fp16, var_55780_cast_fp16))[name = tensor<string, []>("op_56021_cast_fp16")];
            tensor<fp16, []> var_56022_to_fp16 = const()[name = tensor<string, []>("op_56022_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4783_cast_fp16 = mul(x = var_56021_cast_fp16, y = var_56022_to_fp16)[name = tensor<string, []>("aw_4783_cast_fp16")];
            tensor<string, []> var_56025_equation_0 = const()[name = tensor<string, []>("op_56025_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56025_cast_fp16 = einsum(equation = var_56025_equation_0, values = (var_55867_cast_fp16, var_55784_cast_fp16))[name = tensor<string, []>("op_56025_cast_fp16")];
            tensor<fp16, []> var_56026_to_fp16 = const()[name = tensor<string, []>("op_56026_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4785_cast_fp16 = mul(x = var_56025_cast_fp16, y = var_56026_to_fp16)[name = tensor<string, []>("aw_4785_cast_fp16")];
            tensor<string, []> var_56029_equation_0 = const()[name = tensor<string, []>("op_56029_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56029_cast_fp16 = einsum(equation = var_56029_equation_0, values = (var_55871_cast_fp16, var_55788_cast_fp16))[name = tensor<string, []>("op_56029_cast_fp16")];
            tensor<fp16, []> var_56030_to_fp16 = const()[name = tensor<string, []>("op_56030_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4787_cast_fp16 = mul(x = var_56029_cast_fp16, y = var_56030_to_fp16)[name = tensor<string, []>("aw_4787_cast_fp16")];
            tensor<string, []> var_56033_equation_0 = const()[name = tensor<string, []>("op_56033_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56033_cast_fp16 = einsum(equation = var_56033_equation_0, values = (var_55875_cast_fp16, var_55792_cast_fp16))[name = tensor<string, []>("op_56033_cast_fp16")];
            tensor<fp16, []> var_56034_to_fp16 = const()[name = tensor<string, []>("op_56034_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4789_cast_fp16 = mul(x = var_56033_cast_fp16, y = var_56034_to_fp16)[name = tensor<string, []>("aw_4789_cast_fp16")];
            tensor<string, []> var_56037_equation_0 = const()[name = tensor<string, []>("op_56037_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56037_cast_fp16 = einsum(equation = var_56037_equation_0, values = (var_55879_cast_fp16, var_55796_cast_fp16))[name = tensor<string, []>("op_56037_cast_fp16")];
            tensor<fp16, []> var_56038_to_fp16 = const()[name = tensor<string, []>("op_56038_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4791_cast_fp16 = mul(x = var_56037_cast_fp16, y = var_56038_to_fp16)[name = tensor<string, []>("aw_4791_cast_fp16")];
            tensor<string, []> var_56041_equation_0 = const()[name = tensor<string, []>("op_56041_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56041_cast_fp16 = einsum(equation = var_56041_equation_0, values = (var_55883_cast_fp16, var_55800_cast_fp16))[name = tensor<string, []>("op_56041_cast_fp16")];
            tensor<fp16, []> var_56042_to_fp16 = const()[name = tensor<string, []>("op_56042_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4793_cast_fp16 = mul(x = var_56041_cast_fp16, y = var_56042_to_fp16)[name = tensor<string, []>("aw_4793_cast_fp16")];
            tensor<string, []> var_56045_equation_0 = const()[name = tensor<string, []>("op_56045_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56045_cast_fp16 = einsum(equation = var_56045_equation_0, values = (var_55887_cast_fp16, var_55804_cast_fp16))[name = tensor<string, []>("op_56045_cast_fp16")];
            tensor<fp16, []> var_56046_to_fp16 = const()[name = tensor<string, []>("op_56046_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4795_cast_fp16 = mul(x = var_56045_cast_fp16, y = var_56046_to_fp16)[name = tensor<string, []>("aw_4795_cast_fp16")];
            tensor<string, []> var_56049_equation_0 = const()[name = tensor<string, []>("op_56049_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56049_cast_fp16 = einsum(equation = var_56049_equation_0, values = (var_55891_cast_fp16, var_55808_cast_fp16))[name = tensor<string, []>("op_56049_cast_fp16")];
            tensor<fp16, []> var_56050_to_fp16 = const()[name = tensor<string, []>("op_56050_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4797_cast_fp16 = mul(x = var_56049_cast_fp16, y = var_56050_to_fp16)[name = tensor<string, []>("aw_4797_cast_fp16")];
            tensor<string, []> var_56053_equation_0 = const()[name = tensor<string, []>("op_56053_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56053_cast_fp16 = einsum(equation = var_56053_equation_0, values = (var_55895_cast_fp16, var_55812_cast_fp16))[name = tensor<string, []>("op_56053_cast_fp16")];
            tensor<fp16, []> var_56054_to_fp16 = const()[name = tensor<string, []>("op_56054_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4799_cast_fp16 = mul(x = var_56053_cast_fp16, y = var_56054_to_fp16)[name = tensor<string, []>("aw_4799_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_56056_cast_fp16 = softmax(axis = var_30355, x = aw_4761_cast_fp16)[name = tensor<string, []>("op_56056_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_56057_cast_fp16 = softmax(axis = var_30355, x = aw_4763_cast_fp16)[name = tensor<string, []>("op_56057_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_56058_cast_fp16 = softmax(axis = var_30355, x = aw_4765_cast_fp16)[name = tensor<string, []>("op_56058_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_56059_cast_fp16 = softmax(axis = var_30355, x = aw_4767_cast_fp16)[name = tensor<string, []>("op_56059_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_56060_cast_fp16 = softmax(axis = var_30355, x = aw_4769_cast_fp16)[name = tensor<string, []>("op_56060_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_56061_cast_fp16 = softmax(axis = var_30355, x = aw_4771_cast_fp16)[name = tensor<string, []>("op_56061_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_56062_cast_fp16 = softmax(axis = var_30355, x = aw_4773_cast_fp16)[name = tensor<string, []>("op_56062_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_56063_cast_fp16 = softmax(axis = var_30355, x = aw_4775_cast_fp16)[name = tensor<string, []>("op_56063_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_56064_cast_fp16 = softmax(axis = var_30355, x = aw_4777_cast_fp16)[name = tensor<string, []>("op_56064_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_56065_cast_fp16 = softmax(axis = var_30355, x = aw_4779_cast_fp16)[name = tensor<string, []>("op_56065_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_56066_cast_fp16 = softmax(axis = var_30355, x = aw_4781_cast_fp16)[name = tensor<string, []>("op_56066_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_56067_cast_fp16 = softmax(axis = var_30355, x = aw_4783_cast_fp16)[name = tensor<string, []>("op_56067_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_56068_cast_fp16 = softmax(axis = var_30355, x = aw_4785_cast_fp16)[name = tensor<string, []>("op_56068_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_56069_cast_fp16 = softmax(axis = var_30355, x = aw_4787_cast_fp16)[name = tensor<string, []>("op_56069_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_56070_cast_fp16 = softmax(axis = var_30355, x = aw_4789_cast_fp16)[name = tensor<string, []>("op_56070_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_56071_cast_fp16 = softmax(axis = var_30355, x = aw_4791_cast_fp16)[name = tensor<string, []>("op_56071_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_56072_cast_fp16 = softmax(axis = var_30355, x = aw_4793_cast_fp16)[name = tensor<string, []>("op_56072_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_56073_cast_fp16 = softmax(axis = var_30355, x = aw_4795_cast_fp16)[name = tensor<string, []>("op_56073_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_56074_cast_fp16 = softmax(axis = var_30355, x = aw_4797_cast_fp16)[name = tensor<string, []>("op_56074_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_56075_cast_fp16 = softmax(axis = var_30355, x = aw_4799_cast_fp16)[name = tensor<string, []>("op_56075_cast_fp16")];
            tensor<string, []> var_56077_equation_0 = const()[name = tensor<string, []>("op_56077_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56077_cast_fp16 = einsum(equation = var_56077_equation_0, values = (var_55897_cast_fp16, var_56056_cast_fp16))[name = tensor<string, []>("op_56077_cast_fp16")];
            tensor<string, []> var_56079_equation_0 = const()[name = tensor<string, []>("op_56079_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56079_cast_fp16 = einsum(equation = var_56079_equation_0, values = (var_55901_cast_fp16, var_56057_cast_fp16))[name = tensor<string, []>("op_56079_cast_fp16")];
            tensor<string, []> var_56081_equation_0 = const()[name = tensor<string, []>("op_56081_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56081_cast_fp16 = einsum(equation = var_56081_equation_0, values = (var_55905_cast_fp16, var_56058_cast_fp16))[name = tensor<string, []>("op_56081_cast_fp16")];
            tensor<string, []> var_56083_equation_0 = const()[name = tensor<string, []>("op_56083_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56083_cast_fp16 = einsum(equation = var_56083_equation_0, values = (var_55909_cast_fp16, var_56059_cast_fp16))[name = tensor<string, []>("op_56083_cast_fp16")];
            tensor<string, []> var_56085_equation_0 = const()[name = tensor<string, []>("op_56085_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56085_cast_fp16 = einsum(equation = var_56085_equation_0, values = (var_55913_cast_fp16, var_56060_cast_fp16))[name = tensor<string, []>("op_56085_cast_fp16")];
            tensor<string, []> var_56087_equation_0 = const()[name = tensor<string, []>("op_56087_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56087_cast_fp16 = einsum(equation = var_56087_equation_0, values = (var_55917_cast_fp16, var_56061_cast_fp16))[name = tensor<string, []>("op_56087_cast_fp16")];
            tensor<string, []> var_56089_equation_0 = const()[name = tensor<string, []>("op_56089_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56089_cast_fp16 = einsum(equation = var_56089_equation_0, values = (var_55921_cast_fp16, var_56062_cast_fp16))[name = tensor<string, []>("op_56089_cast_fp16")];
            tensor<string, []> var_56091_equation_0 = const()[name = tensor<string, []>("op_56091_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56091_cast_fp16 = einsum(equation = var_56091_equation_0, values = (var_55925_cast_fp16, var_56063_cast_fp16))[name = tensor<string, []>("op_56091_cast_fp16")];
            tensor<string, []> var_56093_equation_0 = const()[name = tensor<string, []>("op_56093_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56093_cast_fp16 = einsum(equation = var_56093_equation_0, values = (var_55929_cast_fp16, var_56064_cast_fp16))[name = tensor<string, []>("op_56093_cast_fp16")];
            tensor<string, []> var_56095_equation_0 = const()[name = tensor<string, []>("op_56095_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56095_cast_fp16 = einsum(equation = var_56095_equation_0, values = (var_55933_cast_fp16, var_56065_cast_fp16))[name = tensor<string, []>("op_56095_cast_fp16")];
            tensor<string, []> var_56097_equation_0 = const()[name = tensor<string, []>("op_56097_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56097_cast_fp16 = einsum(equation = var_56097_equation_0, values = (var_55937_cast_fp16, var_56066_cast_fp16))[name = tensor<string, []>("op_56097_cast_fp16")];
            tensor<string, []> var_56099_equation_0 = const()[name = tensor<string, []>("op_56099_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56099_cast_fp16 = einsum(equation = var_56099_equation_0, values = (var_55941_cast_fp16, var_56067_cast_fp16))[name = tensor<string, []>("op_56099_cast_fp16")];
            tensor<string, []> var_56101_equation_0 = const()[name = tensor<string, []>("op_56101_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56101_cast_fp16 = einsum(equation = var_56101_equation_0, values = (var_55945_cast_fp16, var_56068_cast_fp16))[name = tensor<string, []>("op_56101_cast_fp16")];
            tensor<string, []> var_56103_equation_0 = const()[name = tensor<string, []>("op_56103_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56103_cast_fp16 = einsum(equation = var_56103_equation_0, values = (var_55949_cast_fp16, var_56069_cast_fp16))[name = tensor<string, []>("op_56103_cast_fp16")];
            tensor<string, []> var_56105_equation_0 = const()[name = tensor<string, []>("op_56105_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56105_cast_fp16 = einsum(equation = var_56105_equation_0, values = (var_55953_cast_fp16, var_56070_cast_fp16))[name = tensor<string, []>("op_56105_cast_fp16")];
            tensor<string, []> var_56107_equation_0 = const()[name = tensor<string, []>("op_56107_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56107_cast_fp16 = einsum(equation = var_56107_equation_0, values = (var_55957_cast_fp16, var_56071_cast_fp16))[name = tensor<string, []>("op_56107_cast_fp16")];
            tensor<string, []> var_56109_equation_0 = const()[name = tensor<string, []>("op_56109_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56109_cast_fp16 = einsum(equation = var_56109_equation_0, values = (var_55961_cast_fp16, var_56072_cast_fp16))[name = tensor<string, []>("op_56109_cast_fp16")];
            tensor<string, []> var_56111_equation_0 = const()[name = tensor<string, []>("op_56111_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56111_cast_fp16 = einsum(equation = var_56111_equation_0, values = (var_55965_cast_fp16, var_56073_cast_fp16))[name = tensor<string, []>("op_56111_cast_fp16")];
            tensor<string, []> var_56113_equation_0 = const()[name = tensor<string, []>("op_56113_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56113_cast_fp16 = einsum(equation = var_56113_equation_0, values = (var_55969_cast_fp16, var_56074_cast_fp16))[name = tensor<string, []>("op_56113_cast_fp16")];
            tensor<string, []> var_56115_equation_0 = const()[name = tensor<string, []>("op_56115_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56115_cast_fp16 = einsum(equation = var_56115_equation_0, values = (var_55973_cast_fp16, var_56075_cast_fp16))[name = tensor<string, []>("op_56115_cast_fp16")];
            tensor<bool, []> input_693_interleave_0 = const()[name = tensor<string, []>("input_693_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_693_cast_fp16 = concat(axis = var_30355, interleave = input_693_interleave_0, values = (var_56077_cast_fp16, var_56079_cast_fp16, var_56081_cast_fp16, var_56083_cast_fp16, var_56085_cast_fp16, var_56087_cast_fp16, var_56089_cast_fp16, var_56091_cast_fp16, var_56093_cast_fp16, var_56095_cast_fp16, var_56097_cast_fp16, var_56099_cast_fp16, var_56101_cast_fp16, var_56103_cast_fp16, var_56105_cast_fp16, var_56107_cast_fp16, var_56109_cast_fp16, var_56111_cast_fp16, var_56113_cast_fp16, var_56115_cast_fp16))[name = tensor<string, []>("input_693_cast_fp16")];
            tensor<string, []> var_56125_pad_type_0 = const()[name = tensor<string, []>("op_56125_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_56125_strides_0 = const()[name = tensor<string, []>("op_56125_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_56125_pad_0 = const()[name = tensor<string, []>("op_56125_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_56125_dilations_0 = const()[name = tensor<string, []>("op_56125_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_56125_groups_0 = const()[name = tensor<string, []>("op_56125_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_7_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(791015552))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(792244416))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_7_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_7_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_7_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(792244608)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_56125_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_7_attn2_to_out_0_bias_to_fp16, dilations = var_56125_dilations_0, groups = var_56125_groups_0, pad = var_56125_pad_0, pad_type = var_56125_pad_type_0, strides = var_56125_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_7_attn2_to_out_0_weight_to_fp16_palettized, x = input_693_cast_fp16)[name = tensor<string, []>("op_56125_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_371_cast_fp16 = add(x = var_56125_cast_fp16, y = inputs_369_cast_fp16)[name = tensor<string, []>("inputs_371_cast_fp16")];
            tensor<int32, [1]> input_695_axes_0 = const()[name = tensor<string, []>("input_695_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> input_695_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_695_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(792247232)))];
            tensor<fp16, [1280]> input_695_beta_0_to_fp16 = const()[name = tensor<string, []>("input_695_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(792249856)))];
            tensor<fp16, []> var_56135_to_fp16 = const()[name = tensor<string, []>("op_56135_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> input_695_cast_fp16 = layer_norm(axes = input_695_axes_0, beta = input_695_beta_0_to_fp16, epsilon = var_56135_to_fp16, gamma = input_695_gamma_0_to_fp16, x = inputs_371_cast_fp16)[name = tensor<string, []>("input_695_cast_fp16")];
            tensor<string, []> var_56155_pad_type_0 = const()[name = tensor<string, []>("op_56155_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_56155_strides_0 = const()[name = tensor<string, []>("op_56155_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_56155_pad_0 = const()[name = tensor<string, []>("op_56155_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_56155_dilations_0 = const()[name = tensor<string, []>("op_56155_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_56155_groups_0 = const()[name = tensor<string, []>("op_56155_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_7_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [9830400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(792252480))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(802082944))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_7_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([10240, 1280, 1, 1])];
            tensor<fp16, [10240]> up_blocks_0_attentions_2_transformer_blocks_7_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_7_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(802083136)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_56155_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_7_ff_net_0_proj_bias_to_fp16, dilations = var_56155_dilations_0, groups = var_56155_groups_0, pad = var_56155_pad_0, pad_type = var_56155_pad_type_0, strides = var_56155_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_7_ff_net_0_proj_weight_to_fp16_palettized, x = input_695_cast_fp16)[name = tensor<string, []>("op_56155_cast_fp16")];
            tensor<int32, [2]> var_56156_split_sizes_0 = const()[name = tensor<string, []>("op_56156_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_56156_axis_0 = const()[name = tensor<string, []>("op_56156_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_56156_cast_fp16_0, tensor<fp16, [2, 5120, 1, 1024]> var_56156_cast_fp16_1 = split(axis = var_56156_axis_0, split_sizes = var_56156_split_sizes_0, x = var_56155_cast_fp16)[name = tensor<string, []>("op_56156_cast_fp16")];
            tensor<string, []> var_56158_mode_0 = const()[name = tensor<string, []>("op_56158_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_56158_cast_fp16 = gelu(mode = var_56158_mode_0, x = var_56156_cast_fp16_1)[name = tensor<string, []>("op_56158_cast_fp16")];
            tensor<fp16, [2, 5120, 1, 1024]> input_697_cast_fp16 = mul(x = var_56156_cast_fp16_0, y = var_56158_cast_fp16)[name = tensor<string, []>("input_697_cast_fp16")];
            tensor<string, []> var_56166_pad_type_0 = const()[name = tensor<string, []>("op_56166_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_56166_strides_0 = const()[name = tensor<string, []>("op_56166_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_56166_pad_0 = const()[name = tensor<string, []>("op_56166_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_56166_dilations_0 = const()[name = tensor<string, []>("op_56166_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_56166_groups_0 = const()[name = tensor<string, []>("op_56166_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_7_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [4915200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(802103680))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(807018944))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_7_ff_net_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 5120, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_7_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_7_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(807019136)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_56166_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_7_ff_net_2_bias_to_fp16, dilations = var_56166_dilations_0, groups = var_56166_groups_0, pad = var_56166_pad_0, pad_type = var_56166_pad_type_0, strides = var_56166_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_7_ff_net_2_weight_to_fp16_palettized, x = input_697_cast_fp16)[name = tensor<string, []>("op_56166_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_373_cast_fp16 = add(x = var_56166_cast_fp16, y = inputs_371_cast_fp16)[name = tensor<string, []>("inputs_373_cast_fp16")];
            tensor<int32, [1]> hidden_states_483_axes_0 = const()[name = tensor<string, []>("hidden_states_483_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_483_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_483_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(807021760)))];
            tensor<fp16, [1280]> hidden_states_483_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_483_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(807024384)))];
            tensor<fp16, []> var_56182_to_fp16 = const()[name = tensor<string, []>("op_56182_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_483_cast_fp16 = layer_norm(axes = hidden_states_483_axes_0, beta = hidden_states_483_beta_0_to_fp16, epsilon = var_56182_to_fp16, gamma = hidden_states_483_gamma_0_to_fp16, x = inputs_373_cast_fp16)[name = tensor<string, []>("hidden_states_483_cast_fp16")];
            tensor<string, []> q_249_pad_type_0 = const()[name = tensor<string, []>("q_249_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_249_strides_0 = const()[name = tensor<string, []>("q_249_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_249_pad_0 = const()[name = tensor<string, []>("q_249_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_249_dilations_0 = const()[name = tensor<string, []>("q_249_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_249_groups_0 = const()[name = tensor<string, []>("q_249_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_8_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(807027008))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(808255872))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_8_attn1_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_249_cast_fp16 = conv(dilations = q_249_dilations_0, groups = q_249_groups_0, pad = q_249_pad_0, pad_type = q_249_pad_type_0, strides = q_249_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_8_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_483_cast_fp16)[name = tensor<string, []>("q_249_cast_fp16")];
            tensor<string, []> k_497_pad_type_0 = const()[name = tensor<string, []>("k_497_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_497_strides_0 = const()[name = tensor<string, []>("k_497_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_497_pad_0 = const()[name = tensor<string, []>("k_497_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_497_dilations_0 = const()[name = tensor<string, []>("k_497_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_497_groups_0 = const()[name = tensor<string, []>("k_497_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_8_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(808256064))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(809484928))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_8_attn1_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> k_497_cast_fp16 = conv(dilations = k_497_dilations_0, groups = k_497_groups_0, pad = k_497_pad_0, pad_type = k_497_pad_type_0, strides = k_497_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_8_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_483_cast_fp16)[name = tensor<string, []>("k_497_cast_fp16")];
            tensor<string, []> v_249_pad_type_0 = const()[name = tensor<string, []>("v_249_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_249_strides_0 = const()[name = tensor<string, []>("v_249_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_249_pad_0 = const()[name = tensor<string, []>("v_249_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_249_dilations_0 = const()[name = tensor<string, []>("v_249_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_249_groups_0 = const()[name = tensor<string, []>("v_249_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_8_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(809485120))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(810713984))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_8_attn1_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> v_249_cast_fp16 = conv(dilations = v_249_dilations_0, groups = v_249_groups_0, pad = v_249_pad_0, pad_type = v_249_pad_type_0, strides = v_249_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_8_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_483_cast_fp16)[name = tensor<string, []>("v_249_cast_fp16")];
            tensor<int32, [4]> var_56215_begin_0 = const()[name = tensor<string, []>("op_56215_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_56215_end_0 = const()[name = tensor<string, []>("op_56215_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_56215_end_mask_0 = const()[name = tensor<string, []>("op_56215_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56215_cast_fp16 = slice_by_index(begin = var_56215_begin_0, end = var_56215_end_0, end_mask = var_56215_end_mask_0, x = q_249_cast_fp16)[name = tensor<string, []>("op_56215_cast_fp16")];
            tensor<int32, [4]> var_56219_begin_0 = const()[name = tensor<string, []>("op_56219_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_56219_end_0 = const()[name = tensor<string, []>("op_56219_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_56219_end_mask_0 = const()[name = tensor<string, []>("op_56219_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56219_cast_fp16 = slice_by_index(begin = var_56219_begin_0, end = var_56219_end_0, end_mask = var_56219_end_mask_0, x = q_249_cast_fp16)[name = tensor<string, []>("op_56219_cast_fp16")];
            tensor<int32, [4]> var_56223_begin_0 = const()[name = tensor<string, []>("op_56223_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_56223_end_0 = const()[name = tensor<string, []>("op_56223_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_56223_end_mask_0 = const()[name = tensor<string, []>("op_56223_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56223_cast_fp16 = slice_by_index(begin = var_56223_begin_0, end = var_56223_end_0, end_mask = var_56223_end_mask_0, x = q_249_cast_fp16)[name = tensor<string, []>("op_56223_cast_fp16")];
            tensor<int32, [4]> var_56227_begin_0 = const()[name = tensor<string, []>("op_56227_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_56227_end_0 = const()[name = tensor<string, []>("op_56227_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_56227_end_mask_0 = const()[name = tensor<string, []>("op_56227_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56227_cast_fp16 = slice_by_index(begin = var_56227_begin_0, end = var_56227_end_0, end_mask = var_56227_end_mask_0, x = q_249_cast_fp16)[name = tensor<string, []>("op_56227_cast_fp16")];
            tensor<int32, [4]> var_56231_begin_0 = const()[name = tensor<string, []>("op_56231_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_56231_end_0 = const()[name = tensor<string, []>("op_56231_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_56231_end_mask_0 = const()[name = tensor<string, []>("op_56231_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56231_cast_fp16 = slice_by_index(begin = var_56231_begin_0, end = var_56231_end_0, end_mask = var_56231_end_mask_0, x = q_249_cast_fp16)[name = tensor<string, []>("op_56231_cast_fp16")];
            tensor<int32, [4]> var_56235_begin_0 = const()[name = tensor<string, []>("op_56235_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_56235_end_0 = const()[name = tensor<string, []>("op_56235_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_56235_end_mask_0 = const()[name = tensor<string, []>("op_56235_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56235_cast_fp16 = slice_by_index(begin = var_56235_begin_0, end = var_56235_end_0, end_mask = var_56235_end_mask_0, x = q_249_cast_fp16)[name = tensor<string, []>("op_56235_cast_fp16")];
            tensor<int32, [4]> var_56239_begin_0 = const()[name = tensor<string, []>("op_56239_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_56239_end_0 = const()[name = tensor<string, []>("op_56239_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_56239_end_mask_0 = const()[name = tensor<string, []>("op_56239_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56239_cast_fp16 = slice_by_index(begin = var_56239_begin_0, end = var_56239_end_0, end_mask = var_56239_end_mask_0, x = q_249_cast_fp16)[name = tensor<string, []>("op_56239_cast_fp16")];
            tensor<int32, [4]> var_56243_begin_0 = const()[name = tensor<string, []>("op_56243_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_56243_end_0 = const()[name = tensor<string, []>("op_56243_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_56243_end_mask_0 = const()[name = tensor<string, []>("op_56243_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56243_cast_fp16 = slice_by_index(begin = var_56243_begin_0, end = var_56243_end_0, end_mask = var_56243_end_mask_0, x = q_249_cast_fp16)[name = tensor<string, []>("op_56243_cast_fp16")];
            tensor<int32, [4]> var_56247_begin_0 = const()[name = tensor<string, []>("op_56247_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_56247_end_0 = const()[name = tensor<string, []>("op_56247_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_56247_end_mask_0 = const()[name = tensor<string, []>("op_56247_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56247_cast_fp16 = slice_by_index(begin = var_56247_begin_0, end = var_56247_end_0, end_mask = var_56247_end_mask_0, x = q_249_cast_fp16)[name = tensor<string, []>("op_56247_cast_fp16")];
            tensor<int32, [4]> var_56251_begin_0 = const()[name = tensor<string, []>("op_56251_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_56251_end_0 = const()[name = tensor<string, []>("op_56251_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_56251_end_mask_0 = const()[name = tensor<string, []>("op_56251_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56251_cast_fp16 = slice_by_index(begin = var_56251_begin_0, end = var_56251_end_0, end_mask = var_56251_end_mask_0, x = q_249_cast_fp16)[name = tensor<string, []>("op_56251_cast_fp16")];
            tensor<int32, [4]> var_56255_begin_0 = const()[name = tensor<string, []>("op_56255_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_56255_end_0 = const()[name = tensor<string, []>("op_56255_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_56255_end_mask_0 = const()[name = tensor<string, []>("op_56255_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56255_cast_fp16 = slice_by_index(begin = var_56255_begin_0, end = var_56255_end_0, end_mask = var_56255_end_mask_0, x = q_249_cast_fp16)[name = tensor<string, []>("op_56255_cast_fp16")];
            tensor<int32, [4]> var_56259_begin_0 = const()[name = tensor<string, []>("op_56259_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_56259_end_0 = const()[name = tensor<string, []>("op_56259_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_56259_end_mask_0 = const()[name = tensor<string, []>("op_56259_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56259_cast_fp16 = slice_by_index(begin = var_56259_begin_0, end = var_56259_end_0, end_mask = var_56259_end_mask_0, x = q_249_cast_fp16)[name = tensor<string, []>("op_56259_cast_fp16")];
            tensor<int32, [4]> var_56263_begin_0 = const()[name = tensor<string, []>("op_56263_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_56263_end_0 = const()[name = tensor<string, []>("op_56263_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_56263_end_mask_0 = const()[name = tensor<string, []>("op_56263_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56263_cast_fp16 = slice_by_index(begin = var_56263_begin_0, end = var_56263_end_0, end_mask = var_56263_end_mask_0, x = q_249_cast_fp16)[name = tensor<string, []>("op_56263_cast_fp16")];
            tensor<int32, [4]> var_56267_begin_0 = const()[name = tensor<string, []>("op_56267_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_56267_end_0 = const()[name = tensor<string, []>("op_56267_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_56267_end_mask_0 = const()[name = tensor<string, []>("op_56267_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56267_cast_fp16 = slice_by_index(begin = var_56267_begin_0, end = var_56267_end_0, end_mask = var_56267_end_mask_0, x = q_249_cast_fp16)[name = tensor<string, []>("op_56267_cast_fp16")];
            tensor<int32, [4]> var_56271_begin_0 = const()[name = tensor<string, []>("op_56271_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_56271_end_0 = const()[name = tensor<string, []>("op_56271_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_56271_end_mask_0 = const()[name = tensor<string, []>("op_56271_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56271_cast_fp16 = slice_by_index(begin = var_56271_begin_0, end = var_56271_end_0, end_mask = var_56271_end_mask_0, x = q_249_cast_fp16)[name = tensor<string, []>("op_56271_cast_fp16")];
            tensor<int32, [4]> var_56275_begin_0 = const()[name = tensor<string, []>("op_56275_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_56275_end_0 = const()[name = tensor<string, []>("op_56275_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_56275_end_mask_0 = const()[name = tensor<string, []>("op_56275_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56275_cast_fp16 = slice_by_index(begin = var_56275_begin_0, end = var_56275_end_0, end_mask = var_56275_end_mask_0, x = q_249_cast_fp16)[name = tensor<string, []>("op_56275_cast_fp16")];
            tensor<int32, [4]> var_56279_begin_0 = const()[name = tensor<string, []>("op_56279_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_56279_end_0 = const()[name = tensor<string, []>("op_56279_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_56279_end_mask_0 = const()[name = tensor<string, []>("op_56279_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56279_cast_fp16 = slice_by_index(begin = var_56279_begin_0, end = var_56279_end_0, end_mask = var_56279_end_mask_0, x = q_249_cast_fp16)[name = tensor<string, []>("op_56279_cast_fp16")];
            tensor<int32, [4]> var_56283_begin_0 = const()[name = tensor<string, []>("op_56283_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_56283_end_0 = const()[name = tensor<string, []>("op_56283_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_56283_end_mask_0 = const()[name = tensor<string, []>("op_56283_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56283_cast_fp16 = slice_by_index(begin = var_56283_begin_0, end = var_56283_end_0, end_mask = var_56283_end_mask_0, x = q_249_cast_fp16)[name = tensor<string, []>("op_56283_cast_fp16")];
            tensor<int32, [4]> var_56287_begin_0 = const()[name = tensor<string, []>("op_56287_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_56287_end_0 = const()[name = tensor<string, []>("op_56287_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_56287_end_mask_0 = const()[name = tensor<string, []>("op_56287_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56287_cast_fp16 = slice_by_index(begin = var_56287_begin_0, end = var_56287_end_0, end_mask = var_56287_end_mask_0, x = q_249_cast_fp16)[name = tensor<string, []>("op_56287_cast_fp16")];
            tensor<int32, [4]> var_56291_begin_0 = const()[name = tensor<string, []>("op_56291_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_56291_end_0 = const()[name = tensor<string, []>("op_56291_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_56291_end_mask_0 = const()[name = tensor<string, []>("op_56291_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56291_cast_fp16 = slice_by_index(begin = var_56291_begin_0, end = var_56291_end_0, end_mask = var_56291_end_mask_0, x = q_249_cast_fp16)[name = tensor<string, []>("op_56291_cast_fp16")];
            tensor<int32, [4]> k_499_perm_0 = const()[name = tensor<string, []>("k_499_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_56298_begin_0 = const()[name = tensor<string, []>("op_56298_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_56298_end_0 = const()[name = tensor<string, []>("op_56298_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_56298_end_mask_0 = const()[name = tensor<string, []>("op_56298_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> k_499_cast_fp16 = transpose(perm = k_499_perm_0, x = k_497_cast_fp16)[name = tensor<string, []>("transpose_15")];
            tensor<fp16, [2, 1024, 1, 64]> var_56298_cast_fp16 = slice_by_index(begin = var_56298_begin_0, end = var_56298_end_0, end_mask = var_56298_end_mask_0, x = k_499_cast_fp16)[name = tensor<string, []>("op_56298_cast_fp16")];
            tensor<int32, [4]> var_56302_begin_0 = const()[name = tensor<string, []>("op_56302_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_56302_end_0 = const()[name = tensor<string, []>("op_56302_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_56302_end_mask_0 = const()[name = tensor<string, []>("op_56302_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_56302_cast_fp16 = slice_by_index(begin = var_56302_begin_0, end = var_56302_end_0, end_mask = var_56302_end_mask_0, x = k_499_cast_fp16)[name = tensor<string, []>("op_56302_cast_fp16")];
            tensor<int32, [4]> var_56306_begin_0 = const()[name = tensor<string, []>("op_56306_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_56306_end_0 = const()[name = tensor<string, []>("op_56306_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_56306_end_mask_0 = const()[name = tensor<string, []>("op_56306_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_56306_cast_fp16 = slice_by_index(begin = var_56306_begin_0, end = var_56306_end_0, end_mask = var_56306_end_mask_0, x = k_499_cast_fp16)[name = tensor<string, []>("op_56306_cast_fp16")];
            tensor<int32, [4]> var_56310_begin_0 = const()[name = tensor<string, []>("op_56310_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_56310_end_0 = const()[name = tensor<string, []>("op_56310_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_56310_end_mask_0 = const()[name = tensor<string, []>("op_56310_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_56310_cast_fp16 = slice_by_index(begin = var_56310_begin_0, end = var_56310_end_0, end_mask = var_56310_end_mask_0, x = k_499_cast_fp16)[name = tensor<string, []>("op_56310_cast_fp16")];
            tensor<int32, [4]> var_56314_begin_0 = const()[name = tensor<string, []>("op_56314_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_56314_end_0 = const()[name = tensor<string, []>("op_56314_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_56314_end_mask_0 = const()[name = tensor<string, []>("op_56314_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_56314_cast_fp16 = slice_by_index(begin = var_56314_begin_0, end = var_56314_end_0, end_mask = var_56314_end_mask_0, x = k_499_cast_fp16)[name = tensor<string, []>("op_56314_cast_fp16")];
            tensor<int32, [4]> var_56318_begin_0 = const()[name = tensor<string, []>("op_56318_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_56318_end_0 = const()[name = tensor<string, []>("op_56318_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_56318_end_mask_0 = const()[name = tensor<string, []>("op_56318_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_56318_cast_fp16 = slice_by_index(begin = var_56318_begin_0, end = var_56318_end_0, end_mask = var_56318_end_mask_0, x = k_499_cast_fp16)[name = tensor<string, []>("op_56318_cast_fp16")];
            tensor<int32, [4]> var_56322_begin_0 = const()[name = tensor<string, []>("op_56322_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_56322_end_0 = const()[name = tensor<string, []>("op_56322_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_56322_end_mask_0 = const()[name = tensor<string, []>("op_56322_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_56322_cast_fp16 = slice_by_index(begin = var_56322_begin_0, end = var_56322_end_0, end_mask = var_56322_end_mask_0, x = k_499_cast_fp16)[name = tensor<string, []>("op_56322_cast_fp16")];
            tensor<int32, [4]> var_56326_begin_0 = const()[name = tensor<string, []>("op_56326_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_56326_end_0 = const()[name = tensor<string, []>("op_56326_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_56326_end_mask_0 = const()[name = tensor<string, []>("op_56326_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_56326_cast_fp16 = slice_by_index(begin = var_56326_begin_0, end = var_56326_end_0, end_mask = var_56326_end_mask_0, x = k_499_cast_fp16)[name = tensor<string, []>("op_56326_cast_fp16")];
            tensor<int32, [4]> var_56330_begin_0 = const()[name = tensor<string, []>("op_56330_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_56330_end_0 = const()[name = tensor<string, []>("op_56330_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_56330_end_mask_0 = const()[name = tensor<string, []>("op_56330_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_56330_cast_fp16 = slice_by_index(begin = var_56330_begin_0, end = var_56330_end_0, end_mask = var_56330_end_mask_0, x = k_499_cast_fp16)[name = tensor<string, []>("op_56330_cast_fp16")];
            tensor<int32, [4]> var_56334_begin_0 = const()[name = tensor<string, []>("op_56334_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_56334_end_0 = const()[name = tensor<string, []>("op_56334_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_56334_end_mask_0 = const()[name = tensor<string, []>("op_56334_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_56334_cast_fp16 = slice_by_index(begin = var_56334_begin_0, end = var_56334_end_0, end_mask = var_56334_end_mask_0, x = k_499_cast_fp16)[name = tensor<string, []>("op_56334_cast_fp16")];
            tensor<int32, [4]> var_56338_begin_0 = const()[name = tensor<string, []>("op_56338_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_56338_end_0 = const()[name = tensor<string, []>("op_56338_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_56338_end_mask_0 = const()[name = tensor<string, []>("op_56338_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_56338_cast_fp16 = slice_by_index(begin = var_56338_begin_0, end = var_56338_end_0, end_mask = var_56338_end_mask_0, x = k_499_cast_fp16)[name = tensor<string, []>("op_56338_cast_fp16")];
            tensor<int32, [4]> var_56342_begin_0 = const()[name = tensor<string, []>("op_56342_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_56342_end_0 = const()[name = tensor<string, []>("op_56342_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_56342_end_mask_0 = const()[name = tensor<string, []>("op_56342_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_56342_cast_fp16 = slice_by_index(begin = var_56342_begin_0, end = var_56342_end_0, end_mask = var_56342_end_mask_0, x = k_499_cast_fp16)[name = tensor<string, []>("op_56342_cast_fp16")];
            tensor<int32, [4]> var_56346_begin_0 = const()[name = tensor<string, []>("op_56346_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_56346_end_0 = const()[name = tensor<string, []>("op_56346_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_56346_end_mask_0 = const()[name = tensor<string, []>("op_56346_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_56346_cast_fp16 = slice_by_index(begin = var_56346_begin_0, end = var_56346_end_0, end_mask = var_56346_end_mask_0, x = k_499_cast_fp16)[name = tensor<string, []>("op_56346_cast_fp16")];
            tensor<int32, [4]> var_56350_begin_0 = const()[name = tensor<string, []>("op_56350_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_56350_end_0 = const()[name = tensor<string, []>("op_56350_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_56350_end_mask_0 = const()[name = tensor<string, []>("op_56350_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_56350_cast_fp16 = slice_by_index(begin = var_56350_begin_0, end = var_56350_end_0, end_mask = var_56350_end_mask_0, x = k_499_cast_fp16)[name = tensor<string, []>("op_56350_cast_fp16")];
            tensor<int32, [4]> var_56354_begin_0 = const()[name = tensor<string, []>("op_56354_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_56354_end_0 = const()[name = tensor<string, []>("op_56354_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_56354_end_mask_0 = const()[name = tensor<string, []>("op_56354_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_56354_cast_fp16 = slice_by_index(begin = var_56354_begin_0, end = var_56354_end_0, end_mask = var_56354_end_mask_0, x = k_499_cast_fp16)[name = tensor<string, []>("op_56354_cast_fp16")];
            tensor<int32, [4]> var_56358_begin_0 = const()[name = tensor<string, []>("op_56358_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_56358_end_0 = const()[name = tensor<string, []>("op_56358_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_56358_end_mask_0 = const()[name = tensor<string, []>("op_56358_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_56358_cast_fp16 = slice_by_index(begin = var_56358_begin_0, end = var_56358_end_0, end_mask = var_56358_end_mask_0, x = k_499_cast_fp16)[name = tensor<string, []>("op_56358_cast_fp16")];
            tensor<int32, [4]> var_56362_begin_0 = const()[name = tensor<string, []>("op_56362_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_56362_end_0 = const()[name = tensor<string, []>("op_56362_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_56362_end_mask_0 = const()[name = tensor<string, []>("op_56362_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_56362_cast_fp16 = slice_by_index(begin = var_56362_begin_0, end = var_56362_end_0, end_mask = var_56362_end_mask_0, x = k_499_cast_fp16)[name = tensor<string, []>("op_56362_cast_fp16")];
            tensor<int32, [4]> var_56366_begin_0 = const()[name = tensor<string, []>("op_56366_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_56366_end_0 = const()[name = tensor<string, []>("op_56366_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_56366_end_mask_0 = const()[name = tensor<string, []>("op_56366_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_56366_cast_fp16 = slice_by_index(begin = var_56366_begin_0, end = var_56366_end_0, end_mask = var_56366_end_mask_0, x = k_499_cast_fp16)[name = tensor<string, []>("op_56366_cast_fp16")];
            tensor<int32, [4]> var_56370_begin_0 = const()[name = tensor<string, []>("op_56370_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_56370_end_0 = const()[name = tensor<string, []>("op_56370_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_56370_end_mask_0 = const()[name = tensor<string, []>("op_56370_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_56370_cast_fp16 = slice_by_index(begin = var_56370_begin_0, end = var_56370_end_0, end_mask = var_56370_end_mask_0, x = k_499_cast_fp16)[name = tensor<string, []>("op_56370_cast_fp16")];
            tensor<int32, [4]> var_56374_begin_0 = const()[name = tensor<string, []>("op_56374_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_56374_end_0 = const()[name = tensor<string, []>("op_56374_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_56374_end_mask_0 = const()[name = tensor<string, []>("op_56374_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_56374_cast_fp16 = slice_by_index(begin = var_56374_begin_0, end = var_56374_end_0, end_mask = var_56374_end_mask_0, x = k_499_cast_fp16)[name = tensor<string, []>("op_56374_cast_fp16")];
            tensor<int32, [4]> var_56376_begin_0 = const()[name = tensor<string, []>("op_56376_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_56376_end_0 = const()[name = tensor<string, []>("op_56376_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_56376_end_mask_0 = const()[name = tensor<string, []>("op_56376_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56376_cast_fp16 = slice_by_index(begin = var_56376_begin_0, end = var_56376_end_0, end_mask = var_56376_end_mask_0, x = v_249_cast_fp16)[name = tensor<string, []>("op_56376_cast_fp16")];
            tensor<int32, [4]> var_56380_begin_0 = const()[name = tensor<string, []>("op_56380_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_56380_end_0 = const()[name = tensor<string, []>("op_56380_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_56380_end_mask_0 = const()[name = tensor<string, []>("op_56380_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56380_cast_fp16 = slice_by_index(begin = var_56380_begin_0, end = var_56380_end_0, end_mask = var_56380_end_mask_0, x = v_249_cast_fp16)[name = tensor<string, []>("op_56380_cast_fp16")];
            tensor<int32, [4]> var_56384_begin_0 = const()[name = tensor<string, []>("op_56384_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_56384_end_0 = const()[name = tensor<string, []>("op_56384_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_56384_end_mask_0 = const()[name = tensor<string, []>("op_56384_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56384_cast_fp16 = slice_by_index(begin = var_56384_begin_0, end = var_56384_end_0, end_mask = var_56384_end_mask_0, x = v_249_cast_fp16)[name = tensor<string, []>("op_56384_cast_fp16")];
            tensor<int32, [4]> var_56388_begin_0 = const()[name = tensor<string, []>("op_56388_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_56388_end_0 = const()[name = tensor<string, []>("op_56388_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_56388_end_mask_0 = const()[name = tensor<string, []>("op_56388_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56388_cast_fp16 = slice_by_index(begin = var_56388_begin_0, end = var_56388_end_0, end_mask = var_56388_end_mask_0, x = v_249_cast_fp16)[name = tensor<string, []>("op_56388_cast_fp16")];
            tensor<int32, [4]> var_56392_begin_0 = const()[name = tensor<string, []>("op_56392_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_56392_end_0 = const()[name = tensor<string, []>("op_56392_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_56392_end_mask_0 = const()[name = tensor<string, []>("op_56392_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56392_cast_fp16 = slice_by_index(begin = var_56392_begin_0, end = var_56392_end_0, end_mask = var_56392_end_mask_0, x = v_249_cast_fp16)[name = tensor<string, []>("op_56392_cast_fp16")];
            tensor<int32, [4]> var_56396_begin_0 = const()[name = tensor<string, []>("op_56396_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_56396_end_0 = const()[name = tensor<string, []>("op_56396_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_56396_end_mask_0 = const()[name = tensor<string, []>("op_56396_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56396_cast_fp16 = slice_by_index(begin = var_56396_begin_0, end = var_56396_end_0, end_mask = var_56396_end_mask_0, x = v_249_cast_fp16)[name = tensor<string, []>("op_56396_cast_fp16")];
            tensor<int32, [4]> var_56400_begin_0 = const()[name = tensor<string, []>("op_56400_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_56400_end_0 = const()[name = tensor<string, []>("op_56400_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_56400_end_mask_0 = const()[name = tensor<string, []>("op_56400_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56400_cast_fp16 = slice_by_index(begin = var_56400_begin_0, end = var_56400_end_0, end_mask = var_56400_end_mask_0, x = v_249_cast_fp16)[name = tensor<string, []>("op_56400_cast_fp16")];
            tensor<int32, [4]> var_56404_begin_0 = const()[name = tensor<string, []>("op_56404_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_56404_end_0 = const()[name = tensor<string, []>("op_56404_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_56404_end_mask_0 = const()[name = tensor<string, []>("op_56404_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56404_cast_fp16 = slice_by_index(begin = var_56404_begin_0, end = var_56404_end_0, end_mask = var_56404_end_mask_0, x = v_249_cast_fp16)[name = tensor<string, []>("op_56404_cast_fp16")];
            tensor<int32, [4]> var_56408_begin_0 = const()[name = tensor<string, []>("op_56408_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_56408_end_0 = const()[name = tensor<string, []>("op_56408_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_56408_end_mask_0 = const()[name = tensor<string, []>("op_56408_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56408_cast_fp16 = slice_by_index(begin = var_56408_begin_0, end = var_56408_end_0, end_mask = var_56408_end_mask_0, x = v_249_cast_fp16)[name = tensor<string, []>("op_56408_cast_fp16")];
            tensor<int32, [4]> var_56412_begin_0 = const()[name = tensor<string, []>("op_56412_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_56412_end_0 = const()[name = tensor<string, []>("op_56412_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_56412_end_mask_0 = const()[name = tensor<string, []>("op_56412_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56412_cast_fp16 = slice_by_index(begin = var_56412_begin_0, end = var_56412_end_0, end_mask = var_56412_end_mask_0, x = v_249_cast_fp16)[name = tensor<string, []>("op_56412_cast_fp16")];
            tensor<int32, [4]> var_56416_begin_0 = const()[name = tensor<string, []>("op_56416_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_56416_end_0 = const()[name = tensor<string, []>("op_56416_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_56416_end_mask_0 = const()[name = tensor<string, []>("op_56416_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56416_cast_fp16 = slice_by_index(begin = var_56416_begin_0, end = var_56416_end_0, end_mask = var_56416_end_mask_0, x = v_249_cast_fp16)[name = tensor<string, []>("op_56416_cast_fp16")];
            tensor<int32, [4]> var_56420_begin_0 = const()[name = tensor<string, []>("op_56420_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_56420_end_0 = const()[name = tensor<string, []>("op_56420_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_56420_end_mask_0 = const()[name = tensor<string, []>("op_56420_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56420_cast_fp16 = slice_by_index(begin = var_56420_begin_0, end = var_56420_end_0, end_mask = var_56420_end_mask_0, x = v_249_cast_fp16)[name = tensor<string, []>("op_56420_cast_fp16")];
            tensor<int32, [4]> var_56424_begin_0 = const()[name = tensor<string, []>("op_56424_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_56424_end_0 = const()[name = tensor<string, []>("op_56424_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_56424_end_mask_0 = const()[name = tensor<string, []>("op_56424_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56424_cast_fp16 = slice_by_index(begin = var_56424_begin_0, end = var_56424_end_0, end_mask = var_56424_end_mask_0, x = v_249_cast_fp16)[name = tensor<string, []>("op_56424_cast_fp16")];
            tensor<int32, [4]> var_56428_begin_0 = const()[name = tensor<string, []>("op_56428_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_56428_end_0 = const()[name = tensor<string, []>("op_56428_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_56428_end_mask_0 = const()[name = tensor<string, []>("op_56428_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56428_cast_fp16 = slice_by_index(begin = var_56428_begin_0, end = var_56428_end_0, end_mask = var_56428_end_mask_0, x = v_249_cast_fp16)[name = tensor<string, []>("op_56428_cast_fp16")];
            tensor<int32, [4]> var_56432_begin_0 = const()[name = tensor<string, []>("op_56432_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_56432_end_0 = const()[name = tensor<string, []>("op_56432_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_56432_end_mask_0 = const()[name = tensor<string, []>("op_56432_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56432_cast_fp16 = slice_by_index(begin = var_56432_begin_0, end = var_56432_end_0, end_mask = var_56432_end_mask_0, x = v_249_cast_fp16)[name = tensor<string, []>("op_56432_cast_fp16")];
            tensor<int32, [4]> var_56436_begin_0 = const()[name = tensor<string, []>("op_56436_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_56436_end_0 = const()[name = tensor<string, []>("op_56436_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_56436_end_mask_0 = const()[name = tensor<string, []>("op_56436_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56436_cast_fp16 = slice_by_index(begin = var_56436_begin_0, end = var_56436_end_0, end_mask = var_56436_end_mask_0, x = v_249_cast_fp16)[name = tensor<string, []>("op_56436_cast_fp16")];
            tensor<int32, [4]> var_56440_begin_0 = const()[name = tensor<string, []>("op_56440_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_56440_end_0 = const()[name = tensor<string, []>("op_56440_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_56440_end_mask_0 = const()[name = tensor<string, []>("op_56440_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56440_cast_fp16 = slice_by_index(begin = var_56440_begin_0, end = var_56440_end_0, end_mask = var_56440_end_mask_0, x = v_249_cast_fp16)[name = tensor<string, []>("op_56440_cast_fp16")];
            tensor<int32, [4]> var_56444_begin_0 = const()[name = tensor<string, []>("op_56444_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_56444_end_0 = const()[name = tensor<string, []>("op_56444_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_56444_end_mask_0 = const()[name = tensor<string, []>("op_56444_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56444_cast_fp16 = slice_by_index(begin = var_56444_begin_0, end = var_56444_end_0, end_mask = var_56444_end_mask_0, x = v_249_cast_fp16)[name = tensor<string, []>("op_56444_cast_fp16")];
            tensor<int32, [4]> var_56448_begin_0 = const()[name = tensor<string, []>("op_56448_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_56448_end_0 = const()[name = tensor<string, []>("op_56448_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_56448_end_mask_0 = const()[name = tensor<string, []>("op_56448_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56448_cast_fp16 = slice_by_index(begin = var_56448_begin_0, end = var_56448_end_0, end_mask = var_56448_end_mask_0, x = v_249_cast_fp16)[name = tensor<string, []>("op_56448_cast_fp16")];
            tensor<int32, [4]> var_56452_begin_0 = const()[name = tensor<string, []>("op_56452_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_56452_end_0 = const()[name = tensor<string, []>("op_56452_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_56452_end_mask_0 = const()[name = tensor<string, []>("op_56452_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56452_cast_fp16 = slice_by_index(begin = var_56452_begin_0, end = var_56452_end_0, end_mask = var_56452_end_mask_0, x = v_249_cast_fp16)[name = tensor<string, []>("op_56452_cast_fp16")];
            tensor<string, []> var_56456_equation_0 = const()[name = tensor<string, []>("op_56456_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56456_cast_fp16 = einsum(equation = var_56456_equation_0, values = (var_56298_cast_fp16, var_56215_cast_fp16))[name = tensor<string, []>("op_56456_cast_fp16")];
            tensor<fp16, []> var_56457_to_fp16 = const()[name = tensor<string, []>("op_56457_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4801_cast_fp16 = mul(x = var_56456_cast_fp16, y = var_56457_to_fp16)[name = tensor<string, []>("aw_4801_cast_fp16")];
            tensor<string, []> var_56460_equation_0 = const()[name = tensor<string, []>("op_56460_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56460_cast_fp16 = einsum(equation = var_56460_equation_0, values = (var_56302_cast_fp16, var_56219_cast_fp16))[name = tensor<string, []>("op_56460_cast_fp16")];
            tensor<fp16, []> var_56461_to_fp16 = const()[name = tensor<string, []>("op_56461_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4803_cast_fp16 = mul(x = var_56460_cast_fp16, y = var_56461_to_fp16)[name = tensor<string, []>("aw_4803_cast_fp16")];
            tensor<string, []> var_56464_equation_0 = const()[name = tensor<string, []>("op_56464_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56464_cast_fp16 = einsum(equation = var_56464_equation_0, values = (var_56306_cast_fp16, var_56223_cast_fp16))[name = tensor<string, []>("op_56464_cast_fp16")];
            tensor<fp16, []> var_56465_to_fp16 = const()[name = tensor<string, []>("op_56465_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4805_cast_fp16 = mul(x = var_56464_cast_fp16, y = var_56465_to_fp16)[name = tensor<string, []>("aw_4805_cast_fp16")];
            tensor<string, []> var_56468_equation_0 = const()[name = tensor<string, []>("op_56468_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56468_cast_fp16 = einsum(equation = var_56468_equation_0, values = (var_56310_cast_fp16, var_56227_cast_fp16))[name = tensor<string, []>("op_56468_cast_fp16")];
            tensor<fp16, []> var_56469_to_fp16 = const()[name = tensor<string, []>("op_56469_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4807_cast_fp16 = mul(x = var_56468_cast_fp16, y = var_56469_to_fp16)[name = tensor<string, []>("aw_4807_cast_fp16")];
            tensor<string, []> var_56472_equation_0 = const()[name = tensor<string, []>("op_56472_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56472_cast_fp16 = einsum(equation = var_56472_equation_0, values = (var_56314_cast_fp16, var_56231_cast_fp16))[name = tensor<string, []>("op_56472_cast_fp16")];
            tensor<fp16, []> var_56473_to_fp16 = const()[name = tensor<string, []>("op_56473_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4809_cast_fp16 = mul(x = var_56472_cast_fp16, y = var_56473_to_fp16)[name = tensor<string, []>("aw_4809_cast_fp16")];
            tensor<string, []> var_56476_equation_0 = const()[name = tensor<string, []>("op_56476_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56476_cast_fp16 = einsum(equation = var_56476_equation_0, values = (var_56318_cast_fp16, var_56235_cast_fp16))[name = tensor<string, []>("op_56476_cast_fp16")];
            tensor<fp16, []> var_56477_to_fp16 = const()[name = tensor<string, []>("op_56477_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4811_cast_fp16 = mul(x = var_56476_cast_fp16, y = var_56477_to_fp16)[name = tensor<string, []>("aw_4811_cast_fp16")];
            tensor<string, []> var_56480_equation_0 = const()[name = tensor<string, []>("op_56480_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56480_cast_fp16 = einsum(equation = var_56480_equation_0, values = (var_56322_cast_fp16, var_56239_cast_fp16))[name = tensor<string, []>("op_56480_cast_fp16")];
            tensor<fp16, []> var_56481_to_fp16 = const()[name = tensor<string, []>("op_56481_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4813_cast_fp16 = mul(x = var_56480_cast_fp16, y = var_56481_to_fp16)[name = tensor<string, []>("aw_4813_cast_fp16")];
            tensor<string, []> var_56484_equation_0 = const()[name = tensor<string, []>("op_56484_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56484_cast_fp16 = einsum(equation = var_56484_equation_0, values = (var_56326_cast_fp16, var_56243_cast_fp16))[name = tensor<string, []>("op_56484_cast_fp16")];
            tensor<fp16, []> var_56485_to_fp16 = const()[name = tensor<string, []>("op_56485_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4815_cast_fp16 = mul(x = var_56484_cast_fp16, y = var_56485_to_fp16)[name = tensor<string, []>("aw_4815_cast_fp16")];
            tensor<string, []> var_56488_equation_0 = const()[name = tensor<string, []>("op_56488_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56488_cast_fp16 = einsum(equation = var_56488_equation_0, values = (var_56330_cast_fp16, var_56247_cast_fp16))[name = tensor<string, []>("op_56488_cast_fp16")];
            tensor<fp16, []> var_56489_to_fp16 = const()[name = tensor<string, []>("op_56489_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4817_cast_fp16 = mul(x = var_56488_cast_fp16, y = var_56489_to_fp16)[name = tensor<string, []>("aw_4817_cast_fp16")];
            tensor<string, []> var_56492_equation_0 = const()[name = tensor<string, []>("op_56492_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56492_cast_fp16 = einsum(equation = var_56492_equation_0, values = (var_56334_cast_fp16, var_56251_cast_fp16))[name = tensor<string, []>("op_56492_cast_fp16")];
            tensor<fp16, []> var_56493_to_fp16 = const()[name = tensor<string, []>("op_56493_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4819_cast_fp16 = mul(x = var_56492_cast_fp16, y = var_56493_to_fp16)[name = tensor<string, []>("aw_4819_cast_fp16")];
            tensor<string, []> var_56496_equation_0 = const()[name = tensor<string, []>("op_56496_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56496_cast_fp16 = einsum(equation = var_56496_equation_0, values = (var_56338_cast_fp16, var_56255_cast_fp16))[name = tensor<string, []>("op_56496_cast_fp16")];
            tensor<fp16, []> var_56497_to_fp16 = const()[name = tensor<string, []>("op_56497_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4821_cast_fp16 = mul(x = var_56496_cast_fp16, y = var_56497_to_fp16)[name = tensor<string, []>("aw_4821_cast_fp16")];
            tensor<string, []> var_56500_equation_0 = const()[name = tensor<string, []>("op_56500_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56500_cast_fp16 = einsum(equation = var_56500_equation_0, values = (var_56342_cast_fp16, var_56259_cast_fp16))[name = tensor<string, []>("op_56500_cast_fp16")];
            tensor<fp16, []> var_56501_to_fp16 = const()[name = tensor<string, []>("op_56501_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4823_cast_fp16 = mul(x = var_56500_cast_fp16, y = var_56501_to_fp16)[name = tensor<string, []>("aw_4823_cast_fp16")];
            tensor<string, []> var_56504_equation_0 = const()[name = tensor<string, []>("op_56504_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56504_cast_fp16 = einsum(equation = var_56504_equation_0, values = (var_56346_cast_fp16, var_56263_cast_fp16))[name = tensor<string, []>("op_56504_cast_fp16")];
            tensor<fp16, []> var_56505_to_fp16 = const()[name = tensor<string, []>("op_56505_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4825_cast_fp16 = mul(x = var_56504_cast_fp16, y = var_56505_to_fp16)[name = tensor<string, []>("aw_4825_cast_fp16")];
            tensor<string, []> var_56508_equation_0 = const()[name = tensor<string, []>("op_56508_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56508_cast_fp16 = einsum(equation = var_56508_equation_0, values = (var_56350_cast_fp16, var_56267_cast_fp16))[name = tensor<string, []>("op_56508_cast_fp16")];
            tensor<fp16, []> var_56509_to_fp16 = const()[name = tensor<string, []>("op_56509_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4827_cast_fp16 = mul(x = var_56508_cast_fp16, y = var_56509_to_fp16)[name = tensor<string, []>("aw_4827_cast_fp16")];
            tensor<string, []> var_56512_equation_0 = const()[name = tensor<string, []>("op_56512_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56512_cast_fp16 = einsum(equation = var_56512_equation_0, values = (var_56354_cast_fp16, var_56271_cast_fp16))[name = tensor<string, []>("op_56512_cast_fp16")];
            tensor<fp16, []> var_56513_to_fp16 = const()[name = tensor<string, []>("op_56513_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4829_cast_fp16 = mul(x = var_56512_cast_fp16, y = var_56513_to_fp16)[name = tensor<string, []>("aw_4829_cast_fp16")];
            tensor<string, []> var_56516_equation_0 = const()[name = tensor<string, []>("op_56516_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56516_cast_fp16 = einsum(equation = var_56516_equation_0, values = (var_56358_cast_fp16, var_56275_cast_fp16))[name = tensor<string, []>("op_56516_cast_fp16")];
            tensor<fp16, []> var_56517_to_fp16 = const()[name = tensor<string, []>("op_56517_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4831_cast_fp16 = mul(x = var_56516_cast_fp16, y = var_56517_to_fp16)[name = tensor<string, []>("aw_4831_cast_fp16")];
            tensor<string, []> var_56520_equation_0 = const()[name = tensor<string, []>("op_56520_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56520_cast_fp16 = einsum(equation = var_56520_equation_0, values = (var_56362_cast_fp16, var_56279_cast_fp16))[name = tensor<string, []>("op_56520_cast_fp16")];
            tensor<fp16, []> var_56521_to_fp16 = const()[name = tensor<string, []>("op_56521_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4833_cast_fp16 = mul(x = var_56520_cast_fp16, y = var_56521_to_fp16)[name = tensor<string, []>("aw_4833_cast_fp16")];
            tensor<string, []> var_56524_equation_0 = const()[name = tensor<string, []>("op_56524_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56524_cast_fp16 = einsum(equation = var_56524_equation_0, values = (var_56366_cast_fp16, var_56283_cast_fp16))[name = tensor<string, []>("op_56524_cast_fp16")];
            tensor<fp16, []> var_56525_to_fp16 = const()[name = tensor<string, []>("op_56525_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4835_cast_fp16 = mul(x = var_56524_cast_fp16, y = var_56525_to_fp16)[name = tensor<string, []>("aw_4835_cast_fp16")];
            tensor<string, []> var_56528_equation_0 = const()[name = tensor<string, []>("op_56528_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56528_cast_fp16 = einsum(equation = var_56528_equation_0, values = (var_56370_cast_fp16, var_56287_cast_fp16))[name = tensor<string, []>("op_56528_cast_fp16")];
            tensor<fp16, []> var_56529_to_fp16 = const()[name = tensor<string, []>("op_56529_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4837_cast_fp16 = mul(x = var_56528_cast_fp16, y = var_56529_to_fp16)[name = tensor<string, []>("aw_4837_cast_fp16")];
            tensor<string, []> var_56532_equation_0 = const()[name = tensor<string, []>("op_56532_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56532_cast_fp16 = einsum(equation = var_56532_equation_0, values = (var_56374_cast_fp16, var_56291_cast_fp16))[name = tensor<string, []>("op_56532_cast_fp16")];
            tensor<fp16, []> var_56533_to_fp16 = const()[name = tensor<string, []>("op_56533_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4839_cast_fp16 = mul(x = var_56532_cast_fp16, y = var_56533_to_fp16)[name = tensor<string, []>("aw_4839_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56535_cast_fp16 = softmax(axis = var_30355, x = aw_4801_cast_fp16)[name = tensor<string, []>("op_56535_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56536_cast_fp16 = softmax(axis = var_30355, x = aw_4803_cast_fp16)[name = tensor<string, []>("op_56536_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56537_cast_fp16 = softmax(axis = var_30355, x = aw_4805_cast_fp16)[name = tensor<string, []>("op_56537_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56538_cast_fp16 = softmax(axis = var_30355, x = aw_4807_cast_fp16)[name = tensor<string, []>("op_56538_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56539_cast_fp16 = softmax(axis = var_30355, x = aw_4809_cast_fp16)[name = tensor<string, []>("op_56539_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56540_cast_fp16 = softmax(axis = var_30355, x = aw_4811_cast_fp16)[name = tensor<string, []>("op_56540_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56541_cast_fp16 = softmax(axis = var_30355, x = aw_4813_cast_fp16)[name = tensor<string, []>("op_56541_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56542_cast_fp16 = softmax(axis = var_30355, x = aw_4815_cast_fp16)[name = tensor<string, []>("op_56542_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56543_cast_fp16 = softmax(axis = var_30355, x = aw_4817_cast_fp16)[name = tensor<string, []>("op_56543_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56544_cast_fp16 = softmax(axis = var_30355, x = aw_4819_cast_fp16)[name = tensor<string, []>("op_56544_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56545_cast_fp16 = softmax(axis = var_30355, x = aw_4821_cast_fp16)[name = tensor<string, []>("op_56545_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56546_cast_fp16 = softmax(axis = var_30355, x = aw_4823_cast_fp16)[name = tensor<string, []>("op_56546_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56547_cast_fp16 = softmax(axis = var_30355, x = aw_4825_cast_fp16)[name = tensor<string, []>("op_56547_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56548_cast_fp16 = softmax(axis = var_30355, x = aw_4827_cast_fp16)[name = tensor<string, []>("op_56548_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56549_cast_fp16 = softmax(axis = var_30355, x = aw_4829_cast_fp16)[name = tensor<string, []>("op_56549_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56550_cast_fp16 = softmax(axis = var_30355, x = aw_4831_cast_fp16)[name = tensor<string, []>("op_56550_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56551_cast_fp16 = softmax(axis = var_30355, x = aw_4833_cast_fp16)[name = tensor<string, []>("op_56551_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56552_cast_fp16 = softmax(axis = var_30355, x = aw_4835_cast_fp16)[name = tensor<string, []>("op_56552_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56553_cast_fp16 = softmax(axis = var_30355, x = aw_4837_cast_fp16)[name = tensor<string, []>("op_56553_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56554_cast_fp16 = softmax(axis = var_30355, x = aw_4839_cast_fp16)[name = tensor<string, []>("op_56554_cast_fp16")];
            tensor<string, []> var_56556_equation_0 = const()[name = tensor<string, []>("op_56556_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56556_cast_fp16 = einsum(equation = var_56556_equation_0, values = (var_56376_cast_fp16, var_56535_cast_fp16))[name = tensor<string, []>("op_56556_cast_fp16")];
            tensor<string, []> var_56558_equation_0 = const()[name = tensor<string, []>("op_56558_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56558_cast_fp16 = einsum(equation = var_56558_equation_0, values = (var_56380_cast_fp16, var_56536_cast_fp16))[name = tensor<string, []>("op_56558_cast_fp16")];
            tensor<string, []> var_56560_equation_0 = const()[name = tensor<string, []>("op_56560_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56560_cast_fp16 = einsum(equation = var_56560_equation_0, values = (var_56384_cast_fp16, var_56537_cast_fp16))[name = tensor<string, []>("op_56560_cast_fp16")];
            tensor<string, []> var_56562_equation_0 = const()[name = tensor<string, []>("op_56562_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56562_cast_fp16 = einsum(equation = var_56562_equation_0, values = (var_56388_cast_fp16, var_56538_cast_fp16))[name = tensor<string, []>("op_56562_cast_fp16")];
            tensor<string, []> var_56564_equation_0 = const()[name = tensor<string, []>("op_56564_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56564_cast_fp16 = einsum(equation = var_56564_equation_0, values = (var_56392_cast_fp16, var_56539_cast_fp16))[name = tensor<string, []>("op_56564_cast_fp16")];
            tensor<string, []> var_56566_equation_0 = const()[name = tensor<string, []>("op_56566_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56566_cast_fp16 = einsum(equation = var_56566_equation_0, values = (var_56396_cast_fp16, var_56540_cast_fp16))[name = tensor<string, []>("op_56566_cast_fp16")];
            tensor<string, []> var_56568_equation_0 = const()[name = tensor<string, []>("op_56568_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56568_cast_fp16 = einsum(equation = var_56568_equation_0, values = (var_56400_cast_fp16, var_56541_cast_fp16))[name = tensor<string, []>("op_56568_cast_fp16")];
            tensor<string, []> var_56570_equation_0 = const()[name = tensor<string, []>("op_56570_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56570_cast_fp16 = einsum(equation = var_56570_equation_0, values = (var_56404_cast_fp16, var_56542_cast_fp16))[name = tensor<string, []>("op_56570_cast_fp16")];
            tensor<string, []> var_56572_equation_0 = const()[name = tensor<string, []>("op_56572_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56572_cast_fp16 = einsum(equation = var_56572_equation_0, values = (var_56408_cast_fp16, var_56543_cast_fp16))[name = tensor<string, []>("op_56572_cast_fp16")];
            tensor<string, []> var_56574_equation_0 = const()[name = tensor<string, []>("op_56574_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56574_cast_fp16 = einsum(equation = var_56574_equation_0, values = (var_56412_cast_fp16, var_56544_cast_fp16))[name = tensor<string, []>("op_56574_cast_fp16")];
            tensor<string, []> var_56576_equation_0 = const()[name = tensor<string, []>("op_56576_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56576_cast_fp16 = einsum(equation = var_56576_equation_0, values = (var_56416_cast_fp16, var_56545_cast_fp16))[name = tensor<string, []>("op_56576_cast_fp16")];
            tensor<string, []> var_56578_equation_0 = const()[name = tensor<string, []>("op_56578_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56578_cast_fp16 = einsum(equation = var_56578_equation_0, values = (var_56420_cast_fp16, var_56546_cast_fp16))[name = tensor<string, []>("op_56578_cast_fp16")];
            tensor<string, []> var_56580_equation_0 = const()[name = tensor<string, []>("op_56580_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56580_cast_fp16 = einsum(equation = var_56580_equation_0, values = (var_56424_cast_fp16, var_56547_cast_fp16))[name = tensor<string, []>("op_56580_cast_fp16")];
            tensor<string, []> var_56582_equation_0 = const()[name = tensor<string, []>("op_56582_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56582_cast_fp16 = einsum(equation = var_56582_equation_0, values = (var_56428_cast_fp16, var_56548_cast_fp16))[name = tensor<string, []>("op_56582_cast_fp16")];
            tensor<string, []> var_56584_equation_0 = const()[name = tensor<string, []>("op_56584_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56584_cast_fp16 = einsum(equation = var_56584_equation_0, values = (var_56432_cast_fp16, var_56549_cast_fp16))[name = tensor<string, []>("op_56584_cast_fp16")];
            tensor<string, []> var_56586_equation_0 = const()[name = tensor<string, []>("op_56586_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56586_cast_fp16 = einsum(equation = var_56586_equation_0, values = (var_56436_cast_fp16, var_56550_cast_fp16))[name = tensor<string, []>("op_56586_cast_fp16")];
            tensor<string, []> var_56588_equation_0 = const()[name = tensor<string, []>("op_56588_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56588_cast_fp16 = einsum(equation = var_56588_equation_0, values = (var_56440_cast_fp16, var_56551_cast_fp16))[name = tensor<string, []>("op_56588_cast_fp16")];
            tensor<string, []> var_56590_equation_0 = const()[name = tensor<string, []>("op_56590_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56590_cast_fp16 = einsum(equation = var_56590_equation_0, values = (var_56444_cast_fp16, var_56552_cast_fp16))[name = tensor<string, []>("op_56590_cast_fp16")];
            tensor<string, []> var_56592_equation_0 = const()[name = tensor<string, []>("op_56592_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56592_cast_fp16 = einsum(equation = var_56592_equation_0, values = (var_56448_cast_fp16, var_56553_cast_fp16))[name = tensor<string, []>("op_56592_cast_fp16")];
            tensor<string, []> var_56594_equation_0 = const()[name = tensor<string, []>("op_56594_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56594_cast_fp16 = einsum(equation = var_56594_equation_0, values = (var_56452_cast_fp16, var_56554_cast_fp16))[name = tensor<string, []>("op_56594_cast_fp16")];
            tensor<bool, []> input_699_interleave_0 = const()[name = tensor<string, []>("input_699_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_699_cast_fp16 = concat(axis = var_30355, interleave = input_699_interleave_0, values = (var_56556_cast_fp16, var_56558_cast_fp16, var_56560_cast_fp16, var_56562_cast_fp16, var_56564_cast_fp16, var_56566_cast_fp16, var_56568_cast_fp16, var_56570_cast_fp16, var_56572_cast_fp16, var_56574_cast_fp16, var_56576_cast_fp16, var_56578_cast_fp16, var_56580_cast_fp16, var_56582_cast_fp16, var_56584_cast_fp16, var_56586_cast_fp16, var_56588_cast_fp16, var_56590_cast_fp16, var_56592_cast_fp16, var_56594_cast_fp16))[name = tensor<string, []>("input_699_cast_fp16")];
            tensor<string, []> var_56604_pad_type_0 = const()[name = tensor<string, []>("op_56604_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_56604_strides_0 = const()[name = tensor<string, []>("op_56604_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_56604_pad_0 = const()[name = tensor<string, []>("op_56604_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_56604_dilations_0 = const()[name = tensor<string, []>("op_56604_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_56604_groups_0 = const()[name = tensor<string, []>("op_56604_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_8_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(810714176))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(811943040))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_8_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_8_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_8_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(811943232)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_56604_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_8_attn1_to_out_0_bias_to_fp16, dilations = var_56604_dilations_0, groups = var_56604_groups_0, pad = var_56604_pad_0, pad_type = var_56604_pad_type_0, strides = var_56604_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_8_attn1_to_out_0_weight_to_fp16_palettized, x = input_699_cast_fp16)[name = tensor<string, []>("op_56604_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_375_cast_fp16 = add(x = var_56604_cast_fp16, y = inputs_373_cast_fp16)[name = tensor<string, []>("inputs_375_cast_fp16")];
            tensor<int32, [1]> hidden_states_485_axes_0 = const()[name = tensor<string, []>("hidden_states_485_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_485_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_485_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(811945856)))];
            tensor<fp16, [1280]> hidden_states_485_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_485_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(811948480)))];
            tensor<fp16, []> var_56614_to_fp16 = const()[name = tensor<string, []>("op_56614_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_485_cast_fp16 = layer_norm(axes = hidden_states_485_axes_0, beta = hidden_states_485_beta_0_to_fp16, epsilon = var_56614_to_fp16, gamma = hidden_states_485_gamma_0_to_fp16, x = inputs_375_cast_fp16)[name = tensor<string, []>("hidden_states_485_cast_fp16")];
            tensor<string, []> q_251_pad_type_0 = const()[name = tensor<string, []>("q_251_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_251_strides_0 = const()[name = tensor<string, []>("q_251_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_251_pad_0 = const()[name = tensor<string, []>("q_251_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_251_dilations_0 = const()[name = tensor<string, []>("q_251_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_251_groups_0 = const()[name = tensor<string, []>("q_251_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_8_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(811951104))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(813179968))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_8_attn2_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_251_cast_fp16 = conv(dilations = q_251_dilations_0, groups = q_251_groups_0, pad = q_251_pad_0, pad_type = q_251_pad_type_0, strides = q_251_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_8_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_485_cast_fp16)[name = tensor<string, []>("q_251_cast_fp16")];
            tensor<string, []> k_501_pad_type_0 = const()[name = tensor<string, []>("k_501_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_501_strides_0 = const()[name = tensor<string, []>("k_501_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_501_pad_0 = const()[name = tensor<string, []>("k_501_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_501_dilations_0 = const()[name = tensor<string, []>("k_501_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_501_groups_0 = const()[name = tensor<string, []>("k_501_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_8_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(813180160))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(815146304))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_8_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> k_501_cast_fp16 = conv(dilations = k_501_dilations_0, groups = k_501_groups_0, pad = k_501_pad_0, pad_type = k_501_pad_type_0, strides = k_501_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_8_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("k_501_cast_fp16")];
            tensor<string, []> v_251_pad_type_0 = const()[name = tensor<string, []>("v_251_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_251_strides_0 = const()[name = tensor<string, []>("v_251_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_251_pad_0 = const()[name = tensor<string, []>("v_251_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_251_dilations_0 = const()[name = tensor<string, []>("v_251_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_251_groups_0 = const()[name = tensor<string, []>("v_251_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_8_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(815146496))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(817112640))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_8_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> v_251_cast_fp16 = conv(dilations = v_251_dilations_0, groups = v_251_groups_0, pad = v_251_pad_0, pad_type = v_251_pad_type_0, strides = v_251_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_8_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("v_251_cast_fp16")];
            tensor<int32, [4]> var_56647_begin_0 = const()[name = tensor<string, []>("op_56647_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_56647_end_0 = const()[name = tensor<string, []>("op_56647_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_56647_end_mask_0 = const()[name = tensor<string, []>("op_56647_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56647_cast_fp16 = slice_by_index(begin = var_56647_begin_0, end = var_56647_end_0, end_mask = var_56647_end_mask_0, x = q_251_cast_fp16)[name = tensor<string, []>("op_56647_cast_fp16")];
            tensor<int32, [4]> var_56651_begin_0 = const()[name = tensor<string, []>("op_56651_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_56651_end_0 = const()[name = tensor<string, []>("op_56651_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_56651_end_mask_0 = const()[name = tensor<string, []>("op_56651_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56651_cast_fp16 = slice_by_index(begin = var_56651_begin_0, end = var_56651_end_0, end_mask = var_56651_end_mask_0, x = q_251_cast_fp16)[name = tensor<string, []>("op_56651_cast_fp16")];
            tensor<int32, [4]> var_56655_begin_0 = const()[name = tensor<string, []>("op_56655_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_56655_end_0 = const()[name = tensor<string, []>("op_56655_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_56655_end_mask_0 = const()[name = tensor<string, []>("op_56655_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56655_cast_fp16 = slice_by_index(begin = var_56655_begin_0, end = var_56655_end_0, end_mask = var_56655_end_mask_0, x = q_251_cast_fp16)[name = tensor<string, []>("op_56655_cast_fp16")];
            tensor<int32, [4]> var_56659_begin_0 = const()[name = tensor<string, []>("op_56659_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_56659_end_0 = const()[name = tensor<string, []>("op_56659_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_56659_end_mask_0 = const()[name = tensor<string, []>("op_56659_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56659_cast_fp16 = slice_by_index(begin = var_56659_begin_0, end = var_56659_end_0, end_mask = var_56659_end_mask_0, x = q_251_cast_fp16)[name = tensor<string, []>("op_56659_cast_fp16")];
            tensor<int32, [4]> var_56663_begin_0 = const()[name = tensor<string, []>("op_56663_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_56663_end_0 = const()[name = tensor<string, []>("op_56663_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_56663_end_mask_0 = const()[name = tensor<string, []>("op_56663_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56663_cast_fp16 = slice_by_index(begin = var_56663_begin_0, end = var_56663_end_0, end_mask = var_56663_end_mask_0, x = q_251_cast_fp16)[name = tensor<string, []>("op_56663_cast_fp16")];
            tensor<int32, [4]> var_56667_begin_0 = const()[name = tensor<string, []>("op_56667_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_56667_end_0 = const()[name = tensor<string, []>("op_56667_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_56667_end_mask_0 = const()[name = tensor<string, []>("op_56667_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56667_cast_fp16 = slice_by_index(begin = var_56667_begin_0, end = var_56667_end_0, end_mask = var_56667_end_mask_0, x = q_251_cast_fp16)[name = tensor<string, []>("op_56667_cast_fp16")];
            tensor<int32, [4]> var_56671_begin_0 = const()[name = tensor<string, []>("op_56671_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_56671_end_0 = const()[name = tensor<string, []>("op_56671_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_56671_end_mask_0 = const()[name = tensor<string, []>("op_56671_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56671_cast_fp16 = slice_by_index(begin = var_56671_begin_0, end = var_56671_end_0, end_mask = var_56671_end_mask_0, x = q_251_cast_fp16)[name = tensor<string, []>("op_56671_cast_fp16")];
            tensor<int32, [4]> var_56675_begin_0 = const()[name = tensor<string, []>("op_56675_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_56675_end_0 = const()[name = tensor<string, []>("op_56675_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_56675_end_mask_0 = const()[name = tensor<string, []>("op_56675_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56675_cast_fp16 = slice_by_index(begin = var_56675_begin_0, end = var_56675_end_0, end_mask = var_56675_end_mask_0, x = q_251_cast_fp16)[name = tensor<string, []>("op_56675_cast_fp16")];
            tensor<int32, [4]> var_56679_begin_0 = const()[name = tensor<string, []>("op_56679_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_56679_end_0 = const()[name = tensor<string, []>("op_56679_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_56679_end_mask_0 = const()[name = tensor<string, []>("op_56679_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56679_cast_fp16 = slice_by_index(begin = var_56679_begin_0, end = var_56679_end_0, end_mask = var_56679_end_mask_0, x = q_251_cast_fp16)[name = tensor<string, []>("op_56679_cast_fp16")];
            tensor<int32, [4]> var_56683_begin_0 = const()[name = tensor<string, []>("op_56683_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_56683_end_0 = const()[name = tensor<string, []>("op_56683_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_56683_end_mask_0 = const()[name = tensor<string, []>("op_56683_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56683_cast_fp16 = slice_by_index(begin = var_56683_begin_0, end = var_56683_end_0, end_mask = var_56683_end_mask_0, x = q_251_cast_fp16)[name = tensor<string, []>("op_56683_cast_fp16")];
            tensor<int32, [4]> var_56687_begin_0 = const()[name = tensor<string, []>("op_56687_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_56687_end_0 = const()[name = tensor<string, []>("op_56687_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_56687_end_mask_0 = const()[name = tensor<string, []>("op_56687_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56687_cast_fp16 = slice_by_index(begin = var_56687_begin_0, end = var_56687_end_0, end_mask = var_56687_end_mask_0, x = q_251_cast_fp16)[name = tensor<string, []>("op_56687_cast_fp16")];
            tensor<int32, [4]> var_56691_begin_0 = const()[name = tensor<string, []>("op_56691_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_56691_end_0 = const()[name = tensor<string, []>("op_56691_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_56691_end_mask_0 = const()[name = tensor<string, []>("op_56691_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56691_cast_fp16 = slice_by_index(begin = var_56691_begin_0, end = var_56691_end_0, end_mask = var_56691_end_mask_0, x = q_251_cast_fp16)[name = tensor<string, []>("op_56691_cast_fp16")];
            tensor<int32, [4]> var_56695_begin_0 = const()[name = tensor<string, []>("op_56695_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_56695_end_0 = const()[name = tensor<string, []>("op_56695_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_56695_end_mask_0 = const()[name = tensor<string, []>("op_56695_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56695_cast_fp16 = slice_by_index(begin = var_56695_begin_0, end = var_56695_end_0, end_mask = var_56695_end_mask_0, x = q_251_cast_fp16)[name = tensor<string, []>("op_56695_cast_fp16")];
            tensor<int32, [4]> var_56699_begin_0 = const()[name = tensor<string, []>("op_56699_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_56699_end_0 = const()[name = tensor<string, []>("op_56699_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_56699_end_mask_0 = const()[name = tensor<string, []>("op_56699_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56699_cast_fp16 = slice_by_index(begin = var_56699_begin_0, end = var_56699_end_0, end_mask = var_56699_end_mask_0, x = q_251_cast_fp16)[name = tensor<string, []>("op_56699_cast_fp16")];
            tensor<int32, [4]> var_56703_begin_0 = const()[name = tensor<string, []>("op_56703_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_56703_end_0 = const()[name = tensor<string, []>("op_56703_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_56703_end_mask_0 = const()[name = tensor<string, []>("op_56703_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56703_cast_fp16 = slice_by_index(begin = var_56703_begin_0, end = var_56703_end_0, end_mask = var_56703_end_mask_0, x = q_251_cast_fp16)[name = tensor<string, []>("op_56703_cast_fp16")];
            tensor<int32, [4]> var_56707_begin_0 = const()[name = tensor<string, []>("op_56707_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_56707_end_0 = const()[name = tensor<string, []>("op_56707_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_56707_end_mask_0 = const()[name = tensor<string, []>("op_56707_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56707_cast_fp16 = slice_by_index(begin = var_56707_begin_0, end = var_56707_end_0, end_mask = var_56707_end_mask_0, x = q_251_cast_fp16)[name = tensor<string, []>("op_56707_cast_fp16")];
            tensor<int32, [4]> var_56711_begin_0 = const()[name = tensor<string, []>("op_56711_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_56711_end_0 = const()[name = tensor<string, []>("op_56711_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_56711_end_mask_0 = const()[name = tensor<string, []>("op_56711_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56711_cast_fp16 = slice_by_index(begin = var_56711_begin_0, end = var_56711_end_0, end_mask = var_56711_end_mask_0, x = q_251_cast_fp16)[name = tensor<string, []>("op_56711_cast_fp16")];
            tensor<int32, [4]> var_56715_begin_0 = const()[name = tensor<string, []>("op_56715_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_56715_end_0 = const()[name = tensor<string, []>("op_56715_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_56715_end_mask_0 = const()[name = tensor<string, []>("op_56715_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56715_cast_fp16 = slice_by_index(begin = var_56715_begin_0, end = var_56715_end_0, end_mask = var_56715_end_mask_0, x = q_251_cast_fp16)[name = tensor<string, []>("op_56715_cast_fp16")];
            tensor<int32, [4]> var_56719_begin_0 = const()[name = tensor<string, []>("op_56719_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_56719_end_0 = const()[name = tensor<string, []>("op_56719_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_56719_end_mask_0 = const()[name = tensor<string, []>("op_56719_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56719_cast_fp16 = slice_by_index(begin = var_56719_begin_0, end = var_56719_end_0, end_mask = var_56719_end_mask_0, x = q_251_cast_fp16)[name = tensor<string, []>("op_56719_cast_fp16")];
            tensor<int32, [4]> var_56723_begin_0 = const()[name = tensor<string, []>("op_56723_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_56723_end_0 = const()[name = tensor<string, []>("op_56723_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_56723_end_mask_0 = const()[name = tensor<string, []>("op_56723_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56723_cast_fp16 = slice_by_index(begin = var_56723_begin_0, end = var_56723_end_0, end_mask = var_56723_end_mask_0, x = q_251_cast_fp16)[name = tensor<string, []>("op_56723_cast_fp16")];
            tensor<int32, [4]> k_503_perm_0 = const()[name = tensor<string, []>("k_503_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_56730_begin_0 = const()[name = tensor<string, []>("op_56730_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_56730_end_0 = const()[name = tensor<string, []>("op_56730_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_56730_end_mask_0 = const()[name = tensor<string, []>("op_56730_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> k_503_cast_fp16 = transpose(perm = k_503_perm_0, x = k_501_cast_fp16)[name = tensor<string, []>("transpose_14")];
            tensor<fp16, [2, 77, 1, 64]> var_56730_cast_fp16 = slice_by_index(begin = var_56730_begin_0, end = var_56730_end_0, end_mask = var_56730_end_mask_0, x = k_503_cast_fp16)[name = tensor<string, []>("op_56730_cast_fp16")];
            tensor<int32, [4]> var_56734_begin_0 = const()[name = tensor<string, []>("op_56734_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_56734_end_0 = const()[name = tensor<string, []>("op_56734_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_56734_end_mask_0 = const()[name = tensor<string, []>("op_56734_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_56734_cast_fp16 = slice_by_index(begin = var_56734_begin_0, end = var_56734_end_0, end_mask = var_56734_end_mask_0, x = k_503_cast_fp16)[name = tensor<string, []>("op_56734_cast_fp16")];
            tensor<int32, [4]> var_56738_begin_0 = const()[name = tensor<string, []>("op_56738_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_56738_end_0 = const()[name = tensor<string, []>("op_56738_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_56738_end_mask_0 = const()[name = tensor<string, []>("op_56738_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_56738_cast_fp16 = slice_by_index(begin = var_56738_begin_0, end = var_56738_end_0, end_mask = var_56738_end_mask_0, x = k_503_cast_fp16)[name = tensor<string, []>("op_56738_cast_fp16")];
            tensor<int32, [4]> var_56742_begin_0 = const()[name = tensor<string, []>("op_56742_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_56742_end_0 = const()[name = tensor<string, []>("op_56742_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_56742_end_mask_0 = const()[name = tensor<string, []>("op_56742_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_56742_cast_fp16 = slice_by_index(begin = var_56742_begin_0, end = var_56742_end_0, end_mask = var_56742_end_mask_0, x = k_503_cast_fp16)[name = tensor<string, []>("op_56742_cast_fp16")];
            tensor<int32, [4]> var_56746_begin_0 = const()[name = tensor<string, []>("op_56746_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_56746_end_0 = const()[name = tensor<string, []>("op_56746_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_56746_end_mask_0 = const()[name = tensor<string, []>("op_56746_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_56746_cast_fp16 = slice_by_index(begin = var_56746_begin_0, end = var_56746_end_0, end_mask = var_56746_end_mask_0, x = k_503_cast_fp16)[name = tensor<string, []>("op_56746_cast_fp16")];
            tensor<int32, [4]> var_56750_begin_0 = const()[name = tensor<string, []>("op_56750_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_56750_end_0 = const()[name = tensor<string, []>("op_56750_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_56750_end_mask_0 = const()[name = tensor<string, []>("op_56750_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_56750_cast_fp16 = slice_by_index(begin = var_56750_begin_0, end = var_56750_end_0, end_mask = var_56750_end_mask_0, x = k_503_cast_fp16)[name = tensor<string, []>("op_56750_cast_fp16")];
            tensor<int32, [4]> var_56754_begin_0 = const()[name = tensor<string, []>("op_56754_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_56754_end_0 = const()[name = tensor<string, []>("op_56754_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_56754_end_mask_0 = const()[name = tensor<string, []>("op_56754_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_56754_cast_fp16 = slice_by_index(begin = var_56754_begin_0, end = var_56754_end_0, end_mask = var_56754_end_mask_0, x = k_503_cast_fp16)[name = tensor<string, []>("op_56754_cast_fp16")];
            tensor<int32, [4]> var_56758_begin_0 = const()[name = tensor<string, []>("op_56758_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_56758_end_0 = const()[name = tensor<string, []>("op_56758_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_56758_end_mask_0 = const()[name = tensor<string, []>("op_56758_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_56758_cast_fp16 = slice_by_index(begin = var_56758_begin_0, end = var_56758_end_0, end_mask = var_56758_end_mask_0, x = k_503_cast_fp16)[name = tensor<string, []>("op_56758_cast_fp16")];
            tensor<int32, [4]> var_56762_begin_0 = const()[name = tensor<string, []>("op_56762_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_56762_end_0 = const()[name = tensor<string, []>("op_56762_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_56762_end_mask_0 = const()[name = tensor<string, []>("op_56762_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_56762_cast_fp16 = slice_by_index(begin = var_56762_begin_0, end = var_56762_end_0, end_mask = var_56762_end_mask_0, x = k_503_cast_fp16)[name = tensor<string, []>("op_56762_cast_fp16")];
            tensor<int32, [4]> var_56766_begin_0 = const()[name = tensor<string, []>("op_56766_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_56766_end_0 = const()[name = tensor<string, []>("op_56766_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_56766_end_mask_0 = const()[name = tensor<string, []>("op_56766_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_56766_cast_fp16 = slice_by_index(begin = var_56766_begin_0, end = var_56766_end_0, end_mask = var_56766_end_mask_0, x = k_503_cast_fp16)[name = tensor<string, []>("op_56766_cast_fp16")];
            tensor<int32, [4]> var_56770_begin_0 = const()[name = tensor<string, []>("op_56770_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_56770_end_0 = const()[name = tensor<string, []>("op_56770_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_56770_end_mask_0 = const()[name = tensor<string, []>("op_56770_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_56770_cast_fp16 = slice_by_index(begin = var_56770_begin_0, end = var_56770_end_0, end_mask = var_56770_end_mask_0, x = k_503_cast_fp16)[name = tensor<string, []>("op_56770_cast_fp16")];
            tensor<int32, [4]> var_56774_begin_0 = const()[name = tensor<string, []>("op_56774_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_56774_end_0 = const()[name = tensor<string, []>("op_56774_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_56774_end_mask_0 = const()[name = tensor<string, []>("op_56774_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_56774_cast_fp16 = slice_by_index(begin = var_56774_begin_0, end = var_56774_end_0, end_mask = var_56774_end_mask_0, x = k_503_cast_fp16)[name = tensor<string, []>("op_56774_cast_fp16")];
            tensor<int32, [4]> var_56778_begin_0 = const()[name = tensor<string, []>("op_56778_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_56778_end_0 = const()[name = tensor<string, []>("op_56778_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_56778_end_mask_0 = const()[name = tensor<string, []>("op_56778_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_56778_cast_fp16 = slice_by_index(begin = var_56778_begin_0, end = var_56778_end_0, end_mask = var_56778_end_mask_0, x = k_503_cast_fp16)[name = tensor<string, []>("op_56778_cast_fp16")];
            tensor<int32, [4]> var_56782_begin_0 = const()[name = tensor<string, []>("op_56782_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_56782_end_0 = const()[name = tensor<string, []>("op_56782_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_56782_end_mask_0 = const()[name = tensor<string, []>("op_56782_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_56782_cast_fp16 = slice_by_index(begin = var_56782_begin_0, end = var_56782_end_0, end_mask = var_56782_end_mask_0, x = k_503_cast_fp16)[name = tensor<string, []>("op_56782_cast_fp16")];
            tensor<int32, [4]> var_56786_begin_0 = const()[name = tensor<string, []>("op_56786_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_56786_end_0 = const()[name = tensor<string, []>("op_56786_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_56786_end_mask_0 = const()[name = tensor<string, []>("op_56786_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_56786_cast_fp16 = slice_by_index(begin = var_56786_begin_0, end = var_56786_end_0, end_mask = var_56786_end_mask_0, x = k_503_cast_fp16)[name = tensor<string, []>("op_56786_cast_fp16")];
            tensor<int32, [4]> var_56790_begin_0 = const()[name = tensor<string, []>("op_56790_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_56790_end_0 = const()[name = tensor<string, []>("op_56790_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_56790_end_mask_0 = const()[name = tensor<string, []>("op_56790_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_56790_cast_fp16 = slice_by_index(begin = var_56790_begin_0, end = var_56790_end_0, end_mask = var_56790_end_mask_0, x = k_503_cast_fp16)[name = tensor<string, []>("op_56790_cast_fp16")];
            tensor<int32, [4]> var_56794_begin_0 = const()[name = tensor<string, []>("op_56794_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_56794_end_0 = const()[name = tensor<string, []>("op_56794_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_56794_end_mask_0 = const()[name = tensor<string, []>("op_56794_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_56794_cast_fp16 = slice_by_index(begin = var_56794_begin_0, end = var_56794_end_0, end_mask = var_56794_end_mask_0, x = k_503_cast_fp16)[name = tensor<string, []>("op_56794_cast_fp16")];
            tensor<int32, [4]> var_56798_begin_0 = const()[name = tensor<string, []>("op_56798_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_56798_end_0 = const()[name = tensor<string, []>("op_56798_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_56798_end_mask_0 = const()[name = tensor<string, []>("op_56798_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_56798_cast_fp16 = slice_by_index(begin = var_56798_begin_0, end = var_56798_end_0, end_mask = var_56798_end_mask_0, x = k_503_cast_fp16)[name = tensor<string, []>("op_56798_cast_fp16")];
            tensor<int32, [4]> var_56802_begin_0 = const()[name = tensor<string, []>("op_56802_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_56802_end_0 = const()[name = tensor<string, []>("op_56802_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_56802_end_mask_0 = const()[name = tensor<string, []>("op_56802_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_56802_cast_fp16 = slice_by_index(begin = var_56802_begin_0, end = var_56802_end_0, end_mask = var_56802_end_mask_0, x = k_503_cast_fp16)[name = tensor<string, []>("op_56802_cast_fp16")];
            tensor<int32, [4]> var_56806_begin_0 = const()[name = tensor<string, []>("op_56806_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_56806_end_0 = const()[name = tensor<string, []>("op_56806_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_56806_end_mask_0 = const()[name = tensor<string, []>("op_56806_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_56806_cast_fp16 = slice_by_index(begin = var_56806_begin_0, end = var_56806_end_0, end_mask = var_56806_end_mask_0, x = k_503_cast_fp16)[name = tensor<string, []>("op_56806_cast_fp16")];
            tensor<int32, [4]> var_56808_begin_0 = const()[name = tensor<string, []>("op_56808_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_56808_end_0 = const()[name = tensor<string, []>("op_56808_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_56808_end_mask_0 = const()[name = tensor<string, []>("op_56808_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_56808_cast_fp16 = slice_by_index(begin = var_56808_begin_0, end = var_56808_end_0, end_mask = var_56808_end_mask_0, x = v_251_cast_fp16)[name = tensor<string, []>("op_56808_cast_fp16")];
            tensor<int32, [4]> var_56812_begin_0 = const()[name = tensor<string, []>("op_56812_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_56812_end_0 = const()[name = tensor<string, []>("op_56812_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_56812_end_mask_0 = const()[name = tensor<string, []>("op_56812_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_56812_cast_fp16 = slice_by_index(begin = var_56812_begin_0, end = var_56812_end_0, end_mask = var_56812_end_mask_0, x = v_251_cast_fp16)[name = tensor<string, []>("op_56812_cast_fp16")];
            tensor<int32, [4]> var_56816_begin_0 = const()[name = tensor<string, []>("op_56816_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_56816_end_0 = const()[name = tensor<string, []>("op_56816_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_56816_end_mask_0 = const()[name = tensor<string, []>("op_56816_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_56816_cast_fp16 = slice_by_index(begin = var_56816_begin_0, end = var_56816_end_0, end_mask = var_56816_end_mask_0, x = v_251_cast_fp16)[name = tensor<string, []>("op_56816_cast_fp16")];
            tensor<int32, [4]> var_56820_begin_0 = const()[name = tensor<string, []>("op_56820_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_56820_end_0 = const()[name = tensor<string, []>("op_56820_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_56820_end_mask_0 = const()[name = tensor<string, []>("op_56820_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_56820_cast_fp16 = slice_by_index(begin = var_56820_begin_0, end = var_56820_end_0, end_mask = var_56820_end_mask_0, x = v_251_cast_fp16)[name = tensor<string, []>("op_56820_cast_fp16")];
            tensor<int32, [4]> var_56824_begin_0 = const()[name = tensor<string, []>("op_56824_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_56824_end_0 = const()[name = tensor<string, []>("op_56824_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_56824_end_mask_0 = const()[name = tensor<string, []>("op_56824_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_56824_cast_fp16 = slice_by_index(begin = var_56824_begin_0, end = var_56824_end_0, end_mask = var_56824_end_mask_0, x = v_251_cast_fp16)[name = tensor<string, []>("op_56824_cast_fp16")];
            tensor<int32, [4]> var_56828_begin_0 = const()[name = tensor<string, []>("op_56828_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_56828_end_0 = const()[name = tensor<string, []>("op_56828_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_56828_end_mask_0 = const()[name = tensor<string, []>("op_56828_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_56828_cast_fp16 = slice_by_index(begin = var_56828_begin_0, end = var_56828_end_0, end_mask = var_56828_end_mask_0, x = v_251_cast_fp16)[name = tensor<string, []>("op_56828_cast_fp16")];
            tensor<int32, [4]> var_56832_begin_0 = const()[name = tensor<string, []>("op_56832_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_56832_end_0 = const()[name = tensor<string, []>("op_56832_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_56832_end_mask_0 = const()[name = tensor<string, []>("op_56832_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_56832_cast_fp16 = slice_by_index(begin = var_56832_begin_0, end = var_56832_end_0, end_mask = var_56832_end_mask_0, x = v_251_cast_fp16)[name = tensor<string, []>("op_56832_cast_fp16")];
            tensor<int32, [4]> var_56836_begin_0 = const()[name = tensor<string, []>("op_56836_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_56836_end_0 = const()[name = tensor<string, []>("op_56836_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_56836_end_mask_0 = const()[name = tensor<string, []>("op_56836_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_56836_cast_fp16 = slice_by_index(begin = var_56836_begin_0, end = var_56836_end_0, end_mask = var_56836_end_mask_0, x = v_251_cast_fp16)[name = tensor<string, []>("op_56836_cast_fp16")];
            tensor<int32, [4]> var_56840_begin_0 = const()[name = tensor<string, []>("op_56840_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_56840_end_0 = const()[name = tensor<string, []>("op_56840_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_56840_end_mask_0 = const()[name = tensor<string, []>("op_56840_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_56840_cast_fp16 = slice_by_index(begin = var_56840_begin_0, end = var_56840_end_0, end_mask = var_56840_end_mask_0, x = v_251_cast_fp16)[name = tensor<string, []>("op_56840_cast_fp16")];
            tensor<int32, [4]> var_56844_begin_0 = const()[name = tensor<string, []>("op_56844_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_56844_end_0 = const()[name = tensor<string, []>("op_56844_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_56844_end_mask_0 = const()[name = tensor<string, []>("op_56844_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_56844_cast_fp16 = slice_by_index(begin = var_56844_begin_0, end = var_56844_end_0, end_mask = var_56844_end_mask_0, x = v_251_cast_fp16)[name = tensor<string, []>("op_56844_cast_fp16")];
            tensor<int32, [4]> var_56848_begin_0 = const()[name = tensor<string, []>("op_56848_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_56848_end_0 = const()[name = tensor<string, []>("op_56848_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_56848_end_mask_0 = const()[name = tensor<string, []>("op_56848_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_56848_cast_fp16 = slice_by_index(begin = var_56848_begin_0, end = var_56848_end_0, end_mask = var_56848_end_mask_0, x = v_251_cast_fp16)[name = tensor<string, []>("op_56848_cast_fp16")];
            tensor<int32, [4]> var_56852_begin_0 = const()[name = tensor<string, []>("op_56852_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_56852_end_0 = const()[name = tensor<string, []>("op_56852_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_56852_end_mask_0 = const()[name = tensor<string, []>("op_56852_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_56852_cast_fp16 = slice_by_index(begin = var_56852_begin_0, end = var_56852_end_0, end_mask = var_56852_end_mask_0, x = v_251_cast_fp16)[name = tensor<string, []>("op_56852_cast_fp16")];
            tensor<int32, [4]> var_56856_begin_0 = const()[name = tensor<string, []>("op_56856_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_56856_end_0 = const()[name = tensor<string, []>("op_56856_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_56856_end_mask_0 = const()[name = tensor<string, []>("op_56856_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_56856_cast_fp16 = slice_by_index(begin = var_56856_begin_0, end = var_56856_end_0, end_mask = var_56856_end_mask_0, x = v_251_cast_fp16)[name = tensor<string, []>("op_56856_cast_fp16")];
            tensor<int32, [4]> var_56860_begin_0 = const()[name = tensor<string, []>("op_56860_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_56860_end_0 = const()[name = tensor<string, []>("op_56860_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_56860_end_mask_0 = const()[name = tensor<string, []>("op_56860_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_56860_cast_fp16 = slice_by_index(begin = var_56860_begin_0, end = var_56860_end_0, end_mask = var_56860_end_mask_0, x = v_251_cast_fp16)[name = tensor<string, []>("op_56860_cast_fp16")];
            tensor<int32, [4]> var_56864_begin_0 = const()[name = tensor<string, []>("op_56864_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_56864_end_0 = const()[name = tensor<string, []>("op_56864_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_56864_end_mask_0 = const()[name = tensor<string, []>("op_56864_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_56864_cast_fp16 = slice_by_index(begin = var_56864_begin_0, end = var_56864_end_0, end_mask = var_56864_end_mask_0, x = v_251_cast_fp16)[name = tensor<string, []>("op_56864_cast_fp16")];
            tensor<int32, [4]> var_56868_begin_0 = const()[name = tensor<string, []>("op_56868_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_56868_end_0 = const()[name = tensor<string, []>("op_56868_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_56868_end_mask_0 = const()[name = tensor<string, []>("op_56868_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_56868_cast_fp16 = slice_by_index(begin = var_56868_begin_0, end = var_56868_end_0, end_mask = var_56868_end_mask_0, x = v_251_cast_fp16)[name = tensor<string, []>("op_56868_cast_fp16")];
            tensor<int32, [4]> var_56872_begin_0 = const()[name = tensor<string, []>("op_56872_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_56872_end_0 = const()[name = tensor<string, []>("op_56872_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_56872_end_mask_0 = const()[name = tensor<string, []>("op_56872_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_56872_cast_fp16 = slice_by_index(begin = var_56872_begin_0, end = var_56872_end_0, end_mask = var_56872_end_mask_0, x = v_251_cast_fp16)[name = tensor<string, []>("op_56872_cast_fp16")];
            tensor<int32, [4]> var_56876_begin_0 = const()[name = tensor<string, []>("op_56876_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_56876_end_0 = const()[name = tensor<string, []>("op_56876_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_56876_end_mask_0 = const()[name = tensor<string, []>("op_56876_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_56876_cast_fp16 = slice_by_index(begin = var_56876_begin_0, end = var_56876_end_0, end_mask = var_56876_end_mask_0, x = v_251_cast_fp16)[name = tensor<string, []>("op_56876_cast_fp16")];
            tensor<int32, [4]> var_56880_begin_0 = const()[name = tensor<string, []>("op_56880_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_56880_end_0 = const()[name = tensor<string, []>("op_56880_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_56880_end_mask_0 = const()[name = tensor<string, []>("op_56880_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_56880_cast_fp16 = slice_by_index(begin = var_56880_begin_0, end = var_56880_end_0, end_mask = var_56880_end_mask_0, x = v_251_cast_fp16)[name = tensor<string, []>("op_56880_cast_fp16")];
            tensor<int32, [4]> var_56884_begin_0 = const()[name = tensor<string, []>("op_56884_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_56884_end_0 = const()[name = tensor<string, []>("op_56884_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_56884_end_mask_0 = const()[name = tensor<string, []>("op_56884_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_56884_cast_fp16 = slice_by_index(begin = var_56884_begin_0, end = var_56884_end_0, end_mask = var_56884_end_mask_0, x = v_251_cast_fp16)[name = tensor<string, []>("op_56884_cast_fp16")];
            tensor<string, []> var_56888_equation_0 = const()[name = tensor<string, []>("op_56888_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56888_cast_fp16 = einsum(equation = var_56888_equation_0, values = (var_56730_cast_fp16, var_56647_cast_fp16))[name = tensor<string, []>("op_56888_cast_fp16")];
            tensor<fp16, []> var_56889_to_fp16 = const()[name = tensor<string, []>("op_56889_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4841_cast_fp16 = mul(x = var_56888_cast_fp16, y = var_56889_to_fp16)[name = tensor<string, []>("aw_4841_cast_fp16")];
            tensor<string, []> var_56892_equation_0 = const()[name = tensor<string, []>("op_56892_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56892_cast_fp16 = einsum(equation = var_56892_equation_0, values = (var_56734_cast_fp16, var_56651_cast_fp16))[name = tensor<string, []>("op_56892_cast_fp16")];
            tensor<fp16, []> var_56893_to_fp16 = const()[name = tensor<string, []>("op_56893_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4843_cast_fp16 = mul(x = var_56892_cast_fp16, y = var_56893_to_fp16)[name = tensor<string, []>("aw_4843_cast_fp16")];
            tensor<string, []> var_56896_equation_0 = const()[name = tensor<string, []>("op_56896_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56896_cast_fp16 = einsum(equation = var_56896_equation_0, values = (var_56738_cast_fp16, var_56655_cast_fp16))[name = tensor<string, []>("op_56896_cast_fp16")];
            tensor<fp16, []> var_56897_to_fp16 = const()[name = tensor<string, []>("op_56897_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4845_cast_fp16 = mul(x = var_56896_cast_fp16, y = var_56897_to_fp16)[name = tensor<string, []>("aw_4845_cast_fp16")];
            tensor<string, []> var_56900_equation_0 = const()[name = tensor<string, []>("op_56900_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56900_cast_fp16 = einsum(equation = var_56900_equation_0, values = (var_56742_cast_fp16, var_56659_cast_fp16))[name = tensor<string, []>("op_56900_cast_fp16")];
            tensor<fp16, []> var_56901_to_fp16 = const()[name = tensor<string, []>("op_56901_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4847_cast_fp16 = mul(x = var_56900_cast_fp16, y = var_56901_to_fp16)[name = tensor<string, []>("aw_4847_cast_fp16")];
            tensor<string, []> var_56904_equation_0 = const()[name = tensor<string, []>("op_56904_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56904_cast_fp16 = einsum(equation = var_56904_equation_0, values = (var_56746_cast_fp16, var_56663_cast_fp16))[name = tensor<string, []>("op_56904_cast_fp16")];
            tensor<fp16, []> var_56905_to_fp16 = const()[name = tensor<string, []>("op_56905_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4849_cast_fp16 = mul(x = var_56904_cast_fp16, y = var_56905_to_fp16)[name = tensor<string, []>("aw_4849_cast_fp16")];
            tensor<string, []> var_56908_equation_0 = const()[name = tensor<string, []>("op_56908_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56908_cast_fp16 = einsum(equation = var_56908_equation_0, values = (var_56750_cast_fp16, var_56667_cast_fp16))[name = tensor<string, []>("op_56908_cast_fp16")];
            tensor<fp16, []> var_56909_to_fp16 = const()[name = tensor<string, []>("op_56909_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4851_cast_fp16 = mul(x = var_56908_cast_fp16, y = var_56909_to_fp16)[name = tensor<string, []>("aw_4851_cast_fp16")];
            tensor<string, []> var_56912_equation_0 = const()[name = tensor<string, []>("op_56912_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56912_cast_fp16 = einsum(equation = var_56912_equation_0, values = (var_56754_cast_fp16, var_56671_cast_fp16))[name = tensor<string, []>("op_56912_cast_fp16")];
            tensor<fp16, []> var_56913_to_fp16 = const()[name = tensor<string, []>("op_56913_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4853_cast_fp16 = mul(x = var_56912_cast_fp16, y = var_56913_to_fp16)[name = tensor<string, []>("aw_4853_cast_fp16")];
            tensor<string, []> var_56916_equation_0 = const()[name = tensor<string, []>("op_56916_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56916_cast_fp16 = einsum(equation = var_56916_equation_0, values = (var_56758_cast_fp16, var_56675_cast_fp16))[name = tensor<string, []>("op_56916_cast_fp16")];
            tensor<fp16, []> var_56917_to_fp16 = const()[name = tensor<string, []>("op_56917_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4855_cast_fp16 = mul(x = var_56916_cast_fp16, y = var_56917_to_fp16)[name = tensor<string, []>("aw_4855_cast_fp16")];
            tensor<string, []> var_56920_equation_0 = const()[name = tensor<string, []>("op_56920_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56920_cast_fp16 = einsum(equation = var_56920_equation_0, values = (var_56762_cast_fp16, var_56679_cast_fp16))[name = tensor<string, []>("op_56920_cast_fp16")];
            tensor<fp16, []> var_56921_to_fp16 = const()[name = tensor<string, []>("op_56921_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4857_cast_fp16 = mul(x = var_56920_cast_fp16, y = var_56921_to_fp16)[name = tensor<string, []>("aw_4857_cast_fp16")];
            tensor<string, []> var_56924_equation_0 = const()[name = tensor<string, []>("op_56924_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56924_cast_fp16 = einsum(equation = var_56924_equation_0, values = (var_56766_cast_fp16, var_56683_cast_fp16))[name = tensor<string, []>("op_56924_cast_fp16")];
            tensor<fp16, []> var_56925_to_fp16 = const()[name = tensor<string, []>("op_56925_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4859_cast_fp16 = mul(x = var_56924_cast_fp16, y = var_56925_to_fp16)[name = tensor<string, []>("aw_4859_cast_fp16")];
            tensor<string, []> var_56928_equation_0 = const()[name = tensor<string, []>("op_56928_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56928_cast_fp16 = einsum(equation = var_56928_equation_0, values = (var_56770_cast_fp16, var_56687_cast_fp16))[name = tensor<string, []>("op_56928_cast_fp16")];
            tensor<fp16, []> var_56929_to_fp16 = const()[name = tensor<string, []>("op_56929_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4861_cast_fp16 = mul(x = var_56928_cast_fp16, y = var_56929_to_fp16)[name = tensor<string, []>("aw_4861_cast_fp16")];
            tensor<string, []> var_56932_equation_0 = const()[name = tensor<string, []>("op_56932_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56932_cast_fp16 = einsum(equation = var_56932_equation_0, values = (var_56774_cast_fp16, var_56691_cast_fp16))[name = tensor<string, []>("op_56932_cast_fp16")];
            tensor<fp16, []> var_56933_to_fp16 = const()[name = tensor<string, []>("op_56933_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4863_cast_fp16 = mul(x = var_56932_cast_fp16, y = var_56933_to_fp16)[name = tensor<string, []>("aw_4863_cast_fp16")];
            tensor<string, []> var_56936_equation_0 = const()[name = tensor<string, []>("op_56936_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56936_cast_fp16 = einsum(equation = var_56936_equation_0, values = (var_56778_cast_fp16, var_56695_cast_fp16))[name = tensor<string, []>("op_56936_cast_fp16")];
            tensor<fp16, []> var_56937_to_fp16 = const()[name = tensor<string, []>("op_56937_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4865_cast_fp16 = mul(x = var_56936_cast_fp16, y = var_56937_to_fp16)[name = tensor<string, []>("aw_4865_cast_fp16")];
            tensor<string, []> var_56940_equation_0 = const()[name = tensor<string, []>("op_56940_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56940_cast_fp16 = einsum(equation = var_56940_equation_0, values = (var_56782_cast_fp16, var_56699_cast_fp16))[name = tensor<string, []>("op_56940_cast_fp16")];
            tensor<fp16, []> var_56941_to_fp16 = const()[name = tensor<string, []>("op_56941_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4867_cast_fp16 = mul(x = var_56940_cast_fp16, y = var_56941_to_fp16)[name = tensor<string, []>("aw_4867_cast_fp16")];
            tensor<string, []> var_56944_equation_0 = const()[name = tensor<string, []>("op_56944_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56944_cast_fp16 = einsum(equation = var_56944_equation_0, values = (var_56786_cast_fp16, var_56703_cast_fp16))[name = tensor<string, []>("op_56944_cast_fp16")];
            tensor<fp16, []> var_56945_to_fp16 = const()[name = tensor<string, []>("op_56945_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4869_cast_fp16 = mul(x = var_56944_cast_fp16, y = var_56945_to_fp16)[name = tensor<string, []>("aw_4869_cast_fp16")];
            tensor<string, []> var_56948_equation_0 = const()[name = tensor<string, []>("op_56948_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56948_cast_fp16 = einsum(equation = var_56948_equation_0, values = (var_56790_cast_fp16, var_56707_cast_fp16))[name = tensor<string, []>("op_56948_cast_fp16")];
            tensor<fp16, []> var_56949_to_fp16 = const()[name = tensor<string, []>("op_56949_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4871_cast_fp16 = mul(x = var_56948_cast_fp16, y = var_56949_to_fp16)[name = tensor<string, []>("aw_4871_cast_fp16")];
            tensor<string, []> var_56952_equation_0 = const()[name = tensor<string, []>("op_56952_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56952_cast_fp16 = einsum(equation = var_56952_equation_0, values = (var_56794_cast_fp16, var_56711_cast_fp16))[name = tensor<string, []>("op_56952_cast_fp16")];
            tensor<fp16, []> var_56953_to_fp16 = const()[name = tensor<string, []>("op_56953_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4873_cast_fp16 = mul(x = var_56952_cast_fp16, y = var_56953_to_fp16)[name = tensor<string, []>("aw_4873_cast_fp16")];
            tensor<string, []> var_56956_equation_0 = const()[name = tensor<string, []>("op_56956_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56956_cast_fp16 = einsum(equation = var_56956_equation_0, values = (var_56798_cast_fp16, var_56715_cast_fp16))[name = tensor<string, []>("op_56956_cast_fp16")];
            tensor<fp16, []> var_56957_to_fp16 = const()[name = tensor<string, []>("op_56957_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4875_cast_fp16 = mul(x = var_56956_cast_fp16, y = var_56957_to_fp16)[name = tensor<string, []>("aw_4875_cast_fp16")];
            tensor<string, []> var_56960_equation_0 = const()[name = tensor<string, []>("op_56960_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56960_cast_fp16 = einsum(equation = var_56960_equation_0, values = (var_56802_cast_fp16, var_56719_cast_fp16))[name = tensor<string, []>("op_56960_cast_fp16")];
            tensor<fp16, []> var_56961_to_fp16 = const()[name = tensor<string, []>("op_56961_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4877_cast_fp16 = mul(x = var_56960_cast_fp16, y = var_56961_to_fp16)[name = tensor<string, []>("aw_4877_cast_fp16")];
            tensor<string, []> var_56964_equation_0 = const()[name = tensor<string, []>("op_56964_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56964_cast_fp16 = einsum(equation = var_56964_equation_0, values = (var_56806_cast_fp16, var_56723_cast_fp16))[name = tensor<string, []>("op_56964_cast_fp16")];
            tensor<fp16, []> var_56965_to_fp16 = const()[name = tensor<string, []>("op_56965_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4879_cast_fp16 = mul(x = var_56964_cast_fp16, y = var_56965_to_fp16)[name = tensor<string, []>("aw_4879_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_56967_cast_fp16 = softmax(axis = var_30355, x = aw_4841_cast_fp16)[name = tensor<string, []>("op_56967_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_56968_cast_fp16 = softmax(axis = var_30355, x = aw_4843_cast_fp16)[name = tensor<string, []>("op_56968_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_56969_cast_fp16 = softmax(axis = var_30355, x = aw_4845_cast_fp16)[name = tensor<string, []>("op_56969_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_56970_cast_fp16 = softmax(axis = var_30355, x = aw_4847_cast_fp16)[name = tensor<string, []>("op_56970_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_56971_cast_fp16 = softmax(axis = var_30355, x = aw_4849_cast_fp16)[name = tensor<string, []>("op_56971_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_56972_cast_fp16 = softmax(axis = var_30355, x = aw_4851_cast_fp16)[name = tensor<string, []>("op_56972_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_56973_cast_fp16 = softmax(axis = var_30355, x = aw_4853_cast_fp16)[name = tensor<string, []>("op_56973_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_56974_cast_fp16 = softmax(axis = var_30355, x = aw_4855_cast_fp16)[name = tensor<string, []>("op_56974_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_56975_cast_fp16 = softmax(axis = var_30355, x = aw_4857_cast_fp16)[name = tensor<string, []>("op_56975_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_56976_cast_fp16 = softmax(axis = var_30355, x = aw_4859_cast_fp16)[name = tensor<string, []>("op_56976_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_56977_cast_fp16 = softmax(axis = var_30355, x = aw_4861_cast_fp16)[name = tensor<string, []>("op_56977_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_56978_cast_fp16 = softmax(axis = var_30355, x = aw_4863_cast_fp16)[name = tensor<string, []>("op_56978_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_56979_cast_fp16 = softmax(axis = var_30355, x = aw_4865_cast_fp16)[name = tensor<string, []>("op_56979_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_56980_cast_fp16 = softmax(axis = var_30355, x = aw_4867_cast_fp16)[name = tensor<string, []>("op_56980_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_56981_cast_fp16 = softmax(axis = var_30355, x = aw_4869_cast_fp16)[name = tensor<string, []>("op_56981_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_56982_cast_fp16 = softmax(axis = var_30355, x = aw_4871_cast_fp16)[name = tensor<string, []>("op_56982_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_56983_cast_fp16 = softmax(axis = var_30355, x = aw_4873_cast_fp16)[name = tensor<string, []>("op_56983_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_56984_cast_fp16 = softmax(axis = var_30355, x = aw_4875_cast_fp16)[name = tensor<string, []>("op_56984_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_56985_cast_fp16 = softmax(axis = var_30355, x = aw_4877_cast_fp16)[name = tensor<string, []>("op_56985_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_56986_cast_fp16 = softmax(axis = var_30355, x = aw_4879_cast_fp16)[name = tensor<string, []>("op_56986_cast_fp16")];
            tensor<string, []> var_56988_equation_0 = const()[name = tensor<string, []>("op_56988_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56988_cast_fp16 = einsum(equation = var_56988_equation_0, values = (var_56808_cast_fp16, var_56967_cast_fp16))[name = tensor<string, []>("op_56988_cast_fp16")];
            tensor<string, []> var_56990_equation_0 = const()[name = tensor<string, []>("op_56990_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56990_cast_fp16 = einsum(equation = var_56990_equation_0, values = (var_56812_cast_fp16, var_56968_cast_fp16))[name = tensor<string, []>("op_56990_cast_fp16")];
            tensor<string, []> var_56992_equation_0 = const()[name = tensor<string, []>("op_56992_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56992_cast_fp16 = einsum(equation = var_56992_equation_0, values = (var_56816_cast_fp16, var_56969_cast_fp16))[name = tensor<string, []>("op_56992_cast_fp16")];
            tensor<string, []> var_56994_equation_0 = const()[name = tensor<string, []>("op_56994_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56994_cast_fp16 = einsum(equation = var_56994_equation_0, values = (var_56820_cast_fp16, var_56970_cast_fp16))[name = tensor<string, []>("op_56994_cast_fp16")];
            tensor<string, []> var_56996_equation_0 = const()[name = tensor<string, []>("op_56996_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56996_cast_fp16 = einsum(equation = var_56996_equation_0, values = (var_56824_cast_fp16, var_56971_cast_fp16))[name = tensor<string, []>("op_56996_cast_fp16")];
            tensor<string, []> var_56998_equation_0 = const()[name = tensor<string, []>("op_56998_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56998_cast_fp16 = einsum(equation = var_56998_equation_0, values = (var_56828_cast_fp16, var_56972_cast_fp16))[name = tensor<string, []>("op_56998_cast_fp16")];
            tensor<string, []> var_57000_equation_0 = const()[name = tensor<string, []>("op_57000_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57000_cast_fp16 = einsum(equation = var_57000_equation_0, values = (var_56832_cast_fp16, var_56973_cast_fp16))[name = tensor<string, []>("op_57000_cast_fp16")];
            tensor<string, []> var_57002_equation_0 = const()[name = tensor<string, []>("op_57002_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57002_cast_fp16 = einsum(equation = var_57002_equation_0, values = (var_56836_cast_fp16, var_56974_cast_fp16))[name = tensor<string, []>("op_57002_cast_fp16")];
            tensor<string, []> var_57004_equation_0 = const()[name = tensor<string, []>("op_57004_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57004_cast_fp16 = einsum(equation = var_57004_equation_0, values = (var_56840_cast_fp16, var_56975_cast_fp16))[name = tensor<string, []>("op_57004_cast_fp16")];
            tensor<string, []> var_57006_equation_0 = const()[name = tensor<string, []>("op_57006_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57006_cast_fp16 = einsum(equation = var_57006_equation_0, values = (var_56844_cast_fp16, var_56976_cast_fp16))[name = tensor<string, []>("op_57006_cast_fp16")];
            tensor<string, []> var_57008_equation_0 = const()[name = tensor<string, []>("op_57008_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57008_cast_fp16 = einsum(equation = var_57008_equation_0, values = (var_56848_cast_fp16, var_56977_cast_fp16))[name = tensor<string, []>("op_57008_cast_fp16")];
            tensor<string, []> var_57010_equation_0 = const()[name = tensor<string, []>("op_57010_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57010_cast_fp16 = einsum(equation = var_57010_equation_0, values = (var_56852_cast_fp16, var_56978_cast_fp16))[name = tensor<string, []>("op_57010_cast_fp16")];
            tensor<string, []> var_57012_equation_0 = const()[name = tensor<string, []>("op_57012_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57012_cast_fp16 = einsum(equation = var_57012_equation_0, values = (var_56856_cast_fp16, var_56979_cast_fp16))[name = tensor<string, []>("op_57012_cast_fp16")];
            tensor<string, []> var_57014_equation_0 = const()[name = tensor<string, []>("op_57014_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57014_cast_fp16 = einsum(equation = var_57014_equation_0, values = (var_56860_cast_fp16, var_56980_cast_fp16))[name = tensor<string, []>("op_57014_cast_fp16")];
            tensor<string, []> var_57016_equation_0 = const()[name = tensor<string, []>("op_57016_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57016_cast_fp16 = einsum(equation = var_57016_equation_0, values = (var_56864_cast_fp16, var_56981_cast_fp16))[name = tensor<string, []>("op_57016_cast_fp16")];
            tensor<string, []> var_57018_equation_0 = const()[name = tensor<string, []>("op_57018_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57018_cast_fp16 = einsum(equation = var_57018_equation_0, values = (var_56868_cast_fp16, var_56982_cast_fp16))[name = tensor<string, []>("op_57018_cast_fp16")];
            tensor<string, []> var_57020_equation_0 = const()[name = tensor<string, []>("op_57020_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57020_cast_fp16 = einsum(equation = var_57020_equation_0, values = (var_56872_cast_fp16, var_56983_cast_fp16))[name = tensor<string, []>("op_57020_cast_fp16")];
            tensor<string, []> var_57022_equation_0 = const()[name = tensor<string, []>("op_57022_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57022_cast_fp16 = einsum(equation = var_57022_equation_0, values = (var_56876_cast_fp16, var_56984_cast_fp16))[name = tensor<string, []>("op_57022_cast_fp16")];
            tensor<string, []> var_57024_equation_0 = const()[name = tensor<string, []>("op_57024_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57024_cast_fp16 = einsum(equation = var_57024_equation_0, values = (var_56880_cast_fp16, var_56985_cast_fp16))[name = tensor<string, []>("op_57024_cast_fp16")];
            tensor<string, []> var_57026_equation_0 = const()[name = tensor<string, []>("op_57026_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57026_cast_fp16 = einsum(equation = var_57026_equation_0, values = (var_56884_cast_fp16, var_56986_cast_fp16))[name = tensor<string, []>("op_57026_cast_fp16")];
            tensor<bool, []> input_701_interleave_0 = const()[name = tensor<string, []>("input_701_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_701_cast_fp16 = concat(axis = var_30355, interleave = input_701_interleave_0, values = (var_56988_cast_fp16, var_56990_cast_fp16, var_56992_cast_fp16, var_56994_cast_fp16, var_56996_cast_fp16, var_56998_cast_fp16, var_57000_cast_fp16, var_57002_cast_fp16, var_57004_cast_fp16, var_57006_cast_fp16, var_57008_cast_fp16, var_57010_cast_fp16, var_57012_cast_fp16, var_57014_cast_fp16, var_57016_cast_fp16, var_57018_cast_fp16, var_57020_cast_fp16, var_57022_cast_fp16, var_57024_cast_fp16, var_57026_cast_fp16))[name = tensor<string, []>("input_701_cast_fp16")];
            tensor<string, []> var_57036_pad_type_0 = const()[name = tensor<string, []>("op_57036_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_57036_strides_0 = const()[name = tensor<string, []>("op_57036_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_57036_pad_0 = const()[name = tensor<string, []>("op_57036_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_57036_dilations_0 = const()[name = tensor<string, []>("op_57036_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_57036_groups_0 = const()[name = tensor<string, []>("op_57036_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_8_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(817112832))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(818341696))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_8_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_8_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_8_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(818341888)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_57036_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_8_attn2_to_out_0_bias_to_fp16, dilations = var_57036_dilations_0, groups = var_57036_groups_0, pad = var_57036_pad_0, pad_type = var_57036_pad_type_0, strides = var_57036_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_8_attn2_to_out_0_weight_to_fp16_palettized, x = input_701_cast_fp16)[name = tensor<string, []>("op_57036_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_377_cast_fp16 = add(x = var_57036_cast_fp16, y = inputs_375_cast_fp16)[name = tensor<string, []>("inputs_377_cast_fp16")];
            tensor<int32, [1]> input_703_axes_0 = const()[name = tensor<string, []>("input_703_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> input_703_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_703_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(818344512)))];
            tensor<fp16, [1280]> input_703_beta_0_to_fp16 = const()[name = tensor<string, []>("input_703_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(818347136)))];
            tensor<fp16, []> var_57046_to_fp16 = const()[name = tensor<string, []>("op_57046_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> input_703_cast_fp16 = layer_norm(axes = input_703_axes_0, beta = input_703_beta_0_to_fp16, epsilon = var_57046_to_fp16, gamma = input_703_gamma_0_to_fp16, x = inputs_377_cast_fp16)[name = tensor<string, []>("input_703_cast_fp16")];
            tensor<string, []> var_57066_pad_type_0 = const()[name = tensor<string, []>("op_57066_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_57066_strides_0 = const()[name = tensor<string, []>("op_57066_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_57066_pad_0 = const()[name = tensor<string, []>("op_57066_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_57066_dilations_0 = const()[name = tensor<string, []>("op_57066_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_57066_groups_0 = const()[name = tensor<string, []>("op_57066_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_8_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [9830400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(818349760))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(828180224))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_8_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([10240, 1280, 1, 1])];
            tensor<fp16, [10240]> up_blocks_0_attentions_2_transformer_blocks_8_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_8_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(828180416)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_57066_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_8_ff_net_0_proj_bias_to_fp16, dilations = var_57066_dilations_0, groups = var_57066_groups_0, pad = var_57066_pad_0, pad_type = var_57066_pad_type_0, strides = var_57066_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_8_ff_net_0_proj_weight_to_fp16_palettized, x = input_703_cast_fp16)[name = tensor<string, []>("op_57066_cast_fp16")];
            tensor<int32, [2]> var_57067_split_sizes_0 = const()[name = tensor<string, []>("op_57067_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_57067_axis_0 = const()[name = tensor<string, []>("op_57067_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_57067_cast_fp16_0, tensor<fp16, [2, 5120, 1, 1024]> var_57067_cast_fp16_1 = split(axis = var_57067_axis_0, split_sizes = var_57067_split_sizes_0, x = var_57066_cast_fp16)[name = tensor<string, []>("op_57067_cast_fp16")];
            tensor<string, []> var_57069_mode_0 = const()[name = tensor<string, []>("op_57069_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_57069_cast_fp16 = gelu(mode = var_57069_mode_0, x = var_57067_cast_fp16_1)[name = tensor<string, []>("op_57069_cast_fp16")];
            tensor<fp16, [2, 5120, 1, 1024]> input_705_cast_fp16 = mul(x = var_57067_cast_fp16_0, y = var_57069_cast_fp16)[name = tensor<string, []>("input_705_cast_fp16")];
            tensor<string, []> var_57077_pad_type_0 = const()[name = tensor<string, []>("op_57077_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_57077_strides_0 = const()[name = tensor<string, []>("op_57077_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_57077_pad_0 = const()[name = tensor<string, []>("op_57077_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_57077_dilations_0 = const()[name = tensor<string, []>("op_57077_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_57077_groups_0 = const()[name = tensor<string, []>("op_57077_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_8_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [4915200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(828200960))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(833116224))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_8_ff_net_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 5120, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_8_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_8_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(833116416)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_57077_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_8_ff_net_2_bias_to_fp16, dilations = var_57077_dilations_0, groups = var_57077_groups_0, pad = var_57077_pad_0, pad_type = var_57077_pad_type_0, strides = var_57077_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_8_ff_net_2_weight_to_fp16_palettized, x = input_705_cast_fp16)[name = tensor<string, []>("op_57077_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_379_cast_fp16 = add(x = var_57077_cast_fp16, y = inputs_377_cast_fp16)[name = tensor<string, []>("inputs_379_cast_fp16")];
            tensor<int32, [1]> hidden_states_489_axes_0 = const()[name = tensor<string, []>("hidden_states_489_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_489_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_489_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(833119040)))];
            tensor<fp16, [1280]> hidden_states_489_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_489_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(833121664)))];
            tensor<fp16, []> var_57093_to_fp16 = const()[name = tensor<string, []>("op_57093_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_489_cast_fp16 = layer_norm(axes = hidden_states_489_axes_0, beta = hidden_states_489_beta_0_to_fp16, epsilon = var_57093_to_fp16, gamma = hidden_states_489_gamma_0_to_fp16, x = inputs_379_cast_fp16)[name = tensor<string, []>("hidden_states_489_cast_fp16")];
            tensor<string, []> q_253_pad_type_0 = const()[name = tensor<string, []>("q_253_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_253_strides_0 = const()[name = tensor<string, []>("q_253_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_253_pad_0 = const()[name = tensor<string, []>("q_253_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_253_dilations_0 = const()[name = tensor<string, []>("q_253_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_253_groups_0 = const()[name = tensor<string, []>("q_253_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_9_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(833124288))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(834353152))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_9_attn1_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_253_cast_fp16 = conv(dilations = q_253_dilations_0, groups = q_253_groups_0, pad = q_253_pad_0, pad_type = q_253_pad_type_0, strides = q_253_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_9_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_489_cast_fp16)[name = tensor<string, []>("q_253_cast_fp16")];
            tensor<string, []> k_505_pad_type_0 = const()[name = tensor<string, []>("k_505_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_505_strides_0 = const()[name = tensor<string, []>("k_505_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_505_pad_0 = const()[name = tensor<string, []>("k_505_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_505_dilations_0 = const()[name = tensor<string, []>("k_505_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_505_groups_0 = const()[name = tensor<string, []>("k_505_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_9_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(834353344))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(835582208))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_9_attn1_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> k_505_cast_fp16 = conv(dilations = k_505_dilations_0, groups = k_505_groups_0, pad = k_505_pad_0, pad_type = k_505_pad_type_0, strides = k_505_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_9_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_489_cast_fp16)[name = tensor<string, []>("k_505_cast_fp16")];
            tensor<string, []> v_253_pad_type_0 = const()[name = tensor<string, []>("v_253_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_253_strides_0 = const()[name = tensor<string, []>("v_253_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_253_pad_0 = const()[name = tensor<string, []>("v_253_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_253_dilations_0 = const()[name = tensor<string, []>("v_253_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_253_groups_0 = const()[name = tensor<string, []>("v_253_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_9_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(835582400))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(836811264))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_9_attn1_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> v_253_cast_fp16 = conv(dilations = v_253_dilations_0, groups = v_253_groups_0, pad = v_253_pad_0, pad_type = v_253_pad_type_0, strides = v_253_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_9_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_489_cast_fp16)[name = tensor<string, []>("v_253_cast_fp16")];
            tensor<int32, [4]> var_57126_begin_0 = const()[name = tensor<string, []>("op_57126_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_57126_end_0 = const()[name = tensor<string, []>("op_57126_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_57126_end_mask_0 = const()[name = tensor<string, []>("op_57126_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57126_cast_fp16 = slice_by_index(begin = var_57126_begin_0, end = var_57126_end_0, end_mask = var_57126_end_mask_0, x = q_253_cast_fp16)[name = tensor<string, []>("op_57126_cast_fp16")];
            tensor<int32, [4]> var_57130_begin_0 = const()[name = tensor<string, []>("op_57130_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_57130_end_0 = const()[name = tensor<string, []>("op_57130_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_57130_end_mask_0 = const()[name = tensor<string, []>("op_57130_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57130_cast_fp16 = slice_by_index(begin = var_57130_begin_0, end = var_57130_end_0, end_mask = var_57130_end_mask_0, x = q_253_cast_fp16)[name = tensor<string, []>("op_57130_cast_fp16")];
            tensor<int32, [4]> var_57134_begin_0 = const()[name = tensor<string, []>("op_57134_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_57134_end_0 = const()[name = tensor<string, []>("op_57134_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_57134_end_mask_0 = const()[name = tensor<string, []>("op_57134_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57134_cast_fp16 = slice_by_index(begin = var_57134_begin_0, end = var_57134_end_0, end_mask = var_57134_end_mask_0, x = q_253_cast_fp16)[name = tensor<string, []>("op_57134_cast_fp16")];
            tensor<int32, [4]> var_57138_begin_0 = const()[name = tensor<string, []>("op_57138_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_57138_end_0 = const()[name = tensor<string, []>("op_57138_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_57138_end_mask_0 = const()[name = tensor<string, []>("op_57138_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57138_cast_fp16 = slice_by_index(begin = var_57138_begin_0, end = var_57138_end_0, end_mask = var_57138_end_mask_0, x = q_253_cast_fp16)[name = tensor<string, []>("op_57138_cast_fp16")];
            tensor<int32, [4]> var_57142_begin_0 = const()[name = tensor<string, []>("op_57142_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_57142_end_0 = const()[name = tensor<string, []>("op_57142_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_57142_end_mask_0 = const()[name = tensor<string, []>("op_57142_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57142_cast_fp16 = slice_by_index(begin = var_57142_begin_0, end = var_57142_end_0, end_mask = var_57142_end_mask_0, x = q_253_cast_fp16)[name = tensor<string, []>("op_57142_cast_fp16")];
            tensor<int32, [4]> var_57146_begin_0 = const()[name = tensor<string, []>("op_57146_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_57146_end_0 = const()[name = tensor<string, []>("op_57146_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_57146_end_mask_0 = const()[name = tensor<string, []>("op_57146_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57146_cast_fp16 = slice_by_index(begin = var_57146_begin_0, end = var_57146_end_0, end_mask = var_57146_end_mask_0, x = q_253_cast_fp16)[name = tensor<string, []>("op_57146_cast_fp16")];
            tensor<int32, [4]> var_57150_begin_0 = const()[name = tensor<string, []>("op_57150_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_57150_end_0 = const()[name = tensor<string, []>("op_57150_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_57150_end_mask_0 = const()[name = tensor<string, []>("op_57150_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57150_cast_fp16 = slice_by_index(begin = var_57150_begin_0, end = var_57150_end_0, end_mask = var_57150_end_mask_0, x = q_253_cast_fp16)[name = tensor<string, []>("op_57150_cast_fp16")];
            tensor<int32, [4]> var_57154_begin_0 = const()[name = tensor<string, []>("op_57154_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_57154_end_0 = const()[name = tensor<string, []>("op_57154_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_57154_end_mask_0 = const()[name = tensor<string, []>("op_57154_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57154_cast_fp16 = slice_by_index(begin = var_57154_begin_0, end = var_57154_end_0, end_mask = var_57154_end_mask_0, x = q_253_cast_fp16)[name = tensor<string, []>("op_57154_cast_fp16")];
            tensor<int32, [4]> var_57158_begin_0 = const()[name = tensor<string, []>("op_57158_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_57158_end_0 = const()[name = tensor<string, []>("op_57158_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_57158_end_mask_0 = const()[name = tensor<string, []>("op_57158_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57158_cast_fp16 = slice_by_index(begin = var_57158_begin_0, end = var_57158_end_0, end_mask = var_57158_end_mask_0, x = q_253_cast_fp16)[name = tensor<string, []>("op_57158_cast_fp16")];
            tensor<int32, [4]> var_57162_begin_0 = const()[name = tensor<string, []>("op_57162_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_57162_end_0 = const()[name = tensor<string, []>("op_57162_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_57162_end_mask_0 = const()[name = tensor<string, []>("op_57162_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57162_cast_fp16 = slice_by_index(begin = var_57162_begin_0, end = var_57162_end_0, end_mask = var_57162_end_mask_0, x = q_253_cast_fp16)[name = tensor<string, []>("op_57162_cast_fp16")];
            tensor<int32, [4]> var_57166_begin_0 = const()[name = tensor<string, []>("op_57166_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_57166_end_0 = const()[name = tensor<string, []>("op_57166_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_57166_end_mask_0 = const()[name = tensor<string, []>("op_57166_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57166_cast_fp16 = slice_by_index(begin = var_57166_begin_0, end = var_57166_end_0, end_mask = var_57166_end_mask_0, x = q_253_cast_fp16)[name = tensor<string, []>("op_57166_cast_fp16")];
            tensor<int32, [4]> var_57170_begin_0 = const()[name = tensor<string, []>("op_57170_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_57170_end_0 = const()[name = tensor<string, []>("op_57170_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_57170_end_mask_0 = const()[name = tensor<string, []>("op_57170_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57170_cast_fp16 = slice_by_index(begin = var_57170_begin_0, end = var_57170_end_0, end_mask = var_57170_end_mask_0, x = q_253_cast_fp16)[name = tensor<string, []>("op_57170_cast_fp16")];
            tensor<int32, [4]> var_57174_begin_0 = const()[name = tensor<string, []>("op_57174_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_57174_end_0 = const()[name = tensor<string, []>("op_57174_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_57174_end_mask_0 = const()[name = tensor<string, []>("op_57174_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57174_cast_fp16 = slice_by_index(begin = var_57174_begin_0, end = var_57174_end_0, end_mask = var_57174_end_mask_0, x = q_253_cast_fp16)[name = tensor<string, []>("op_57174_cast_fp16")];
            tensor<int32, [4]> var_57178_begin_0 = const()[name = tensor<string, []>("op_57178_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_57178_end_0 = const()[name = tensor<string, []>("op_57178_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_57178_end_mask_0 = const()[name = tensor<string, []>("op_57178_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57178_cast_fp16 = slice_by_index(begin = var_57178_begin_0, end = var_57178_end_0, end_mask = var_57178_end_mask_0, x = q_253_cast_fp16)[name = tensor<string, []>("op_57178_cast_fp16")];
            tensor<int32, [4]> var_57182_begin_0 = const()[name = tensor<string, []>("op_57182_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_57182_end_0 = const()[name = tensor<string, []>("op_57182_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_57182_end_mask_0 = const()[name = tensor<string, []>("op_57182_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57182_cast_fp16 = slice_by_index(begin = var_57182_begin_0, end = var_57182_end_0, end_mask = var_57182_end_mask_0, x = q_253_cast_fp16)[name = tensor<string, []>("op_57182_cast_fp16")];
            tensor<int32, [4]> var_57186_begin_0 = const()[name = tensor<string, []>("op_57186_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_57186_end_0 = const()[name = tensor<string, []>("op_57186_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_57186_end_mask_0 = const()[name = tensor<string, []>("op_57186_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57186_cast_fp16 = slice_by_index(begin = var_57186_begin_0, end = var_57186_end_0, end_mask = var_57186_end_mask_0, x = q_253_cast_fp16)[name = tensor<string, []>("op_57186_cast_fp16")];
            tensor<int32, [4]> var_57190_begin_0 = const()[name = tensor<string, []>("op_57190_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_57190_end_0 = const()[name = tensor<string, []>("op_57190_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_57190_end_mask_0 = const()[name = tensor<string, []>("op_57190_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57190_cast_fp16 = slice_by_index(begin = var_57190_begin_0, end = var_57190_end_0, end_mask = var_57190_end_mask_0, x = q_253_cast_fp16)[name = tensor<string, []>("op_57190_cast_fp16")];
            tensor<int32, [4]> var_57194_begin_0 = const()[name = tensor<string, []>("op_57194_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_57194_end_0 = const()[name = tensor<string, []>("op_57194_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_57194_end_mask_0 = const()[name = tensor<string, []>("op_57194_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57194_cast_fp16 = slice_by_index(begin = var_57194_begin_0, end = var_57194_end_0, end_mask = var_57194_end_mask_0, x = q_253_cast_fp16)[name = tensor<string, []>("op_57194_cast_fp16")];
            tensor<int32, [4]> var_57198_begin_0 = const()[name = tensor<string, []>("op_57198_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_57198_end_0 = const()[name = tensor<string, []>("op_57198_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_57198_end_mask_0 = const()[name = tensor<string, []>("op_57198_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57198_cast_fp16 = slice_by_index(begin = var_57198_begin_0, end = var_57198_end_0, end_mask = var_57198_end_mask_0, x = q_253_cast_fp16)[name = tensor<string, []>("op_57198_cast_fp16")];
            tensor<int32, [4]> var_57202_begin_0 = const()[name = tensor<string, []>("op_57202_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_57202_end_0 = const()[name = tensor<string, []>("op_57202_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_57202_end_mask_0 = const()[name = tensor<string, []>("op_57202_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57202_cast_fp16 = slice_by_index(begin = var_57202_begin_0, end = var_57202_end_0, end_mask = var_57202_end_mask_0, x = q_253_cast_fp16)[name = tensor<string, []>("op_57202_cast_fp16")];
            tensor<int32, [4]> k_507_perm_0 = const()[name = tensor<string, []>("k_507_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_57209_begin_0 = const()[name = tensor<string, []>("op_57209_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_57209_end_0 = const()[name = tensor<string, []>("op_57209_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_57209_end_mask_0 = const()[name = tensor<string, []>("op_57209_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> k_507_cast_fp16 = transpose(perm = k_507_perm_0, x = k_505_cast_fp16)[name = tensor<string, []>("transpose_13")];
            tensor<fp16, [2, 1024, 1, 64]> var_57209_cast_fp16 = slice_by_index(begin = var_57209_begin_0, end = var_57209_end_0, end_mask = var_57209_end_mask_0, x = k_507_cast_fp16)[name = tensor<string, []>("op_57209_cast_fp16")];
            tensor<int32, [4]> var_57213_begin_0 = const()[name = tensor<string, []>("op_57213_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_57213_end_0 = const()[name = tensor<string, []>("op_57213_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_57213_end_mask_0 = const()[name = tensor<string, []>("op_57213_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_57213_cast_fp16 = slice_by_index(begin = var_57213_begin_0, end = var_57213_end_0, end_mask = var_57213_end_mask_0, x = k_507_cast_fp16)[name = tensor<string, []>("op_57213_cast_fp16")];
            tensor<int32, [4]> var_57217_begin_0 = const()[name = tensor<string, []>("op_57217_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_57217_end_0 = const()[name = tensor<string, []>("op_57217_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_57217_end_mask_0 = const()[name = tensor<string, []>("op_57217_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_57217_cast_fp16 = slice_by_index(begin = var_57217_begin_0, end = var_57217_end_0, end_mask = var_57217_end_mask_0, x = k_507_cast_fp16)[name = tensor<string, []>("op_57217_cast_fp16")];
            tensor<int32, [4]> var_57221_begin_0 = const()[name = tensor<string, []>("op_57221_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_57221_end_0 = const()[name = tensor<string, []>("op_57221_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_57221_end_mask_0 = const()[name = tensor<string, []>("op_57221_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_57221_cast_fp16 = slice_by_index(begin = var_57221_begin_0, end = var_57221_end_0, end_mask = var_57221_end_mask_0, x = k_507_cast_fp16)[name = tensor<string, []>("op_57221_cast_fp16")];
            tensor<int32, [4]> var_57225_begin_0 = const()[name = tensor<string, []>("op_57225_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_57225_end_0 = const()[name = tensor<string, []>("op_57225_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_57225_end_mask_0 = const()[name = tensor<string, []>("op_57225_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_57225_cast_fp16 = slice_by_index(begin = var_57225_begin_0, end = var_57225_end_0, end_mask = var_57225_end_mask_0, x = k_507_cast_fp16)[name = tensor<string, []>("op_57225_cast_fp16")];
            tensor<int32, [4]> var_57229_begin_0 = const()[name = tensor<string, []>("op_57229_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_57229_end_0 = const()[name = tensor<string, []>("op_57229_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_57229_end_mask_0 = const()[name = tensor<string, []>("op_57229_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_57229_cast_fp16 = slice_by_index(begin = var_57229_begin_0, end = var_57229_end_0, end_mask = var_57229_end_mask_0, x = k_507_cast_fp16)[name = tensor<string, []>("op_57229_cast_fp16")];
            tensor<int32, [4]> var_57233_begin_0 = const()[name = tensor<string, []>("op_57233_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_57233_end_0 = const()[name = tensor<string, []>("op_57233_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_57233_end_mask_0 = const()[name = tensor<string, []>("op_57233_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_57233_cast_fp16 = slice_by_index(begin = var_57233_begin_0, end = var_57233_end_0, end_mask = var_57233_end_mask_0, x = k_507_cast_fp16)[name = tensor<string, []>("op_57233_cast_fp16")];
            tensor<int32, [4]> var_57237_begin_0 = const()[name = tensor<string, []>("op_57237_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_57237_end_0 = const()[name = tensor<string, []>("op_57237_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_57237_end_mask_0 = const()[name = tensor<string, []>("op_57237_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_57237_cast_fp16 = slice_by_index(begin = var_57237_begin_0, end = var_57237_end_0, end_mask = var_57237_end_mask_0, x = k_507_cast_fp16)[name = tensor<string, []>("op_57237_cast_fp16")];
            tensor<int32, [4]> var_57241_begin_0 = const()[name = tensor<string, []>("op_57241_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_57241_end_0 = const()[name = tensor<string, []>("op_57241_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_57241_end_mask_0 = const()[name = tensor<string, []>("op_57241_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_57241_cast_fp16 = slice_by_index(begin = var_57241_begin_0, end = var_57241_end_0, end_mask = var_57241_end_mask_0, x = k_507_cast_fp16)[name = tensor<string, []>("op_57241_cast_fp16")];
            tensor<int32, [4]> var_57245_begin_0 = const()[name = tensor<string, []>("op_57245_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_57245_end_0 = const()[name = tensor<string, []>("op_57245_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_57245_end_mask_0 = const()[name = tensor<string, []>("op_57245_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_57245_cast_fp16 = slice_by_index(begin = var_57245_begin_0, end = var_57245_end_0, end_mask = var_57245_end_mask_0, x = k_507_cast_fp16)[name = tensor<string, []>("op_57245_cast_fp16")];
            tensor<int32, [4]> var_57249_begin_0 = const()[name = tensor<string, []>("op_57249_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_57249_end_0 = const()[name = tensor<string, []>("op_57249_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_57249_end_mask_0 = const()[name = tensor<string, []>("op_57249_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_57249_cast_fp16 = slice_by_index(begin = var_57249_begin_0, end = var_57249_end_0, end_mask = var_57249_end_mask_0, x = k_507_cast_fp16)[name = tensor<string, []>("op_57249_cast_fp16")];
            tensor<int32, [4]> var_57253_begin_0 = const()[name = tensor<string, []>("op_57253_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_57253_end_0 = const()[name = tensor<string, []>("op_57253_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_57253_end_mask_0 = const()[name = tensor<string, []>("op_57253_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_57253_cast_fp16 = slice_by_index(begin = var_57253_begin_0, end = var_57253_end_0, end_mask = var_57253_end_mask_0, x = k_507_cast_fp16)[name = tensor<string, []>("op_57253_cast_fp16")];
            tensor<int32, [4]> var_57257_begin_0 = const()[name = tensor<string, []>("op_57257_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_57257_end_0 = const()[name = tensor<string, []>("op_57257_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_57257_end_mask_0 = const()[name = tensor<string, []>("op_57257_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_57257_cast_fp16 = slice_by_index(begin = var_57257_begin_0, end = var_57257_end_0, end_mask = var_57257_end_mask_0, x = k_507_cast_fp16)[name = tensor<string, []>("op_57257_cast_fp16")];
            tensor<int32, [4]> var_57261_begin_0 = const()[name = tensor<string, []>("op_57261_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_57261_end_0 = const()[name = tensor<string, []>("op_57261_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_57261_end_mask_0 = const()[name = tensor<string, []>("op_57261_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_57261_cast_fp16 = slice_by_index(begin = var_57261_begin_0, end = var_57261_end_0, end_mask = var_57261_end_mask_0, x = k_507_cast_fp16)[name = tensor<string, []>("op_57261_cast_fp16")];
            tensor<int32, [4]> var_57265_begin_0 = const()[name = tensor<string, []>("op_57265_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_57265_end_0 = const()[name = tensor<string, []>("op_57265_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_57265_end_mask_0 = const()[name = tensor<string, []>("op_57265_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_57265_cast_fp16 = slice_by_index(begin = var_57265_begin_0, end = var_57265_end_0, end_mask = var_57265_end_mask_0, x = k_507_cast_fp16)[name = tensor<string, []>("op_57265_cast_fp16")];
            tensor<int32, [4]> var_57269_begin_0 = const()[name = tensor<string, []>("op_57269_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_57269_end_0 = const()[name = tensor<string, []>("op_57269_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_57269_end_mask_0 = const()[name = tensor<string, []>("op_57269_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_57269_cast_fp16 = slice_by_index(begin = var_57269_begin_0, end = var_57269_end_0, end_mask = var_57269_end_mask_0, x = k_507_cast_fp16)[name = tensor<string, []>("op_57269_cast_fp16")];
            tensor<int32, [4]> var_57273_begin_0 = const()[name = tensor<string, []>("op_57273_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_57273_end_0 = const()[name = tensor<string, []>("op_57273_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_57273_end_mask_0 = const()[name = tensor<string, []>("op_57273_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_57273_cast_fp16 = slice_by_index(begin = var_57273_begin_0, end = var_57273_end_0, end_mask = var_57273_end_mask_0, x = k_507_cast_fp16)[name = tensor<string, []>("op_57273_cast_fp16")];
            tensor<int32, [4]> var_57277_begin_0 = const()[name = tensor<string, []>("op_57277_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_57277_end_0 = const()[name = tensor<string, []>("op_57277_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_57277_end_mask_0 = const()[name = tensor<string, []>("op_57277_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_57277_cast_fp16 = slice_by_index(begin = var_57277_begin_0, end = var_57277_end_0, end_mask = var_57277_end_mask_0, x = k_507_cast_fp16)[name = tensor<string, []>("op_57277_cast_fp16")];
            tensor<int32, [4]> var_57281_begin_0 = const()[name = tensor<string, []>("op_57281_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_57281_end_0 = const()[name = tensor<string, []>("op_57281_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_57281_end_mask_0 = const()[name = tensor<string, []>("op_57281_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_57281_cast_fp16 = slice_by_index(begin = var_57281_begin_0, end = var_57281_end_0, end_mask = var_57281_end_mask_0, x = k_507_cast_fp16)[name = tensor<string, []>("op_57281_cast_fp16")];
            tensor<int32, [4]> var_57285_begin_0 = const()[name = tensor<string, []>("op_57285_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_57285_end_0 = const()[name = tensor<string, []>("op_57285_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_57285_end_mask_0 = const()[name = tensor<string, []>("op_57285_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_57285_cast_fp16 = slice_by_index(begin = var_57285_begin_0, end = var_57285_end_0, end_mask = var_57285_end_mask_0, x = k_507_cast_fp16)[name = tensor<string, []>("op_57285_cast_fp16")];
            tensor<int32, [4]> var_57287_begin_0 = const()[name = tensor<string, []>("op_57287_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_57287_end_0 = const()[name = tensor<string, []>("op_57287_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_57287_end_mask_0 = const()[name = tensor<string, []>("op_57287_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57287_cast_fp16 = slice_by_index(begin = var_57287_begin_0, end = var_57287_end_0, end_mask = var_57287_end_mask_0, x = v_253_cast_fp16)[name = tensor<string, []>("op_57287_cast_fp16")];
            tensor<int32, [4]> var_57291_begin_0 = const()[name = tensor<string, []>("op_57291_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_57291_end_0 = const()[name = tensor<string, []>("op_57291_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_57291_end_mask_0 = const()[name = tensor<string, []>("op_57291_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57291_cast_fp16 = slice_by_index(begin = var_57291_begin_0, end = var_57291_end_0, end_mask = var_57291_end_mask_0, x = v_253_cast_fp16)[name = tensor<string, []>("op_57291_cast_fp16")];
            tensor<int32, [4]> var_57295_begin_0 = const()[name = tensor<string, []>("op_57295_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_57295_end_0 = const()[name = tensor<string, []>("op_57295_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_57295_end_mask_0 = const()[name = tensor<string, []>("op_57295_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57295_cast_fp16 = slice_by_index(begin = var_57295_begin_0, end = var_57295_end_0, end_mask = var_57295_end_mask_0, x = v_253_cast_fp16)[name = tensor<string, []>("op_57295_cast_fp16")];
            tensor<int32, [4]> var_57299_begin_0 = const()[name = tensor<string, []>("op_57299_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_57299_end_0 = const()[name = tensor<string, []>("op_57299_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_57299_end_mask_0 = const()[name = tensor<string, []>("op_57299_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57299_cast_fp16 = slice_by_index(begin = var_57299_begin_0, end = var_57299_end_0, end_mask = var_57299_end_mask_0, x = v_253_cast_fp16)[name = tensor<string, []>("op_57299_cast_fp16")];
            tensor<int32, [4]> var_57303_begin_0 = const()[name = tensor<string, []>("op_57303_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_57303_end_0 = const()[name = tensor<string, []>("op_57303_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_57303_end_mask_0 = const()[name = tensor<string, []>("op_57303_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57303_cast_fp16 = slice_by_index(begin = var_57303_begin_0, end = var_57303_end_0, end_mask = var_57303_end_mask_0, x = v_253_cast_fp16)[name = tensor<string, []>("op_57303_cast_fp16")];
            tensor<int32, [4]> var_57307_begin_0 = const()[name = tensor<string, []>("op_57307_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_57307_end_0 = const()[name = tensor<string, []>("op_57307_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_57307_end_mask_0 = const()[name = tensor<string, []>("op_57307_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57307_cast_fp16 = slice_by_index(begin = var_57307_begin_0, end = var_57307_end_0, end_mask = var_57307_end_mask_0, x = v_253_cast_fp16)[name = tensor<string, []>("op_57307_cast_fp16")];
            tensor<int32, [4]> var_57311_begin_0 = const()[name = tensor<string, []>("op_57311_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_57311_end_0 = const()[name = tensor<string, []>("op_57311_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_57311_end_mask_0 = const()[name = tensor<string, []>("op_57311_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57311_cast_fp16 = slice_by_index(begin = var_57311_begin_0, end = var_57311_end_0, end_mask = var_57311_end_mask_0, x = v_253_cast_fp16)[name = tensor<string, []>("op_57311_cast_fp16")];
            tensor<int32, [4]> var_57315_begin_0 = const()[name = tensor<string, []>("op_57315_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_57315_end_0 = const()[name = tensor<string, []>("op_57315_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_57315_end_mask_0 = const()[name = tensor<string, []>("op_57315_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57315_cast_fp16 = slice_by_index(begin = var_57315_begin_0, end = var_57315_end_0, end_mask = var_57315_end_mask_0, x = v_253_cast_fp16)[name = tensor<string, []>("op_57315_cast_fp16")];
            tensor<int32, [4]> var_57319_begin_0 = const()[name = tensor<string, []>("op_57319_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_57319_end_0 = const()[name = tensor<string, []>("op_57319_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_57319_end_mask_0 = const()[name = tensor<string, []>("op_57319_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57319_cast_fp16 = slice_by_index(begin = var_57319_begin_0, end = var_57319_end_0, end_mask = var_57319_end_mask_0, x = v_253_cast_fp16)[name = tensor<string, []>("op_57319_cast_fp16")];
            tensor<int32, [4]> var_57323_begin_0 = const()[name = tensor<string, []>("op_57323_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_57323_end_0 = const()[name = tensor<string, []>("op_57323_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_57323_end_mask_0 = const()[name = tensor<string, []>("op_57323_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57323_cast_fp16 = slice_by_index(begin = var_57323_begin_0, end = var_57323_end_0, end_mask = var_57323_end_mask_0, x = v_253_cast_fp16)[name = tensor<string, []>("op_57323_cast_fp16")];
            tensor<int32, [4]> var_57327_begin_0 = const()[name = tensor<string, []>("op_57327_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_57327_end_0 = const()[name = tensor<string, []>("op_57327_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_57327_end_mask_0 = const()[name = tensor<string, []>("op_57327_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57327_cast_fp16 = slice_by_index(begin = var_57327_begin_0, end = var_57327_end_0, end_mask = var_57327_end_mask_0, x = v_253_cast_fp16)[name = tensor<string, []>("op_57327_cast_fp16")];
            tensor<int32, [4]> var_57331_begin_0 = const()[name = tensor<string, []>("op_57331_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_57331_end_0 = const()[name = tensor<string, []>("op_57331_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_57331_end_mask_0 = const()[name = tensor<string, []>("op_57331_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57331_cast_fp16 = slice_by_index(begin = var_57331_begin_0, end = var_57331_end_0, end_mask = var_57331_end_mask_0, x = v_253_cast_fp16)[name = tensor<string, []>("op_57331_cast_fp16")];
            tensor<int32, [4]> var_57335_begin_0 = const()[name = tensor<string, []>("op_57335_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_57335_end_0 = const()[name = tensor<string, []>("op_57335_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_57335_end_mask_0 = const()[name = tensor<string, []>("op_57335_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57335_cast_fp16 = slice_by_index(begin = var_57335_begin_0, end = var_57335_end_0, end_mask = var_57335_end_mask_0, x = v_253_cast_fp16)[name = tensor<string, []>("op_57335_cast_fp16")];
            tensor<int32, [4]> var_57339_begin_0 = const()[name = tensor<string, []>("op_57339_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_57339_end_0 = const()[name = tensor<string, []>("op_57339_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_57339_end_mask_0 = const()[name = tensor<string, []>("op_57339_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57339_cast_fp16 = slice_by_index(begin = var_57339_begin_0, end = var_57339_end_0, end_mask = var_57339_end_mask_0, x = v_253_cast_fp16)[name = tensor<string, []>("op_57339_cast_fp16")];
            tensor<int32, [4]> var_57343_begin_0 = const()[name = tensor<string, []>("op_57343_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_57343_end_0 = const()[name = tensor<string, []>("op_57343_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_57343_end_mask_0 = const()[name = tensor<string, []>("op_57343_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57343_cast_fp16 = slice_by_index(begin = var_57343_begin_0, end = var_57343_end_0, end_mask = var_57343_end_mask_0, x = v_253_cast_fp16)[name = tensor<string, []>("op_57343_cast_fp16")];
            tensor<int32, [4]> var_57347_begin_0 = const()[name = tensor<string, []>("op_57347_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_57347_end_0 = const()[name = tensor<string, []>("op_57347_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_57347_end_mask_0 = const()[name = tensor<string, []>("op_57347_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57347_cast_fp16 = slice_by_index(begin = var_57347_begin_0, end = var_57347_end_0, end_mask = var_57347_end_mask_0, x = v_253_cast_fp16)[name = tensor<string, []>("op_57347_cast_fp16")];
            tensor<int32, [4]> var_57351_begin_0 = const()[name = tensor<string, []>("op_57351_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_57351_end_0 = const()[name = tensor<string, []>("op_57351_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_57351_end_mask_0 = const()[name = tensor<string, []>("op_57351_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57351_cast_fp16 = slice_by_index(begin = var_57351_begin_0, end = var_57351_end_0, end_mask = var_57351_end_mask_0, x = v_253_cast_fp16)[name = tensor<string, []>("op_57351_cast_fp16")];
            tensor<int32, [4]> var_57355_begin_0 = const()[name = tensor<string, []>("op_57355_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_57355_end_0 = const()[name = tensor<string, []>("op_57355_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_57355_end_mask_0 = const()[name = tensor<string, []>("op_57355_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57355_cast_fp16 = slice_by_index(begin = var_57355_begin_0, end = var_57355_end_0, end_mask = var_57355_end_mask_0, x = v_253_cast_fp16)[name = tensor<string, []>("op_57355_cast_fp16")];
            tensor<int32, [4]> var_57359_begin_0 = const()[name = tensor<string, []>("op_57359_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_57359_end_0 = const()[name = tensor<string, []>("op_57359_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_57359_end_mask_0 = const()[name = tensor<string, []>("op_57359_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57359_cast_fp16 = slice_by_index(begin = var_57359_begin_0, end = var_57359_end_0, end_mask = var_57359_end_mask_0, x = v_253_cast_fp16)[name = tensor<string, []>("op_57359_cast_fp16")];
            tensor<int32, [4]> var_57363_begin_0 = const()[name = tensor<string, []>("op_57363_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_57363_end_0 = const()[name = tensor<string, []>("op_57363_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_57363_end_mask_0 = const()[name = tensor<string, []>("op_57363_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57363_cast_fp16 = slice_by_index(begin = var_57363_begin_0, end = var_57363_end_0, end_mask = var_57363_end_mask_0, x = v_253_cast_fp16)[name = tensor<string, []>("op_57363_cast_fp16")];
            tensor<string, []> var_57367_equation_0 = const()[name = tensor<string, []>("op_57367_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57367_cast_fp16 = einsum(equation = var_57367_equation_0, values = (var_57209_cast_fp16, var_57126_cast_fp16))[name = tensor<string, []>("op_57367_cast_fp16")];
            tensor<fp16, []> var_57368_to_fp16 = const()[name = tensor<string, []>("op_57368_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4881_cast_fp16 = mul(x = var_57367_cast_fp16, y = var_57368_to_fp16)[name = tensor<string, []>("aw_4881_cast_fp16")];
            tensor<string, []> var_57371_equation_0 = const()[name = tensor<string, []>("op_57371_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57371_cast_fp16 = einsum(equation = var_57371_equation_0, values = (var_57213_cast_fp16, var_57130_cast_fp16))[name = tensor<string, []>("op_57371_cast_fp16")];
            tensor<fp16, []> var_57372_to_fp16 = const()[name = tensor<string, []>("op_57372_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4883_cast_fp16 = mul(x = var_57371_cast_fp16, y = var_57372_to_fp16)[name = tensor<string, []>("aw_4883_cast_fp16")];
            tensor<string, []> var_57375_equation_0 = const()[name = tensor<string, []>("op_57375_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57375_cast_fp16 = einsum(equation = var_57375_equation_0, values = (var_57217_cast_fp16, var_57134_cast_fp16))[name = tensor<string, []>("op_57375_cast_fp16")];
            tensor<fp16, []> var_57376_to_fp16 = const()[name = tensor<string, []>("op_57376_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4885_cast_fp16 = mul(x = var_57375_cast_fp16, y = var_57376_to_fp16)[name = tensor<string, []>("aw_4885_cast_fp16")];
            tensor<string, []> var_57379_equation_0 = const()[name = tensor<string, []>("op_57379_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57379_cast_fp16 = einsum(equation = var_57379_equation_0, values = (var_57221_cast_fp16, var_57138_cast_fp16))[name = tensor<string, []>("op_57379_cast_fp16")];
            tensor<fp16, []> var_57380_to_fp16 = const()[name = tensor<string, []>("op_57380_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4887_cast_fp16 = mul(x = var_57379_cast_fp16, y = var_57380_to_fp16)[name = tensor<string, []>("aw_4887_cast_fp16")];
            tensor<string, []> var_57383_equation_0 = const()[name = tensor<string, []>("op_57383_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57383_cast_fp16 = einsum(equation = var_57383_equation_0, values = (var_57225_cast_fp16, var_57142_cast_fp16))[name = tensor<string, []>("op_57383_cast_fp16")];
            tensor<fp16, []> var_57384_to_fp16 = const()[name = tensor<string, []>("op_57384_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4889_cast_fp16 = mul(x = var_57383_cast_fp16, y = var_57384_to_fp16)[name = tensor<string, []>("aw_4889_cast_fp16")];
            tensor<string, []> var_57387_equation_0 = const()[name = tensor<string, []>("op_57387_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57387_cast_fp16 = einsum(equation = var_57387_equation_0, values = (var_57229_cast_fp16, var_57146_cast_fp16))[name = tensor<string, []>("op_57387_cast_fp16")];
            tensor<fp16, []> var_57388_to_fp16 = const()[name = tensor<string, []>("op_57388_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4891_cast_fp16 = mul(x = var_57387_cast_fp16, y = var_57388_to_fp16)[name = tensor<string, []>("aw_4891_cast_fp16")];
            tensor<string, []> var_57391_equation_0 = const()[name = tensor<string, []>("op_57391_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57391_cast_fp16 = einsum(equation = var_57391_equation_0, values = (var_57233_cast_fp16, var_57150_cast_fp16))[name = tensor<string, []>("op_57391_cast_fp16")];
            tensor<fp16, []> var_57392_to_fp16 = const()[name = tensor<string, []>("op_57392_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4893_cast_fp16 = mul(x = var_57391_cast_fp16, y = var_57392_to_fp16)[name = tensor<string, []>("aw_4893_cast_fp16")];
            tensor<string, []> var_57395_equation_0 = const()[name = tensor<string, []>("op_57395_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57395_cast_fp16 = einsum(equation = var_57395_equation_0, values = (var_57237_cast_fp16, var_57154_cast_fp16))[name = tensor<string, []>("op_57395_cast_fp16")];
            tensor<fp16, []> var_57396_to_fp16 = const()[name = tensor<string, []>("op_57396_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4895_cast_fp16 = mul(x = var_57395_cast_fp16, y = var_57396_to_fp16)[name = tensor<string, []>("aw_4895_cast_fp16")];
            tensor<string, []> var_57399_equation_0 = const()[name = tensor<string, []>("op_57399_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57399_cast_fp16 = einsum(equation = var_57399_equation_0, values = (var_57241_cast_fp16, var_57158_cast_fp16))[name = tensor<string, []>("op_57399_cast_fp16")];
            tensor<fp16, []> var_57400_to_fp16 = const()[name = tensor<string, []>("op_57400_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4897_cast_fp16 = mul(x = var_57399_cast_fp16, y = var_57400_to_fp16)[name = tensor<string, []>("aw_4897_cast_fp16")];
            tensor<string, []> var_57403_equation_0 = const()[name = tensor<string, []>("op_57403_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57403_cast_fp16 = einsum(equation = var_57403_equation_0, values = (var_57245_cast_fp16, var_57162_cast_fp16))[name = tensor<string, []>("op_57403_cast_fp16")];
            tensor<fp16, []> var_57404_to_fp16 = const()[name = tensor<string, []>("op_57404_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4899_cast_fp16 = mul(x = var_57403_cast_fp16, y = var_57404_to_fp16)[name = tensor<string, []>("aw_4899_cast_fp16")];
            tensor<string, []> var_57407_equation_0 = const()[name = tensor<string, []>("op_57407_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57407_cast_fp16 = einsum(equation = var_57407_equation_0, values = (var_57249_cast_fp16, var_57166_cast_fp16))[name = tensor<string, []>("op_57407_cast_fp16")];
            tensor<fp16, []> var_57408_to_fp16 = const()[name = tensor<string, []>("op_57408_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4901_cast_fp16 = mul(x = var_57407_cast_fp16, y = var_57408_to_fp16)[name = tensor<string, []>("aw_4901_cast_fp16")];
            tensor<string, []> var_57411_equation_0 = const()[name = tensor<string, []>("op_57411_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57411_cast_fp16 = einsum(equation = var_57411_equation_0, values = (var_57253_cast_fp16, var_57170_cast_fp16))[name = tensor<string, []>("op_57411_cast_fp16")];
            tensor<fp16, []> var_57412_to_fp16 = const()[name = tensor<string, []>("op_57412_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4903_cast_fp16 = mul(x = var_57411_cast_fp16, y = var_57412_to_fp16)[name = tensor<string, []>("aw_4903_cast_fp16")];
            tensor<string, []> var_57415_equation_0 = const()[name = tensor<string, []>("op_57415_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57415_cast_fp16 = einsum(equation = var_57415_equation_0, values = (var_57257_cast_fp16, var_57174_cast_fp16))[name = tensor<string, []>("op_57415_cast_fp16")];
            tensor<fp16, []> var_57416_to_fp16 = const()[name = tensor<string, []>("op_57416_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4905_cast_fp16 = mul(x = var_57415_cast_fp16, y = var_57416_to_fp16)[name = tensor<string, []>("aw_4905_cast_fp16")];
            tensor<string, []> var_57419_equation_0 = const()[name = tensor<string, []>("op_57419_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57419_cast_fp16 = einsum(equation = var_57419_equation_0, values = (var_57261_cast_fp16, var_57178_cast_fp16))[name = tensor<string, []>("op_57419_cast_fp16")];
            tensor<fp16, []> var_57420_to_fp16 = const()[name = tensor<string, []>("op_57420_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4907_cast_fp16 = mul(x = var_57419_cast_fp16, y = var_57420_to_fp16)[name = tensor<string, []>("aw_4907_cast_fp16")];
            tensor<string, []> var_57423_equation_0 = const()[name = tensor<string, []>("op_57423_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57423_cast_fp16 = einsum(equation = var_57423_equation_0, values = (var_57265_cast_fp16, var_57182_cast_fp16))[name = tensor<string, []>("op_57423_cast_fp16")];
            tensor<fp16, []> var_57424_to_fp16 = const()[name = tensor<string, []>("op_57424_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4909_cast_fp16 = mul(x = var_57423_cast_fp16, y = var_57424_to_fp16)[name = tensor<string, []>("aw_4909_cast_fp16")];
            tensor<string, []> var_57427_equation_0 = const()[name = tensor<string, []>("op_57427_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57427_cast_fp16 = einsum(equation = var_57427_equation_0, values = (var_57269_cast_fp16, var_57186_cast_fp16))[name = tensor<string, []>("op_57427_cast_fp16")];
            tensor<fp16, []> var_57428_to_fp16 = const()[name = tensor<string, []>("op_57428_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4911_cast_fp16 = mul(x = var_57427_cast_fp16, y = var_57428_to_fp16)[name = tensor<string, []>("aw_4911_cast_fp16")];
            tensor<string, []> var_57431_equation_0 = const()[name = tensor<string, []>("op_57431_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57431_cast_fp16 = einsum(equation = var_57431_equation_0, values = (var_57273_cast_fp16, var_57190_cast_fp16))[name = tensor<string, []>("op_57431_cast_fp16")];
            tensor<fp16, []> var_57432_to_fp16 = const()[name = tensor<string, []>("op_57432_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4913_cast_fp16 = mul(x = var_57431_cast_fp16, y = var_57432_to_fp16)[name = tensor<string, []>("aw_4913_cast_fp16")];
            tensor<string, []> var_57435_equation_0 = const()[name = tensor<string, []>("op_57435_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57435_cast_fp16 = einsum(equation = var_57435_equation_0, values = (var_57277_cast_fp16, var_57194_cast_fp16))[name = tensor<string, []>("op_57435_cast_fp16")];
            tensor<fp16, []> var_57436_to_fp16 = const()[name = tensor<string, []>("op_57436_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4915_cast_fp16 = mul(x = var_57435_cast_fp16, y = var_57436_to_fp16)[name = tensor<string, []>("aw_4915_cast_fp16")];
            tensor<string, []> var_57439_equation_0 = const()[name = tensor<string, []>("op_57439_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57439_cast_fp16 = einsum(equation = var_57439_equation_0, values = (var_57281_cast_fp16, var_57198_cast_fp16))[name = tensor<string, []>("op_57439_cast_fp16")];
            tensor<fp16, []> var_57440_to_fp16 = const()[name = tensor<string, []>("op_57440_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4917_cast_fp16 = mul(x = var_57439_cast_fp16, y = var_57440_to_fp16)[name = tensor<string, []>("aw_4917_cast_fp16")];
            tensor<string, []> var_57443_equation_0 = const()[name = tensor<string, []>("op_57443_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57443_cast_fp16 = einsum(equation = var_57443_equation_0, values = (var_57285_cast_fp16, var_57202_cast_fp16))[name = tensor<string, []>("op_57443_cast_fp16")];
            tensor<fp16, []> var_57444_to_fp16 = const()[name = tensor<string, []>("op_57444_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4919_cast_fp16 = mul(x = var_57443_cast_fp16, y = var_57444_to_fp16)[name = tensor<string, []>("aw_4919_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57446_cast_fp16 = softmax(axis = var_30355, x = aw_4881_cast_fp16)[name = tensor<string, []>("op_57446_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57447_cast_fp16 = softmax(axis = var_30355, x = aw_4883_cast_fp16)[name = tensor<string, []>("op_57447_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57448_cast_fp16 = softmax(axis = var_30355, x = aw_4885_cast_fp16)[name = tensor<string, []>("op_57448_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57449_cast_fp16 = softmax(axis = var_30355, x = aw_4887_cast_fp16)[name = tensor<string, []>("op_57449_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57450_cast_fp16 = softmax(axis = var_30355, x = aw_4889_cast_fp16)[name = tensor<string, []>("op_57450_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57451_cast_fp16 = softmax(axis = var_30355, x = aw_4891_cast_fp16)[name = tensor<string, []>("op_57451_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57452_cast_fp16 = softmax(axis = var_30355, x = aw_4893_cast_fp16)[name = tensor<string, []>("op_57452_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57453_cast_fp16 = softmax(axis = var_30355, x = aw_4895_cast_fp16)[name = tensor<string, []>("op_57453_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57454_cast_fp16 = softmax(axis = var_30355, x = aw_4897_cast_fp16)[name = tensor<string, []>("op_57454_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57455_cast_fp16 = softmax(axis = var_30355, x = aw_4899_cast_fp16)[name = tensor<string, []>("op_57455_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57456_cast_fp16 = softmax(axis = var_30355, x = aw_4901_cast_fp16)[name = tensor<string, []>("op_57456_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57457_cast_fp16 = softmax(axis = var_30355, x = aw_4903_cast_fp16)[name = tensor<string, []>("op_57457_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57458_cast_fp16 = softmax(axis = var_30355, x = aw_4905_cast_fp16)[name = tensor<string, []>("op_57458_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57459_cast_fp16 = softmax(axis = var_30355, x = aw_4907_cast_fp16)[name = tensor<string, []>("op_57459_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57460_cast_fp16 = softmax(axis = var_30355, x = aw_4909_cast_fp16)[name = tensor<string, []>("op_57460_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57461_cast_fp16 = softmax(axis = var_30355, x = aw_4911_cast_fp16)[name = tensor<string, []>("op_57461_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57462_cast_fp16 = softmax(axis = var_30355, x = aw_4913_cast_fp16)[name = tensor<string, []>("op_57462_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57463_cast_fp16 = softmax(axis = var_30355, x = aw_4915_cast_fp16)[name = tensor<string, []>("op_57463_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57464_cast_fp16 = softmax(axis = var_30355, x = aw_4917_cast_fp16)[name = tensor<string, []>("op_57464_cast_fp16")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57465_cast_fp16 = softmax(axis = var_30355, x = aw_4919_cast_fp16)[name = tensor<string, []>("op_57465_cast_fp16")];
            tensor<string, []> var_57467_equation_0 = const()[name = tensor<string, []>("op_57467_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57467_cast_fp16 = einsum(equation = var_57467_equation_0, values = (var_57287_cast_fp16, var_57446_cast_fp16))[name = tensor<string, []>("op_57467_cast_fp16")];
            tensor<string, []> var_57469_equation_0 = const()[name = tensor<string, []>("op_57469_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57469_cast_fp16 = einsum(equation = var_57469_equation_0, values = (var_57291_cast_fp16, var_57447_cast_fp16))[name = tensor<string, []>("op_57469_cast_fp16")];
            tensor<string, []> var_57471_equation_0 = const()[name = tensor<string, []>("op_57471_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57471_cast_fp16 = einsum(equation = var_57471_equation_0, values = (var_57295_cast_fp16, var_57448_cast_fp16))[name = tensor<string, []>("op_57471_cast_fp16")];
            tensor<string, []> var_57473_equation_0 = const()[name = tensor<string, []>("op_57473_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57473_cast_fp16 = einsum(equation = var_57473_equation_0, values = (var_57299_cast_fp16, var_57449_cast_fp16))[name = tensor<string, []>("op_57473_cast_fp16")];
            tensor<string, []> var_57475_equation_0 = const()[name = tensor<string, []>("op_57475_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57475_cast_fp16 = einsum(equation = var_57475_equation_0, values = (var_57303_cast_fp16, var_57450_cast_fp16))[name = tensor<string, []>("op_57475_cast_fp16")];
            tensor<string, []> var_57477_equation_0 = const()[name = tensor<string, []>("op_57477_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57477_cast_fp16 = einsum(equation = var_57477_equation_0, values = (var_57307_cast_fp16, var_57451_cast_fp16))[name = tensor<string, []>("op_57477_cast_fp16")];
            tensor<string, []> var_57479_equation_0 = const()[name = tensor<string, []>("op_57479_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57479_cast_fp16 = einsum(equation = var_57479_equation_0, values = (var_57311_cast_fp16, var_57452_cast_fp16))[name = tensor<string, []>("op_57479_cast_fp16")];
            tensor<string, []> var_57481_equation_0 = const()[name = tensor<string, []>("op_57481_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57481_cast_fp16 = einsum(equation = var_57481_equation_0, values = (var_57315_cast_fp16, var_57453_cast_fp16))[name = tensor<string, []>("op_57481_cast_fp16")];
            tensor<string, []> var_57483_equation_0 = const()[name = tensor<string, []>("op_57483_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57483_cast_fp16 = einsum(equation = var_57483_equation_0, values = (var_57319_cast_fp16, var_57454_cast_fp16))[name = tensor<string, []>("op_57483_cast_fp16")];
            tensor<string, []> var_57485_equation_0 = const()[name = tensor<string, []>("op_57485_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57485_cast_fp16 = einsum(equation = var_57485_equation_0, values = (var_57323_cast_fp16, var_57455_cast_fp16))[name = tensor<string, []>("op_57485_cast_fp16")];
            tensor<string, []> var_57487_equation_0 = const()[name = tensor<string, []>("op_57487_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57487_cast_fp16 = einsum(equation = var_57487_equation_0, values = (var_57327_cast_fp16, var_57456_cast_fp16))[name = tensor<string, []>("op_57487_cast_fp16")];
            tensor<string, []> var_57489_equation_0 = const()[name = tensor<string, []>("op_57489_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57489_cast_fp16 = einsum(equation = var_57489_equation_0, values = (var_57331_cast_fp16, var_57457_cast_fp16))[name = tensor<string, []>("op_57489_cast_fp16")];
            tensor<string, []> var_57491_equation_0 = const()[name = tensor<string, []>("op_57491_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57491_cast_fp16 = einsum(equation = var_57491_equation_0, values = (var_57335_cast_fp16, var_57458_cast_fp16))[name = tensor<string, []>("op_57491_cast_fp16")];
            tensor<string, []> var_57493_equation_0 = const()[name = tensor<string, []>("op_57493_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57493_cast_fp16 = einsum(equation = var_57493_equation_0, values = (var_57339_cast_fp16, var_57459_cast_fp16))[name = tensor<string, []>("op_57493_cast_fp16")];
            tensor<string, []> var_57495_equation_0 = const()[name = tensor<string, []>("op_57495_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57495_cast_fp16 = einsum(equation = var_57495_equation_0, values = (var_57343_cast_fp16, var_57460_cast_fp16))[name = tensor<string, []>("op_57495_cast_fp16")];
            tensor<string, []> var_57497_equation_0 = const()[name = tensor<string, []>("op_57497_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57497_cast_fp16 = einsum(equation = var_57497_equation_0, values = (var_57347_cast_fp16, var_57461_cast_fp16))[name = tensor<string, []>("op_57497_cast_fp16")];
            tensor<string, []> var_57499_equation_0 = const()[name = tensor<string, []>("op_57499_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57499_cast_fp16 = einsum(equation = var_57499_equation_0, values = (var_57351_cast_fp16, var_57462_cast_fp16))[name = tensor<string, []>("op_57499_cast_fp16")];
            tensor<string, []> var_57501_equation_0 = const()[name = tensor<string, []>("op_57501_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57501_cast_fp16 = einsum(equation = var_57501_equation_0, values = (var_57355_cast_fp16, var_57463_cast_fp16))[name = tensor<string, []>("op_57501_cast_fp16")];
            tensor<string, []> var_57503_equation_0 = const()[name = tensor<string, []>("op_57503_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57503_cast_fp16 = einsum(equation = var_57503_equation_0, values = (var_57359_cast_fp16, var_57464_cast_fp16))[name = tensor<string, []>("op_57503_cast_fp16")];
            tensor<string, []> var_57505_equation_0 = const()[name = tensor<string, []>("op_57505_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57505_cast_fp16 = einsum(equation = var_57505_equation_0, values = (var_57363_cast_fp16, var_57465_cast_fp16))[name = tensor<string, []>("op_57505_cast_fp16")];
            tensor<bool, []> input_707_interleave_0 = const()[name = tensor<string, []>("input_707_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_707_cast_fp16 = concat(axis = var_30355, interleave = input_707_interleave_0, values = (var_57467_cast_fp16, var_57469_cast_fp16, var_57471_cast_fp16, var_57473_cast_fp16, var_57475_cast_fp16, var_57477_cast_fp16, var_57479_cast_fp16, var_57481_cast_fp16, var_57483_cast_fp16, var_57485_cast_fp16, var_57487_cast_fp16, var_57489_cast_fp16, var_57491_cast_fp16, var_57493_cast_fp16, var_57495_cast_fp16, var_57497_cast_fp16, var_57499_cast_fp16, var_57501_cast_fp16, var_57503_cast_fp16, var_57505_cast_fp16))[name = tensor<string, []>("input_707_cast_fp16")];
            tensor<string, []> var_57515_pad_type_0 = const()[name = tensor<string, []>("op_57515_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_57515_strides_0 = const()[name = tensor<string, []>("op_57515_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_57515_pad_0 = const()[name = tensor<string, []>("op_57515_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_57515_dilations_0 = const()[name = tensor<string, []>("op_57515_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_57515_groups_0 = const()[name = tensor<string, []>("op_57515_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_9_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(836811456))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(838040320))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_9_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_9_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_9_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(838040512)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_57515_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_9_attn1_to_out_0_bias_to_fp16, dilations = var_57515_dilations_0, groups = var_57515_groups_0, pad = var_57515_pad_0, pad_type = var_57515_pad_type_0, strides = var_57515_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_9_attn1_to_out_0_weight_to_fp16_palettized, x = input_707_cast_fp16)[name = tensor<string, []>("op_57515_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_381_cast_fp16 = add(x = var_57515_cast_fp16, y = inputs_379_cast_fp16)[name = tensor<string, []>("inputs_381_cast_fp16")];
            tensor<int32, [1]> hidden_states_491_axes_0 = const()[name = tensor<string, []>("hidden_states_491_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_491_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_491_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(838043136)))];
            tensor<fp16, [1280]> hidden_states_491_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_491_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(838045760)))];
            tensor<fp16, []> var_57525_to_fp16 = const()[name = tensor<string, []>("op_57525_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_491_cast_fp16 = layer_norm(axes = hidden_states_491_axes_0, beta = hidden_states_491_beta_0_to_fp16, epsilon = var_57525_to_fp16, gamma = hidden_states_491_gamma_0_to_fp16, x = inputs_381_cast_fp16)[name = tensor<string, []>("hidden_states_491_cast_fp16")];
            tensor<string, []> q_255_pad_type_0 = const()[name = tensor<string, []>("q_255_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_255_strides_0 = const()[name = tensor<string, []>("q_255_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_255_pad_0 = const()[name = tensor<string, []>("q_255_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_255_dilations_0 = const()[name = tensor<string, []>("q_255_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_255_groups_0 = const()[name = tensor<string, []>("q_255_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_9_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(838048384))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(839277248))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_9_attn2_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 1024]> q_255_cast_fp16 = conv(dilations = q_255_dilations_0, groups = q_255_groups_0, pad = q_255_pad_0, pad_type = q_255_pad_type_0, strides = q_255_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_9_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_491_cast_fp16)[name = tensor<string, []>("q_255_cast_fp16")];
            tensor<string, []> k_509_pad_type_0 = const()[name = tensor<string, []>("k_509_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_509_strides_0 = const()[name = tensor<string, []>("k_509_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_509_pad_0 = const()[name = tensor<string, []>("k_509_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_509_dilations_0 = const()[name = tensor<string, []>("k_509_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_509_groups_0 = const()[name = tensor<string, []>("k_509_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_9_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(839277440))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(841243584))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_9_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> k_509_cast_fp16 = conv(dilations = k_509_dilations_0, groups = k_509_groups_0, pad = k_509_pad_0, pad_type = k_509_pad_type_0, strides = k_509_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_9_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("k_509_cast_fp16")];
            tensor<string, []> v_255_pad_type_0 = const()[name = tensor<string, []>("v_255_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_255_strides_0 = const()[name = tensor<string, []>("v_255_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_255_pad_0 = const()[name = tensor<string, []>("v_255_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_255_dilations_0 = const()[name = tensor<string, []>("v_255_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_255_groups_0 = const()[name = tensor<string, []>("v_255_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_9_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1966080]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(841243776))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(843209920))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_9_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2048, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> v_255_cast_fp16 = conv(dilations = v_255_dilations_0, groups = v_255_groups_0, pad = v_255_pad_0, pad_type = v_255_pad_type_0, strides = v_255_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_9_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("v_255_cast_fp16")];
            tensor<int32, [4]> var_57558_begin_0 = const()[name = tensor<string, []>("op_57558_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_57558_end_0 = const()[name = tensor<string, []>("op_57558_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_57558_end_mask_0 = const()[name = tensor<string, []>("op_57558_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57558_cast_fp16 = slice_by_index(begin = var_57558_begin_0, end = var_57558_end_0, end_mask = var_57558_end_mask_0, x = q_255_cast_fp16)[name = tensor<string, []>("op_57558_cast_fp16")];
            tensor<int32, [4]> var_57562_begin_0 = const()[name = tensor<string, []>("op_57562_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_57562_end_0 = const()[name = tensor<string, []>("op_57562_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_57562_end_mask_0 = const()[name = tensor<string, []>("op_57562_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57562_cast_fp16 = slice_by_index(begin = var_57562_begin_0, end = var_57562_end_0, end_mask = var_57562_end_mask_0, x = q_255_cast_fp16)[name = tensor<string, []>("op_57562_cast_fp16")];
            tensor<int32, [4]> var_57566_begin_0 = const()[name = tensor<string, []>("op_57566_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_57566_end_0 = const()[name = tensor<string, []>("op_57566_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_57566_end_mask_0 = const()[name = tensor<string, []>("op_57566_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57566_cast_fp16 = slice_by_index(begin = var_57566_begin_0, end = var_57566_end_0, end_mask = var_57566_end_mask_0, x = q_255_cast_fp16)[name = tensor<string, []>("op_57566_cast_fp16")];
            tensor<int32, [4]> var_57570_begin_0 = const()[name = tensor<string, []>("op_57570_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_57570_end_0 = const()[name = tensor<string, []>("op_57570_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_57570_end_mask_0 = const()[name = tensor<string, []>("op_57570_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57570_cast_fp16 = slice_by_index(begin = var_57570_begin_0, end = var_57570_end_0, end_mask = var_57570_end_mask_0, x = q_255_cast_fp16)[name = tensor<string, []>("op_57570_cast_fp16")];
            tensor<int32, [4]> var_57574_begin_0 = const()[name = tensor<string, []>("op_57574_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_57574_end_0 = const()[name = tensor<string, []>("op_57574_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_57574_end_mask_0 = const()[name = tensor<string, []>("op_57574_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57574_cast_fp16 = slice_by_index(begin = var_57574_begin_0, end = var_57574_end_0, end_mask = var_57574_end_mask_0, x = q_255_cast_fp16)[name = tensor<string, []>("op_57574_cast_fp16")];
            tensor<int32, [4]> var_57578_begin_0 = const()[name = tensor<string, []>("op_57578_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_57578_end_0 = const()[name = tensor<string, []>("op_57578_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_57578_end_mask_0 = const()[name = tensor<string, []>("op_57578_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57578_cast_fp16 = slice_by_index(begin = var_57578_begin_0, end = var_57578_end_0, end_mask = var_57578_end_mask_0, x = q_255_cast_fp16)[name = tensor<string, []>("op_57578_cast_fp16")];
            tensor<int32, [4]> var_57582_begin_0 = const()[name = tensor<string, []>("op_57582_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_57582_end_0 = const()[name = tensor<string, []>("op_57582_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_57582_end_mask_0 = const()[name = tensor<string, []>("op_57582_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57582_cast_fp16 = slice_by_index(begin = var_57582_begin_0, end = var_57582_end_0, end_mask = var_57582_end_mask_0, x = q_255_cast_fp16)[name = tensor<string, []>("op_57582_cast_fp16")];
            tensor<int32, [4]> var_57586_begin_0 = const()[name = tensor<string, []>("op_57586_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_57586_end_0 = const()[name = tensor<string, []>("op_57586_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_57586_end_mask_0 = const()[name = tensor<string, []>("op_57586_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57586_cast_fp16 = slice_by_index(begin = var_57586_begin_0, end = var_57586_end_0, end_mask = var_57586_end_mask_0, x = q_255_cast_fp16)[name = tensor<string, []>("op_57586_cast_fp16")];
            tensor<int32, [4]> var_57590_begin_0 = const()[name = tensor<string, []>("op_57590_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_57590_end_0 = const()[name = tensor<string, []>("op_57590_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_57590_end_mask_0 = const()[name = tensor<string, []>("op_57590_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57590_cast_fp16 = slice_by_index(begin = var_57590_begin_0, end = var_57590_end_0, end_mask = var_57590_end_mask_0, x = q_255_cast_fp16)[name = tensor<string, []>("op_57590_cast_fp16")];
            tensor<int32, [4]> var_57594_begin_0 = const()[name = tensor<string, []>("op_57594_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_57594_end_0 = const()[name = tensor<string, []>("op_57594_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_57594_end_mask_0 = const()[name = tensor<string, []>("op_57594_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57594_cast_fp16 = slice_by_index(begin = var_57594_begin_0, end = var_57594_end_0, end_mask = var_57594_end_mask_0, x = q_255_cast_fp16)[name = tensor<string, []>("op_57594_cast_fp16")];
            tensor<int32, [4]> var_57598_begin_0 = const()[name = tensor<string, []>("op_57598_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_57598_end_0 = const()[name = tensor<string, []>("op_57598_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_57598_end_mask_0 = const()[name = tensor<string, []>("op_57598_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57598_cast_fp16 = slice_by_index(begin = var_57598_begin_0, end = var_57598_end_0, end_mask = var_57598_end_mask_0, x = q_255_cast_fp16)[name = tensor<string, []>("op_57598_cast_fp16")];
            tensor<int32, [4]> var_57602_begin_0 = const()[name = tensor<string, []>("op_57602_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_57602_end_0 = const()[name = tensor<string, []>("op_57602_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_57602_end_mask_0 = const()[name = tensor<string, []>("op_57602_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57602_cast_fp16 = slice_by_index(begin = var_57602_begin_0, end = var_57602_end_0, end_mask = var_57602_end_mask_0, x = q_255_cast_fp16)[name = tensor<string, []>("op_57602_cast_fp16")];
            tensor<int32, [4]> var_57606_begin_0 = const()[name = tensor<string, []>("op_57606_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_57606_end_0 = const()[name = tensor<string, []>("op_57606_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_57606_end_mask_0 = const()[name = tensor<string, []>("op_57606_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57606_cast_fp16 = slice_by_index(begin = var_57606_begin_0, end = var_57606_end_0, end_mask = var_57606_end_mask_0, x = q_255_cast_fp16)[name = tensor<string, []>("op_57606_cast_fp16")];
            tensor<int32, [4]> var_57610_begin_0 = const()[name = tensor<string, []>("op_57610_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_57610_end_0 = const()[name = tensor<string, []>("op_57610_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_57610_end_mask_0 = const()[name = tensor<string, []>("op_57610_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57610_cast_fp16 = slice_by_index(begin = var_57610_begin_0, end = var_57610_end_0, end_mask = var_57610_end_mask_0, x = q_255_cast_fp16)[name = tensor<string, []>("op_57610_cast_fp16")];
            tensor<int32, [4]> var_57614_begin_0 = const()[name = tensor<string, []>("op_57614_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_57614_end_0 = const()[name = tensor<string, []>("op_57614_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_57614_end_mask_0 = const()[name = tensor<string, []>("op_57614_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57614_cast_fp16 = slice_by_index(begin = var_57614_begin_0, end = var_57614_end_0, end_mask = var_57614_end_mask_0, x = q_255_cast_fp16)[name = tensor<string, []>("op_57614_cast_fp16")];
            tensor<int32, [4]> var_57618_begin_0 = const()[name = tensor<string, []>("op_57618_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_57618_end_0 = const()[name = tensor<string, []>("op_57618_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_57618_end_mask_0 = const()[name = tensor<string, []>("op_57618_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57618_cast_fp16 = slice_by_index(begin = var_57618_begin_0, end = var_57618_end_0, end_mask = var_57618_end_mask_0, x = q_255_cast_fp16)[name = tensor<string, []>("op_57618_cast_fp16")];
            tensor<int32, [4]> var_57622_begin_0 = const()[name = tensor<string, []>("op_57622_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_57622_end_0 = const()[name = tensor<string, []>("op_57622_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_57622_end_mask_0 = const()[name = tensor<string, []>("op_57622_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57622_cast_fp16 = slice_by_index(begin = var_57622_begin_0, end = var_57622_end_0, end_mask = var_57622_end_mask_0, x = q_255_cast_fp16)[name = tensor<string, []>("op_57622_cast_fp16")];
            tensor<int32, [4]> var_57626_begin_0 = const()[name = tensor<string, []>("op_57626_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_57626_end_0 = const()[name = tensor<string, []>("op_57626_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_57626_end_mask_0 = const()[name = tensor<string, []>("op_57626_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57626_cast_fp16 = slice_by_index(begin = var_57626_begin_0, end = var_57626_end_0, end_mask = var_57626_end_mask_0, x = q_255_cast_fp16)[name = tensor<string, []>("op_57626_cast_fp16")];
            tensor<int32, [4]> var_57630_begin_0 = const()[name = tensor<string, []>("op_57630_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_57630_end_0 = const()[name = tensor<string, []>("op_57630_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_57630_end_mask_0 = const()[name = tensor<string, []>("op_57630_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57630_cast_fp16 = slice_by_index(begin = var_57630_begin_0, end = var_57630_end_0, end_mask = var_57630_end_mask_0, x = q_255_cast_fp16)[name = tensor<string, []>("op_57630_cast_fp16")];
            tensor<int32, [4]> var_57634_begin_0 = const()[name = tensor<string, []>("op_57634_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_57634_end_0 = const()[name = tensor<string, []>("op_57634_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_57634_end_mask_0 = const()[name = tensor<string, []>("op_57634_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57634_cast_fp16 = slice_by_index(begin = var_57634_begin_0, end = var_57634_end_0, end_mask = var_57634_end_mask_0, x = q_255_cast_fp16)[name = tensor<string, []>("op_57634_cast_fp16")];
            tensor<int32, [4]> k_511_perm_0 = const()[name = tensor<string, []>("k_511_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_57641_begin_0 = const()[name = tensor<string, []>("op_57641_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_57641_end_0 = const()[name = tensor<string, []>("op_57641_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_57641_end_mask_0 = const()[name = tensor<string, []>("op_57641_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> k_511_cast_fp16 = transpose(perm = k_511_perm_0, x = k_509_cast_fp16)[name = tensor<string, []>("transpose_12")];
            tensor<fp16, [2, 77, 1, 64]> var_57641_cast_fp16 = slice_by_index(begin = var_57641_begin_0, end = var_57641_end_0, end_mask = var_57641_end_mask_0, x = k_511_cast_fp16)[name = tensor<string, []>("op_57641_cast_fp16")];
            tensor<int32, [4]> var_57645_begin_0 = const()[name = tensor<string, []>("op_57645_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_57645_end_0 = const()[name = tensor<string, []>("op_57645_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_57645_end_mask_0 = const()[name = tensor<string, []>("op_57645_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_57645_cast_fp16 = slice_by_index(begin = var_57645_begin_0, end = var_57645_end_0, end_mask = var_57645_end_mask_0, x = k_511_cast_fp16)[name = tensor<string, []>("op_57645_cast_fp16")];
            tensor<int32, [4]> var_57649_begin_0 = const()[name = tensor<string, []>("op_57649_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_57649_end_0 = const()[name = tensor<string, []>("op_57649_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_57649_end_mask_0 = const()[name = tensor<string, []>("op_57649_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_57649_cast_fp16 = slice_by_index(begin = var_57649_begin_0, end = var_57649_end_0, end_mask = var_57649_end_mask_0, x = k_511_cast_fp16)[name = tensor<string, []>("op_57649_cast_fp16")];
            tensor<int32, [4]> var_57653_begin_0 = const()[name = tensor<string, []>("op_57653_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_57653_end_0 = const()[name = tensor<string, []>("op_57653_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_57653_end_mask_0 = const()[name = tensor<string, []>("op_57653_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_57653_cast_fp16 = slice_by_index(begin = var_57653_begin_0, end = var_57653_end_0, end_mask = var_57653_end_mask_0, x = k_511_cast_fp16)[name = tensor<string, []>("op_57653_cast_fp16")];
            tensor<int32, [4]> var_57657_begin_0 = const()[name = tensor<string, []>("op_57657_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_57657_end_0 = const()[name = tensor<string, []>("op_57657_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_57657_end_mask_0 = const()[name = tensor<string, []>("op_57657_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_57657_cast_fp16 = slice_by_index(begin = var_57657_begin_0, end = var_57657_end_0, end_mask = var_57657_end_mask_0, x = k_511_cast_fp16)[name = tensor<string, []>("op_57657_cast_fp16")];
            tensor<int32, [4]> var_57661_begin_0 = const()[name = tensor<string, []>("op_57661_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_57661_end_0 = const()[name = tensor<string, []>("op_57661_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_57661_end_mask_0 = const()[name = tensor<string, []>("op_57661_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_57661_cast_fp16 = slice_by_index(begin = var_57661_begin_0, end = var_57661_end_0, end_mask = var_57661_end_mask_0, x = k_511_cast_fp16)[name = tensor<string, []>("op_57661_cast_fp16")];
            tensor<int32, [4]> var_57665_begin_0 = const()[name = tensor<string, []>("op_57665_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_57665_end_0 = const()[name = tensor<string, []>("op_57665_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_57665_end_mask_0 = const()[name = tensor<string, []>("op_57665_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_57665_cast_fp16 = slice_by_index(begin = var_57665_begin_0, end = var_57665_end_0, end_mask = var_57665_end_mask_0, x = k_511_cast_fp16)[name = tensor<string, []>("op_57665_cast_fp16")];
            tensor<int32, [4]> var_57669_begin_0 = const()[name = tensor<string, []>("op_57669_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_57669_end_0 = const()[name = tensor<string, []>("op_57669_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_57669_end_mask_0 = const()[name = tensor<string, []>("op_57669_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_57669_cast_fp16 = slice_by_index(begin = var_57669_begin_0, end = var_57669_end_0, end_mask = var_57669_end_mask_0, x = k_511_cast_fp16)[name = tensor<string, []>("op_57669_cast_fp16")];
            tensor<int32, [4]> var_57673_begin_0 = const()[name = tensor<string, []>("op_57673_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_57673_end_0 = const()[name = tensor<string, []>("op_57673_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_57673_end_mask_0 = const()[name = tensor<string, []>("op_57673_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_57673_cast_fp16 = slice_by_index(begin = var_57673_begin_0, end = var_57673_end_0, end_mask = var_57673_end_mask_0, x = k_511_cast_fp16)[name = tensor<string, []>("op_57673_cast_fp16")];
            tensor<int32, [4]> var_57677_begin_0 = const()[name = tensor<string, []>("op_57677_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_57677_end_0 = const()[name = tensor<string, []>("op_57677_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_57677_end_mask_0 = const()[name = tensor<string, []>("op_57677_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_57677_cast_fp16 = slice_by_index(begin = var_57677_begin_0, end = var_57677_end_0, end_mask = var_57677_end_mask_0, x = k_511_cast_fp16)[name = tensor<string, []>("op_57677_cast_fp16")];
            tensor<int32, [4]> var_57681_begin_0 = const()[name = tensor<string, []>("op_57681_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_57681_end_0 = const()[name = tensor<string, []>("op_57681_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_57681_end_mask_0 = const()[name = tensor<string, []>("op_57681_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_57681_cast_fp16 = slice_by_index(begin = var_57681_begin_0, end = var_57681_end_0, end_mask = var_57681_end_mask_0, x = k_511_cast_fp16)[name = tensor<string, []>("op_57681_cast_fp16")];
            tensor<int32, [4]> var_57685_begin_0 = const()[name = tensor<string, []>("op_57685_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_57685_end_0 = const()[name = tensor<string, []>("op_57685_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_57685_end_mask_0 = const()[name = tensor<string, []>("op_57685_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_57685_cast_fp16 = slice_by_index(begin = var_57685_begin_0, end = var_57685_end_0, end_mask = var_57685_end_mask_0, x = k_511_cast_fp16)[name = tensor<string, []>("op_57685_cast_fp16")];
            tensor<int32, [4]> var_57689_begin_0 = const()[name = tensor<string, []>("op_57689_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_57689_end_0 = const()[name = tensor<string, []>("op_57689_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_57689_end_mask_0 = const()[name = tensor<string, []>("op_57689_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_57689_cast_fp16 = slice_by_index(begin = var_57689_begin_0, end = var_57689_end_0, end_mask = var_57689_end_mask_0, x = k_511_cast_fp16)[name = tensor<string, []>("op_57689_cast_fp16")];
            tensor<int32, [4]> var_57693_begin_0 = const()[name = tensor<string, []>("op_57693_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_57693_end_0 = const()[name = tensor<string, []>("op_57693_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_57693_end_mask_0 = const()[name = tensor<string, []>("op_57693_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_57693_cast_fp16 = slice_by_index(begin = var_57693_begin_0, end = var_57693_end_0, end_mask = var_57693_end_mask_0, x = k_511_cast_fp16)[name = tensor<string, []>("op_57693_cast_fp16")];
            tensor<int32, [4]> var_57697_begin_0 = const()[name = tensor<string, []>("op_57697_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_57697_end_0 = const()[name = tensor<string, []>("op_57697_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_57697_end_mask_0 = const()[name = tensor<string, []>("op_57697_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_57697_cast_fp16 = slice_by_index(begin = var_57697_begin_0, end = var_57697_end_0, end_mask = var_57697_end_mask_0, x = k_511_cast_fp16)[name = tensor<string, []>("op_57697_cast_fp16")];
            tensor<int32, [4]> var_57701_begin_0 = const()[name = tensor<string, []>("op_57701_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_57701_end_0 = const()[name = tensor<string, []>("op_57701_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_57701_end_mask_0 = const()[name = tensor<string, []>("op_57701_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_57701_cast_fp16 = slice_by_index(begin = var_57701_begin_0, end = var_57701_end_0, end_mask = var_57701_end_mask_0, x = k_511_cast_fp16)[name = tensor<string, []>("op_57701_cast_fp16")];
            tensor<int32, [4]> var_57705_begin_0 = const()[name = tensor<string, []>("op_57705_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_57705_end_0 = const()[name = tensor<string, []>("op_57705_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_57705_end_mask_0 = const()[name = tensor<string, []>("op_57705_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_57705_cast_fp16 = slice_by_index(begin = var_57705_begin_0, end = var_57705_end_0, end_mask = var_57705_end_mask_0, x = k_511_cast_fp16)[name = tensor<string, []>("op_57705_cast_fp16")];
            tensor<int32, [4]> var_57709_begin_0 = const()[name = tensor<string, []>("op_57709_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_57709_end_0 = const()[name = tensor<string, []>("op_57709_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_57709_end_mask_0 = const()[name = tensor<string, []>("op_57709_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_57709_cast_fp16 = slice_by_index(begin = var_57709_begin_0, end = var_57709_end_0, end_mask = var_57709_end_mask_0, x = k_511_cast_fp16)[name = tensor<string, []>("op_57709_cast_fp16")];
            tensor<int32, [4]> var_57713_begin_0 = const()[name = tensor<string, []>("op_57713_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_57713_end_0 = const()[name = tensor<string, []>("op_57713_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_57713_end_mask_0 = const()[name = tensor<string, []>("op_57713_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_57713_cast_fp16 = slice_by_index(begin = var_57713_begin_0, end = var_57713_end_0, end_mask = var_57713_end_mask_0, x = k_511_cast_fp16)[name = tensor<string, []>("op_57713_cast_fp16")];
            tensor<int32, [4]> var_57717_begin_0 = const()[name = tensor<string, []>("op_57717_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_57717_end_0 = const()[name = tensor<string, []>("op_57717_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_57717_end_mask_0 = const()[name = tensor<string, []>("op_57717_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_57717_cast_fp16 = slice_by_index(begin = var_57717_begin_0, end = var_57717_end_0, end_mask = var_57717_end_mask_0, x = k_511_cast_fp16)[name = tensor<string, []>("op_57717_cast_fp16")];
            tensor<int32, [4]> var_57719_begin_0 = const()[name = tensor<string, []>("op_57719_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_57719_end_0 = const()[name = tensor<string, []>("op_57719_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_57719_end_mask_0 = const()[name = tensor<string, []>("op_57719_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_57719_cast_fp16 = slice_by_index(begin = var_57719_begin_0, end = var_57719_end_0, end_mask = var_57719_end_mask_0, x = v_255_cast_fp16)[name = tensor<string, []>("op_57719_cast_fp16")];
            tensor<int32, [4]> var_57723_begin_0 = const()[name = tensor<string, []>("op_57723_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_57723_end_0 = const()[name = tensor<string, []>("op_57723_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_57723_end_mask_0 = const()[name = tensor<string, []>("op_57723_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_57723_cast_fp16 = slice_by_index(begin = var_57723_begin_0, end = var_57723_end_0, end_mask = var_57723_end_mask_0, x = v_255_cast_fp16)[name = tensor<string, []>("op_57723_cast_fp16")];
            tensor<int32, [4]> var_57727_begin_0 = const()[name = tensor<string, []>("op_57727_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_57727_end_0 = const()[name = tensor<string, []>("op_57727_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_57727_end_mask_0 = const()[name = tensor<string, []>("op_57727_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_57727_cast_fp16 = slice_by_index(begin = var_57727_begin_0, end = var_57727_end_0, end_mask = var_57727_end_mask_0, x = v_255_cast_fp16)[name = tensor<string, []>("op_57727_cast_fp16")];
            tensor<int32, [4]> var_57731_begin_0 = const()[name = tensor<string, []>("op_57731_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_57731_end_0 = const()[name = tensor<string, []>("op_57731_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_57731_end_mask_0 = const()[name = tensor<string, []>("op_57731_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_57731_cast_fp16 = slice_by_index(begin = var_57731_begin_0, end = var_57731_end_0, end_mask = var_57731_end_mask_0, x = v_255_cast_fp16)[name = tensor<string, []>("op_57731_cast_fp16")];
            tensor<int32, [4]> var_57735_begin_0 = const()[name = tensor<string, []>("op_57735_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_57735_end_0 = const()[name = tensor<string, []>("op_57735_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_57735_end_mask_0 = const()[name = tensor<string, []>("op_57735_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_57735_cast_fp16 = slice_by_index(begin = var_57735_begin_0, end = var_57735_end_0, end_mask = var_57735_end_mask_0, x = v_255_cast_fp16)[name = tensor<string, []>("op_57735_cast_fp16")];
            tensor<int32, [4]> var_57739_begin_0 = const()[name = tensor<string, []>("op_57739_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_57739_end_0 = const()[name = tensor<string, []>("op_57739_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_57739_end_mask_0 = const()[name = tensor<string, []>("op_57739_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_57739_cast_fp16 = slice_by_index(begin = var_57739_begin_0, end = var_57739_end_0, end_mask = var_57739_end_mask_0, x = v_255_cast_fp16)[name = tensor<string, []>("op_57739_cast_fp16")];
            tensor<int32, [4]> var_57743_begin_0 = const()[name = tensor<string, []>("op_57743_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_57743_end_0 = const()[name = tensor<string, []>("op_57743_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_57743_end_mask_0 = const()[name = tensor<string, []>("op_57743_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_57743_cast_fp16 = slice_by_index(begin = var_57743_begin_0, end = var_57743_end_0, end_mask = var_57743_end_mask_0, x = v_255_cast_fp16)[name = tensor<string, []>("op_57743_cast_fp16")];
            tensor<int32, [4]> var_57747_begin_0 = const()[name = tensor<string, []>("op_57747_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_57747_end_0 = const()[name = tensor<string, []>("op_57747_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_57747_end_mask_0 = const()[name = tensor<string, []>("op_57747_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_57747_cast_fp16 = slice_by_index(begin = var_57747_begin_0, end = var_57747_end_0, end_mask = var_57747_end_mask_0, x = v_255_cast_fp16)[name = tensor<string, []>("op_57747_cast_fp16")];
            tensor<int32, [4]> var_57751_begin_0 = const()[name = tensor<string, []>("op_57751_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_57751_end_0 = const()[name = tensor<string, []>("op_57751_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_57751_end_mask_0 = const()[name = tensor<string, []>("op_57751_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_57751_cast_fp16 = slice_by_index(begin = var_57751_begin_0, end = var_57751_end_0, end_mask = var_57751_end_mask_0, x = v_255_cast_fp16)[name = tensor<string, []>("op_57751_cast_fp16")];
            tensor<int32, [4]> var_57755_begin_0 = const()[name = tensor<string, []>("op_57755_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_57755_end_0 = const()[name = tensor<string, []>("op_57755_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_57755_end_mask_0 = const()[name = tensor<string, []>("op_57755_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_57755_cast_fp16 = slice_by_index(begin = var_57755_begin_0, end = var_57755_end_0, end_mask = var_57755_end_mask_0, x = v_255_cast_fp16)[name = tensor<string, []>("op_57755_cast_fp16")];
            tensor<int32, [4]> var_57759_begin_0 = const()[name = tensor<string, []>("op_57759_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_57759_end_0 = const()[name = tensor<string, []>("op_57759_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_57759_end_mask_0 = const()[name = tensor<string, []>("op_57759_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_57759_cast_fp16 = slice_by_index(begin = var_57759_begin_0, end = var_57759_end_0, end_mask = var_57759_end_mask_0, x = v_255_cast_fp16)[name = tensor<string, []>("op_57759_cast_fp16")];
            tensor<int32, [4]> var_57763_begin_0 = const()[name = tensor<string, []>("op_57763_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_57763_end_0 = const()[name = tensor<string, []>("op_57763_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_57763_end_mask_0 = const()[name = tensor<string, []>("op_57763_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_57763_cast_fp16 = slice_by_index(begin = var_57763_begin_0, end = var_57763_end_0, end_mask = var_57763_end_mask_0, x = v_255_cast_fp16)[name = tensor<string, []>("op_57763_cast_fp16")];
            tensor<int32, [4]> var_57767_begin_0 = const()[name = tensor<string, []>("op_57767_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_57767_end_0 = const()[name = tensor<string, []>("op_57767_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_57767_end_mask_0 = const()[name = tensor<string, []>("op_57767_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_57767_cast_fp16 = slice_by_index(begin = var_57767_begin_0, end = var_57767_end_0, end_mask = var_57767_end_mask_0, x = v_255_cast_fp16)[name = tensor<string, []>("op_57767_cast_fp16")];
            tensor<int32, [4]> var_57771_begin_0 = const()[name = tensor<string, []>("op_57771_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_57771_end_0 = const()[name = tensor<string, []>("op_57771_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_57771_end_mask_0 = const()[name = tensor<string, []>("op_57771_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_57771_cast_fp16 = slice_by_index(begin = var_57771_begin_0, end = var_57771_end_0, end_mask = var_57771_end_mask_0, x = v_255_cast_fp16)[name = tensor<string, []>("op_57771_cast_fp16")];
            tensor<int32, [4]> var_57775_begin_0 = const()[name = tensor<string, []>("op_57775_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_57775_end_0 = const()[name = tensor<string, []>("op_57775_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_57775_end_mask_0 = const()[name = tensor<string, []>("op_57775_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_57775_cast_fp16 = slice_by_index(begin = var_57775_begin_0, end = var_57775_end_0, end_mask = var_57775_end_mask_0, x = v_255_cast_fp16)[name = tensor<string, []>("op_57775_cast_fp16")];
            tensor<int32, [4]> var_57779_begin_0 = const()[name = tensor<string, []>("op_57779_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_57779_end_0 = const()[name = tensor<string, []>("op_57779_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_57779_end_mask_0 = const()[name = tensor<string, []>("op_57779_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_57779_cast_fp16 = slice_by_index(begin = var_57779_begin_0, end = var_57779_end_0, end_mask = var_57779_end_mask_0, x = v_255_cast_fp16)[name = tensor<string, []>("op_57779_cast_fp16")];
            tensor<int32, [4]> var_57783_begin_0 = const()[name = tensor<string, []>("op_57783_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_57783_end_0 = const()[name = tensor<string, []>("op_57783_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_57783_end_mask_0 = const()[name = tensor<string, []>("op_57783_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_57783_cast_fp16 = slice_by_index(begin = var_57783_begin_0, end = var_57783_end_0, end_mask = var_57783_end_mask_0, x = v_255_cast_fp16)[name = tensor<string, []>("op_57783_cast_fp16")];
            tensor<int32, [4]> var_57787_begin_0 = const()[name = tensor<string, []>("op_57787_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_57787_end_0 = const()[name = tensor<string, []>("op_57787_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_57787_end_mask_0 = const()[name = tensor<string, []>("op_57787_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_57787_cast_fp16 = slice_by_index(begin = var_57787_begin_0, end = var_57787_end_0, end_mask = var_57787_end_mask_0, x = v_255_cast_fp16)[name = tensor<string, []>("op_57787_cast_fp16")];
            tensor<int32, [4]> var_57791_begin_0 = const()[name = tensor<string, []>("op_57791_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_57791_end_0 = const()[name = tensor<string, []>("op_57791_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_57791_end_mask_0 = const()[name = tensor<string, []>("op_57791_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_57791_cast_fp16 = slice_by_index(begin = var_57791_begin_0, end = var_57791_end_0, end_mask = var_57791_end_mask_0, x = v_255_cast_fp16)[name = tensor<string, []>("op_57791_cast_fp16")];
            tensor<int32, [4]> var_57795_begin_0 = const()[name = tensor<string, []>("op_57795_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_57795_end_0 = const()[name = tensor<string, []>("op_57795_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_57795_end_mask_0 = const()[name = tensor<string, []>("op_57795_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_57795_cast_fp16 = slice_by_index(begin = var_57795_begin_0, end = var_57795_end_0, end_mask = var_57795_end_mask_0, x = v_255_cast_fp16)[name = tensor<string, []>("op_57795_cast_fp16")];
            tensor<string, []> var_57799_equation_0 = const()[name = tensor<string, []>("op_57799_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_57799_cast_fp16 = einsum(equation = var_57799_equation_0, values = (var_57641_cast_fp16, var_57558_cast_fp16))[name = tensor<string, []>("op_57799_cast_fp16")];
            tensor<fp16, []> var_57800_to_fp16 = const()[name = tensor<string, []>("op_57800_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4921_cast_fp16 = mul(x = var_57799_cast_fp16, y = var_57800_to_fp16)[name = tensor<string, []>("aw_4921_cast_fp16")];
            tensor<string, []> var_57803_equation_0 = const()[name = tensor<string, []>("op_57803_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_57803_cast_fp16 = einsum(equation = var_57803_equation_0, values = (var_57645_cast_fp16, var_57562_cast_fp16))[name = tensor<string, []>("op_57803_cast_fp16")];
            tensor<fp16, []> var_57804_to_fp16 = const()[name = tensor<string, []>("op_57804_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4923_cast_fp16 = mul(x = var_57803_cast_fp16, y = var_57804_to_fp16)[name = tensor<string, []>("aw_4923_cast_fp16")];
            tensor<string, []> var_57807_equation_0 = const()[name = tensor<string, []>("op_57807_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_57807_cast_fp16 = einsum(equation = var_57807_equation_0, values = (var_57649_cast_fp16, var_57566_cast_fp16))[name = tensor<string, []>("op_57807_cast_fp16")];
            tensor<fp16, []> var_57808_to_fp16 = const()[name = tensor<string, []>("op_57808_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4925_cast_fp16 = mul(x = var_57807_cast_fp16, y = var_57808_to_fp16)[name = tensor<string, []>("aw_4925_cast_fp16")];
            tensor<string, []> var_57811_equation_0 = const()[name = tensor<string, []>("op_57811_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_57811_cast_fp16 = einsum(equation = var_57811_equation_0, values = (var_57653_cast_fp16, var_57570_cast_fp16))[name = tensor<string, []>("op_57811_cast_fp16")];
            tensor<fp16, []> var_57812_to_fp16 = const()[name = tensor<string, []>("op_57812_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4927_cast_fp16 = mul(x = var_57811_cast_fp16, y = var_57812_to_fp16)[name = tensor<string, []>("aw_4927_cast_fp16")];
            tensor<string, []> var_57815_equation_0 = const()[name = tensor<string, []>("op_57815_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_57815_cast_fp16 = einsum(equation = var_57815_equation_0, values = (var_57657_cast_fp16, var_57574_cast_fp16))[name = tensor<string, []>("op_57815_cast_fp16")];
            tensor<fp16, []> var_57816_to_fp16 = const()[name = tensor<string, []>("op_57816_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4929_cast_fp16 = mul(x = var_57815_cast_fp16, y = var_57816_to_fp16)[name = tensor<string, []>("aw_4929_cast_fp16")];
            tensor<string, []> var_57819_equation_0 = const()[name = tensor<string, []>("op_57819_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_57819_cast_fp16 = einsum(equation = var_57819_equation_0, values = (var_57661_cast_fp16, var_57578_cast_fp16))[name = tensor<string, []>("op_57819_cast_fp16")];
            tensor<fp16, []> var_57820_to_fp16 = const()[name = tensor<string, []>("op_57820_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4931_cast_fp16 = mul(x = var_57819_cast_fp16, y = var_57820_to_fp16)[name = tensor<string, []>("aw_4931_cast_fp16")];
            tensor<string, []> var_57823_equation_0 = const()[name = tensor<string, []>("op_57823_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_57823_cast_fp16 = einsum(equation = var_57823_equation_0, values = (var_57665_cast_fp16, var_57582_cast_fp16))[name = tensor<string, []>("op_57823_cast_fp16")];
            tensor<fp16, []> var_57824_to_fp16 = const()[name = tensor<string, []>("op_57824_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4933_cast_fp16 = mul(x = var_57823_cast_fp16, y = var_57824_to_fp16)[name = tensor<string, []>("aw_4933_cast_fp16")];
            tensor<string, []> var_57827_equation_0 = const()[name = tensor<string, []>("op_57827_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_57827_cast_fp16 = einsum(equation = var_57827_equation_0, values = (var_57669_cast_fp16, var_57586_cast_fp16))[name = tensor<string, []>("op_57827_cast_fp16")];
            tensor<fp16, []> var_57828_to_fp16 = const()[name = tensor<string, []>("op_57828_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4935_cast_fp16 = mul(x = var_57827_cast_fp16, y = var_57828_to_fp16)[name = tensor<string, []>("aw_4935_cast_fp16")];
            tensor<string, []> var_57831_equation_0 = const()[name = tensor<string, []>("op_57831_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_57831_cast_fp16 = einsum(equation = var_57831_equation_0, values = (var_57673_cast_fp16, var_57590_cast_fp16))[name = tensor<string, []>("op_57831_cast_fp16")];
            tensor<fp16, []> var_57832_to_fp16 = const()[name = tensor<string, []>("op_57832_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4937_cast_fp16 = mul(x = var_57831_cast_fp16, y = var_57832_to_fp16)[name = tensor<string, []>("aw_4937_cast_fp16")];
            tensor<string, []> var_57835_equation_0 = const()[name = tensor<string, []>("op_57835_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_57835_cast_fp16 = einsum(equation = var_57835_equation_0, values = (var_57677_cast_fp16, var_57594_cast_fp16))[name = tensor<string, []>("op_57835_cast_fp16")];
            tensor<fp16, []> var_57836_to_fp16 = const()[name = tensor<string, []>("op_57836_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4939_cast_fp16 = mul(x = var_57835_cast_fp16, y = var_57836_to_fp16)[name = tensor<string, []>("aw_4939_cast_fp16")];
            tensor<string, []> var_57839_equation_0 = const()[name = tensor<string, []>("op_57839_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_57839_cast_fp16 = einsum(equation = var_57839_equation_0, values = (var_57681_cast_fp16, var_57598_cast_fp16))[name = tensor<string, []>("op_57839_cast_fp16")];
            tensor<fp16, []> var_57840_to_fp16 = const()[name = tensor<string, []>("op_57840_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4941_cast_fp16 = mul(x = var_57839_cast_fp16, y = var_57840_to_fp16)[name = tensor<string, []>("aw_4941_cast_fp16")];
            tensor<string, []> var_57843_equation_0 = const()[name = tensor<string, []>("op_57843_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_57843_cast_fp16 = einsum(equation = var_57843_equation_0, values = (var_57685_cast_fp16, var_57602_cast_fp16))[name = tensor<string, []>("op_57843_cast_fp16")];
            tensor<fp16, []> var_57844_to_fp16 = const()[name = tensor<string, []>("op_57844_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4943_cast_fp16 = mul(x = var_57843_cast_fp16, y = var_57844_to_fp16)[name = tensor<string, []>("aw_4943_cast_fp16")];
            tensor<string, []> var_57847_equation_0 = const()[name = tensor<string, []>("op_57847_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_57847_cast_fp16 = einsum(equation = var_57847_equation_0, values = (var_57689_cast_fp16, var_57606_cast_fp16))[name = tensor<string, []>("op_57847_cast_fp16")];
            tensor<fp16, []> var_57848_to_fp16 = const()[name = tensor<string, []>("op_57848_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4945_cast_fp16 = mul(x = var_57847_cast_fp16, y = var_57848_to_fp16)[name = tensor<string, []>("aw_4945_cast_fp16")];
            tensor<string, []> var_57851_equation_0 = const()[name = tensor<string, []>("op_57851_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_57851_cast_fp16 = einsum(equation = var_57851_equation_0, values = (var_57693_cast_fp16, var_57610_cast_fp16))[name = tensor<string, []>("op_57851_cast_fp16")];
            tensor<fp16, []> var_57852_to_fp16 = const()[name = tensor<string, []>("op_57852_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4947_cast_fp16 = mul(x = var_57851_cast_fp16, y = var_57852_to_fp16)[name = tensor<string, []>("aw_4947_cast_fp16")];
            tensor<string, []> var_57855_equation_0 = const()[name = tensor<string, []>("op_57855_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_57855_cast_fp16 = einsum(equation = var_57855_equation_0, values = (var_57697_cast_fp16, var_57614_cast_fp16))[name = tensor<string, []>("op_57855_cast_fp16")];
            tensor<fp16, []> var_57856_to_fp16 = const()[name = tensor<string, []>("op_57856_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4949_cast_fp16 = mul(x = var_57855_cast_fp16, y = var_57856_to_fp16)[name = tensor<string, []>("aw_4949_cast_fp16")];
            tensor<string, []> var_57859_equation_0 = const()[name = tensor<string, []>("op_57859_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_57859_cast_fp16 = einsum(equation = var_57859_equation_0, values = (var_57701_cast_fp16, var_57618_cast_fp16))[name = tensor<string, []>("op_57859_cast_fp16")];
            tensor<fp16, []> var_57860_to_fp16 = const()[name = tensor<string, []>("op_57860_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4951_cast_fp16 = mul(x = var_57859_cast_fp16, y = var_57860_to_fp16)[name = tensor<string, []>("aw_4951_cast_fp16")];
            tensor<string, []> var_57863_equation_0 = const()[name = tensor<string, []>("op_57863_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_57863_cast_fp16 = einsum(equation = var_57863_equation_0, values = (var_57705_cast_fp16, var_57622_cast_fp16))[name = tensor<string, []>("op_57863_cast_fp16")];
            tensor<fp16, []> var_57864_to_fp16 = const()[name = tensor<string, []>("op_57864_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4953_cast_fp16 = mul(x = var_57863_cast_fp16, y = var_57864_to_fp16)[name = tensor<string, []>("aw_4953_cast_fp16")];
            tensor<string, []> var_57867_equation_0 = const()[name = tensor<string, []>("op_57867_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_57867_cast_fp16 = einsum(equation = var_57867_equation_0, values = (var_57709_cast_fp16, var_57626_cast_fp16))[name = tensor<string, []>("op_57867_cast_fp16")];
            tensor<fp16, []> var_57868_to_fp16 = const()[name = tensor<string, []>("op_57868_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4955_cast_fp16 = mul(x = var_57867_cast_fp16, y = var_57868_to_fp16)[name = tensor<string, []>("aw_4955_cast_fp16")];
            tensor<string, []> var_57871_equation_0 = const()[name = tensor<string, []>("op_57871_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_57871_cast_fp16 = einsum(equation = var_57871_equation_0, values = (var_57713_cast_fp16, var_57630_cast_fp16))[name = tensor<string, []>("op_57871_cast_fp16")];
            tensor<fp16, []> var_57872_to_fp16 = const()[name = tensor<string, []>("op_57872_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4957_cast_fp16 = mul(x = var_57871_cast_fp16, y = var_57872_to_fp16)[name = tensor<string, []>("aw_4957_cast_fp16")];
            tensor<string, []> var_57875_equation_0 = const()[name = tensor<string, []>("op_57875_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_57875_cast_fp16 = einsum(equation = var_57875_equation_0, values = (var_57717_cast_fp16, var_57634_cast_fp16))[name = tensor<string, []>("op_57875_cast_fp16")];
            tensor<fp16, []> var_57876_to_fp16 = const()[name = tensor<string, []>("op_57876_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4959_cast_fp16 = mul(x = var_57875_cast_fp16, y = var_57876_to_fp16)[name = tensor<string, []>("aw_4959_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_57878_cast_fp16 = softmax(axis = var_30355, x = aw_4921_cast_fp16)[name = tensor<string, []>("op_57878_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_57879_cast_fp16 = softmax(axis = var_30355, x = aw_4923_cast_fp16)[name = tensor<string, []>("op_57879_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_57880_cast_fp16 = softmax(axis = var_30355, x = aw_4925_cast_fp16)[name = tensor<string, []>("op_57880_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_57881_cast_fp16 = softmax(axis = var_30355, x = aw_4927_cast_fp16)[name = tensor<string, []>("op_57881_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_57882_cast_fp16 = softmax(axis = var_30355, x = aw_4929_cast_fp16)[name = tensor<string, []>("op_57882_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_57883_cast_fp16 = softmax(axis = var_30355, x = aw_4931_cast_fp16)[name = tensor<string, []>("op_57883_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_57884_cast_fp16 = softmax(axis = var_30355, x = aw_4933_cast_fp16)[name = tensor<string, []>("op_57884_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_57885_cast_fp16 = softmax(axis = var_30355, x = aw_4935_cast_fp16)[name = tensor<string, []>("op_57885_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_57886_cast_fp16 = softmax(axis = var_30355, x = aw_4937_cast_fp16)[name = tensor<string, []>("op_57886_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_57887_cast_fp16 = softmax(axis = var_30355, x = aw_4939_cast_fp16)[name = tensor<string, []>("op_57887_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_57888_cast_fp16 = softmax(axis = var_30355, x = aw_4941_cast_fp16)[name = tensor<string, []>("op_57888_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_57889_cast_fp16 = softmax(axis = var_30355, x = aw_4943_cast_fp16)[name = tensor<string, []>("op_57889_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_57890_cast_fp16 = softmax(axis = var_30355, x = aw_4945_cast_fp16)[name = tensor<string, []>("op_57890_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_57891_cast_fp16 = softmax(axis = var_30355, x = aw_4947_cast_fp16)[name = tensor<string, []>("op_57891_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_57892_cast_fp16 = softmax(axis = var_30355, x = aw_4949_cast_fp16)[name = tensor<string, []>("op_57892_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_57893_cast_fp16 = softmax(axis = var_30355, x = aw_4951_cast_fp16)[name = tensor<string, []>("op_57893_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_57894_cast_fp16 = softmax(axis = var_30355, x = aw_4953_cast_fp16)[name = tensor<string, []>("op_57894_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_57895_cast_fp16 = softmax(axis = var_30355, x = aw_4955_cast_fp16)[name = tensor<string, []>("op_57895_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_57896_cast_fp16 = softmax(axis = var_30355, x = aw_4957_cast_fp16)[name = tensor<string, []>("op_57896_cast_fp16")];
            tensor<fp16, [2, 77, 1, 1024]> var_57897_cast_fp16 = softmax(axis = var_30355, x = aw_4959_cast_fp16)[name = tensor<string, []>("op_57897_cast_fp16")];
            tensor<string, []> var_57899_equation_0 = const()[name = tensor<string, []>("op_57899_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57899_cast_fp16 = einsum(equation = var_57899_equation_0, values = (var_57719_cast_fp16, var_57878_cast_fp16))[name = tensor<string, []>("op_57899_cast_fp16")];
            tensor<string, []> var_57901_equation_0 = const()[name = tensor<string, []>("op_57901_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57901_cast_fp16 = einsum(equation = var_57901_equation_0, values = (var_57723_cast_fp16, var_57879_cast_fp16))[name = tensor<string, []>("op_57901_cast_fp16")];
            tensor<string, []> var_57903_equation_0 = const()[name = tensor<string, []>("op_57903_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57903_cast_fp16 = einsum(equation = var_57903_equation_0, values = (var_57727_cast_fp16, var_57880_cast_fp16))[name = tensor<string, []>("op_57903_cast_fp16")];
            tensor<string, []> var_57905_equation_0 = const()[name = tensor<string, []>("op_57905_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57905_cast_fp16 = einsum(equation = var_57905_equation_0, values = (var_57731_cast_fp16, var_57881_cast_fp16))[name = tensor<string, []>("op_57905_cast_fp16")];
            tensor<string, []> var_57907_equation_0 = const()[name = tensor<string, []>("op_57907_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57907_cast_fp16 = einsum(equation = var_57907_equation_0, values = (var_57735_cast_fp16, var_57882_cast_fp16))[name = tensor<string, []>("op_57907_cast_fp16")];
            tensor<string, []> var_57909_equation_0 = const()[name = tensor<string, []>("op_57909_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57909_cast_fp16 = einsum(equation = var_57909_equation_0, values = (var_57739_cast_fp16, var_57883_cast_fp16))[name = tensor<string, []>("op_57909_cast_fp16")];
            tensor<string, []> var_57911_equation_0 = const()[name = tensor<string, []>("op_57911_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57911_cast_fp16 = einsum(equation = var_57911_equation_0, values = (var_57743_cast_fp16, var_57884_cast_fp16))[name = tensor<string, []>("op_57911_cast_fp16")];
            tensor<string, []> var_57913_equation_0 = const()[name = tensor<string, []>("op_57913_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57913_cast_fp16 = einsum(equation = var_57913_equation_0, values = (var_57747_cast_fp16, var_57885_cast_fp16))[name = tensor<string, []>("op_57913_cast_fp16")];
            tensor<string, []> var_57915_equation_0 = const()[name = tensor<string, []>("op_57915_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57915_cast_fp16 = einsum(equation = var_57915_equation_0, values = (var_57751_cast_fp16, var_57886_cast_fp16))[name = tensor<string, []>("op_57915_cast_fp16")];
            tensor<string, []> var_57917_equation_0 = const()[name = tensor<string, []>("op_57917_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57917_cast_fp16 = einsum(equation = var_57917_equation_0, values = (var_57755_cast_fp16, var_57887_cast_fp16))[name = tensor<string, []>("op_57917_cast_fp16")];
            tensor<string, []> var_57919_equation_0 = const()[name = tensor<string, []>("op_57919_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57919_cast_fp16 = einsum(equation = var_57919_equation_0, values = (var_57759_cast_fp16, var_57888_cast_fp16))[name = tensor<string, []>("op_57919_cast_fp16")];
            tensor<string, []> var_57921_equation_0 = const()[name = tensor<string, []>("op_57921_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57921_cast_fp16 = einsum(equation = var_57921_equation_0, values = (var_57763_cast_fp16, var_57889_cast_fp16))[name = tensor<string, []>("op_57921_cast_fp16")];
            tensor<string, []> var_57923_equation_0 = const()[name = tensor<string, []>("op_57923_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57923_cast_fp16 = einsum(equation = var_57923_equation_0, values = (var_57767_cast_fp16, var_57890_cast_fp16))[name = tensor<string, []>("op_57923_cast_fp16")];
            tensor<string, []> var_57925_equation_0 = const()[name = tensor<string, []>("op_57925_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57925_cast_fp16 = einsum(equation = var_57925_equation_0, values = (var_57771_cast_fp16, var_57891_cast_fp16))[name = tensor<string, []>("op_57925_cast_fp16")];
            tensor<string, []> var_57927_equation_0 = const()[name = tensor<string, []>("op_57927_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57927_cast_fp16 = einsum(equation = var_57927_equation_0, values = (var_57775_cast_fp16, var_57892_cast_fp16))[name = tensor<string, []>("op_57927_cast_fp16")];
            tensor<string, []> var_57929_equation_0 = const()[name = tensor<string, []>("op_57929_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57929_cast_fp16 = einsum(equation = var_57929_equation_0, values = (var_57779_cast_fp16, var_57893_cast_fp16))[name = tensor<string, []>("op_57929_cast_fp16")];
            tensor<string, []> var_57931_equation_0 = const()[name = tensor<string, []>("op_57931_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57931_cast_fp16 = einsum(equation = var_57931_equation_0, values = (var_57783_cast_fp16, var_57894_cast_fp16))[name = tensor<string, []>("op_57931_cast_fp16")];
            tensor<string, []> var_57933_equation_0 = const()[name = tensor<string, []>("op_57933_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57933_cast_fp16 = einsum(equation = var_57933_equation_0, values = (var_57787_cast_fp16, var_57895_cast_fp16))[name = tensor<string, []>("op_57933_cast_fp16")];
            tensor<string, []> var_57935_equation_0 = const()[name = tensor<string, []>("op_57935_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57935_cast_fp16 = einsum(equation = var_57935_equation_0, values = (var_57791_cast_fp16, var_57896_cast_fp16))[name = tensor<string, []>("op_57935_cast_fp16")];
            tensor<string, []> var_57937_equation_0 = const()[name = tensor<string, []>("op_57937_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57937_cast_fp16 = einsum(equation = var_57937_equation_0, values = (var_57795_cast_fp16, var_57897_cast_fp16))[name = tensor<string, []>("op_57937_cast_fp16")];
            tensor<bool, []> input_709_interleave_0 = const()[name = tensor<string, []>("input_709_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_709_cast_fp16 = concat(axis = var_30355, interleave = input_709_interleave_0, values = (var_57899_cast_fp16, var_57901_cast_fp16, var_57903_cast_fp16, var_57905_cast_fp16, var_57907_cast_fp16, var_57909_cast_fp16, var_57911_cast_fp16, var_57913_cast_fp16, var_57915_cast_fp16, var_57917_cast_fp16, var_57919_cast_fp16, var_57921_cast_fp16, var_57923_cast_fp16, var_57925_cast_fp16, var_57927_cast_fp16, var_57929_cast_fp16, var_57931_cast_fp16, var_57933_cast_fp16, var_57935_cast_fp16, var_57937_cast_fp16))[name = tensor<string, []>("input_709_cast_fp16")];
            tensor<string, []> var_57947_pad_type_0 = const()[name = tensor<string, []>("op_57947_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_57947_strides_0 = const()[name = tensor<string, []>("op_57947_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_57947_pad_0 = const()[name = tensor<string, []>("op_57947_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_57947_dilations_0 = const()[name = tensor<string, []>("op_57947_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_57947_groups_0 = const()[name = tensor<string, []>("op_57947_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_9_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(843210112))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(844438976))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_9_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_9_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_9_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(844439168)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_57947_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_9_attn2_to_out_0_bias_to_fp16, dilations = var_57947_dilations_0, groups = var_57947_groups_0, pad = var_57947_pad_0, pad_type = var_57947_pad_type_0, strides = var_57947_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_9_attn2_to_out_0_weight_to_fp16_palettized, x = input_709_cast_fp16)[name = tensor<string, []>("op_57947_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_383_cast_fp16 = add(x = var_57947_cast_fp16, y = inputs_381_cast_fp16)[name = tensor<string, []>("inputs_383_cast_fp16")];
            tensor<int32, [1]> input_711_axes_0 = const()[name = tensor<string, []>("input_711_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> input_711_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_711_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(844441792)))];
            tensor<fp16, [1280]> input_711_beta_0_to_fp16 = const()[name = tensor<string, []>("input_711_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(844444416)))];
            tensor<fp16, []> var_57957_to_fp16 = const()[name = tensor<string, []>("op_57957_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 1, 1024]> input_711_cast_fp16 = layer_norm(axes = input_711_axes_0, beta = input_711_beta_0_to_fp16, epsilon = var_57957_to_fp16, gamma = input_711_gamma_0_to_fp16, x = inputs_383_cast_fp16)[name = tensor<string, []>("input_711_cast_fp16")];
            tensor<string, []> var_57977_pad_type_0 = const()[name = tensor<string, []>("op_57977_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_57977_strides_0 = const()[name = tensor<string, []>("op_57977_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_57977_pad_0 = const()[name = tensor<string, []>("op_57977_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_57977_dilations_0 = const()[name = tensor<string, []>("op_57977_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_57977_groups_0 = const()[name = tensor<string, []>("op_57977_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_9_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [9830400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(844447040))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(854277504))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_9_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([10240, 1280, 1, 1])];
            tensor<fp16, [10240]> up_blocks_0_attentions_2_transformer_blocks_9_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_9_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(854277696)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_57977_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_9_ff_net_0_proj_bias_to_fp16, dilations = var_57977_dilations_0, groups = var_57977_groups_0, pad = var_57977_pad_0, pad_type = var_57977_pad_type_0, strides = var_57977_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_9_ff_net_0_proj_weight_to_fp16_palettized, x = input_711_cast_fp16)[name = tensor<string, []>("op_57977_cast_fp16")];
            tensor<int32, [2]> var_57978_split_sizes_0 = const()[name = tensor<string, []>("op_57978_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_57978_axis_0 = const()[name = tensor<string, []>("op_57978_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_57978_cast_fp16_0, tensor<fp16, [2, 5120, 1, 1024]> var_57978_cast_fp16_1 = split(axis = var_57978_axis_0, split_sizes = var_57978_split_sizes_0, x = var_57977_cast_fp16)[name = tensor<string, []>("op_57978_cast_fp16")];
            tensor<string, []> var_57980_mode_0 = const()[name = tensor<string, []>("op_57980_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_57980_cast_fp16 = gelu(mode = var_57980_mode_0, x = var_57978_cast_fp16_1)[name = tensor<string, []>("op_57980_cast_fp16")];
            tensor<fp16, [2, 5120, 1, 1024]> input_713_cast_fp16 = mul(x = var_57978_cast_fp16_0, y = var_57980_cast_fp16)[name = tensor<string, []>("input_713_cast_fp16")];
            tensor<string, []> var_57988_pad_type_0 = const()[name = tensor<string, []>("op_57988_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_57988_strides_0 = const()[name = tensor<string, []>("op_57988_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_57988_pad_0 = const()[name = tensor<string, []>("op_57988_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_57988_dilations_0 = const()[name = tensor<string, []>("op_57988_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_57988_groups_0 = const()[name = tensor<string, []>("op_57988_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_9_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [4915200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(854298240))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(859213504))), name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_9_ff_net_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 5120, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_9_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_9_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(859213696)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_57988_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_9_ff_net_2_bias_to_fp16, dilations = var_57988_dilations_0, groups = var_57988_groups_0, pad = var_57988_pad_0, pad_type = var_57988_pad_type_0, strides = var_57988_strides_0, weight = up_blocks_0_attentions_2_transformer_blocks_9_ff_net_2_weight_to_fp16_palettized, x = input_713_cast_fp16)[name = tensor<string, []>("op_57988_cast_fp16")];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_495_cast_fp16 = add(x = var_57988_cast_fp16, y = inputs_383_cast_fp16)[name = tensor<string, []>("hidden_states_495_cast_fp16")];
            tensor<int32, [4]> var_57990 = const()[name = tensor<string, []>("op_57990"), val = tensor<int32, [4]>([2, 1280, 32, 32])];
            tensor<fp16, [2, 1280, 32, 32]> input_715_cast_fp16 = reshape(shape = var_57990, x = hidden_states_495_cast_fp16)[name = tensor<string, []>("input_715_cast_fp16")];
            tensor<string, []> hidden_states_497_pad_type_0 = const()[name = tensor<string, []>("hidden_states_497_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> hidden_states_497_strides_0 = const()[name = tensor<string, []>("hidden_states_497_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_497_pad_0 = const()[name = tensor<string, []>("hidden_states_497_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_497_dilations_0 = const()[name = tensor<string, []>("hidden_states_497_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_497_groups_0 = const()[name = tensor<string, []>("hidden_states_497_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_proj_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(859216320))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(860445184))), name = tensor<string, []>("up_blocks_0_attentions_2_proj_out_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_proj_out_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_proj_out_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(860445376)))];
            tensor<fp16, [2, 1280, 32, 32]> hidden_states_497_cast_fp16 = conv(bias = up_blocks_0_attentions_2_proj_out_bias_to_fp16, dilations = hidden_states_497_dilations_0, groups = hidden_states_497_groups_0, pad = hidden_states_497_pad_0, pad_type = hidden_states_497_pad_type_0, strides = hidden_states_497_strides_0, weight = up_blocks_0_attentions_2_proj_out_weight_to_fp16_palettized, x = input_715_cast_fp16)[name = tensor<string, []>("hidden_states_497_cast_fp16")];
            tensor<fp16, [2, 1280, 32, 32]> input_717_cast_fp16 = add(x = hidden_states_497_cast_fp16, y = hidden_states_431_cast_fp16)[name = tensor<string, []>("input_717_cast_fp16")];
            tensor<fp32, []> input_719_scale_factor_height_0 = const()[name = tensor<string, []>("input_719_scale_factor_height_0"), val = tensor<fp32, []>(0x1p+1)];
            tensor<fp32, []> input_719_scale_factor_width_0 = const()[name = tensor<string, []>("input_719_scale_factor_width_0"), val = tensor<fp32, []>(0x1p+1)];
            tensor<fp16, [2, 1280, 64, 64]> input_719_cast_fp16 = upsample_nearest_neighbor(scale_factor_height = input_719_scale_factor_height_0, scale_factor_width = input_719_scale_factor_width_0, x = input_717_cast_fp16)[name = tensor<string, []>("input_719_cast_fp16")];
            tensor<string, []> hidden_states_499_pad_type_0 = const()[name = tensor<string, []>("hidden_states_499_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_499_pad_0 = const()[name = tensor<string, []>("hidden_states_499_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> hidden_states_499_strides_0 = const()[name = tensor<string, []>("hidden_states_499_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> hidden_states_499_dilations_0 = const()[name = tensor<string, []>("hidden_states_499_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_499_groups_0 = const()[name = tensor<string, []>("hidden_states_499_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1280, 1280, 3, 3]> up_blocks_0_upsamplers_0_conv_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [11059200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(860448000))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(871507264))), 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, []>(871507456)))];
            tensor<fp16, [2, 1280, 64, 64]> hidden_states_499_cast_fp16 = conv(bias = up_blocks_0_upsamplers_0_conv_bias_to_fp16, dilations = hidden_states_499_dilations_0, groups = hidden_states_499_groups_0, pad = hidden_states_499_pad_0, pad_type = hidden_states_499_pad_type_0, strides = hidden_states_499_strides_0, weight = up_blocks_0_upsamplers_0_conv_weight_to_fp16_palettized, x = input_719_cast_fp16)[name = tensor<string, []>("hidden_states_499_cast_fp16")];
            tensor<int32, []> var_58034 = const()[name = tensor<string, []>("op_58034"), val = tensor<int32, []>(1)];
            tensor<bool, []> input_721_interleave_0 = const()[name = tensor<string, []>("input_721_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 640, 64, 64]> cast_9 = cast(dtype = cast_9_dtype_0, x = input_113_cast_fp16)[name = tensor<string, []>("cast_5")];
            tensor<fp16, [2, 1920, 64, 64]> input_721_cast_fp16 = concat(axis = var_58034, interleave = input_721_interleave_0, values = (hidden_states_499_cast_fp16, cast_9))[name = tensor<string, []>("input_721_cast_fp16")];
            tensor<int32, [5]> reshape_120_shape_0 = const()[name = tensor<string, []>("reshape_120_shape_0"), val = tensor<int32, [5]>([2, 32, 60, 64, 64])];
            tensor<fp16, [2, 32, 60, 64, 64]> reshape_120_cast_fp16 = reshape(shape = reshape_120_shape_0, x = input_721_cast_fp16)[name = tensor<string, []>("reshape_120_cast_fp16")];
            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_fp16 = reduce_mean(axes = reduce_mean_90_axes_0, keep_dims = reduce_mean_90_keep_dims_0, x = reshape_120_cast_fp16)[name = tensor<string, []>("reduce_mean_90_cast_fp16")];
            tensor<fp16, [2, 32, 60, 64, 64]> sub_60_cast_fp16 = sub(x = reshape_120_cast_fp16, y = reduce_mean_90_cast_fp16)[name = tensor<string, []>("sub_60_cast_fp16")];
            tensor<fp16, [2, 32, 60, 64, 64]> square_30_cast_fp16 = square(x = sub_60_cast_fp16)[name = tensor<string, []>("square_30_cast_fp16")];
            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_fp16 = reduce_mean(axes = reduce_mean_92_axes_0, keep_dims = reduce_mean_92_keep_dims_0, x = square_30_cast_fp16)[name = tensor<string, []>("reduce_mean_92_cast_fp16")];
            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_fp16 = add(x = reduce_mean_92_cast_fp16, y = add_60_y_0_to_fp16)[name = tensor<string, []>("add_60_cast_fp16")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_30_cast_fp16 = sqrt(x = add_60_cast_fp16)[name = tensor<string, []>("sqrt_30_cast_fp16")];
            tensor<fp16, [2, 32, 60, 64, 64]> real_div_30_cast_fp16 = real_div(x = sub_60_cast_fp16, y = sqrt_30_cast_fp16)[name = tensor<string, []>("real_div_30_cast_fp16")];
            tensor<int32, [4]> reshape_121_shape_0 = const()[name = tensor<string, []>("reshape_121_shape_0"), val = tensor<int32, [4]>([2, 1920, 64, 64])];
            tensor<fp16, [2, 1920, 64, 64]> reshape_121_cast_fp16 = reshape(shape = reshape_121_shape_0, x = real_div_30_cast_fp16)[name = tensor<string, []>("reshape_121_cast_fp16")];
            tensor<fp16, [1920]> add_61_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_61_gamma_0_to_fp16"), val = tensor<fp16, [1920]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(871510080)))];
            tensor<fp16, [1920]> add_61_beta_0_to_fp16 = const()[name = tensor<string, []>("add_61_beta_0_to_fp16"), val = tensor<fp16, [1920]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(871513984)))];
            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, 1920, 64, 64]> add_61_cast_fp16 = 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_55_mean_0_to_fp16, variance = add_55_variance_0_to_fp16, x = reshape_121_cast_fp16)[name = tensor<string, []>("add_61_cast_fp16")];
            tensor<fp16, [2, 1920, 64, 64]> input_725_cast_fp16 = silu(x = add_61_cast_fp16)[name = tensor<string, []>("input_725_cast_fp16")];
            tensor<string, []> hidden_states_501_pad_type_0 = const()[name = tensor<string, []>("hidden_states_501_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_501_pad_0 = const()[name = tensor<string, []>("hidden_states_501_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> hidden_states_501_strides_0 = const()[name = tensor<string, []>("hidden_states_501_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> hidden_states_501_dilations_0 = const()[name = tensor<string, []>("hidden_states_501_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_501_groups_0 = const()[name = tensor<string, []>("hidden_states_501_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 1920, 3, 3]> up_blocks_1_resnets_0_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [8294400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(871517888))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(879812352))), name = tensor<string, []>("up_blocks_1_resnets_0_conv1_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 1920, 3, 3])];
            tensor<fp16, [640]> 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, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(879812544)))];
            tensor<fp16, [2, 640, 64, 64]> hidden_states_501_cast_fp16 = conv(bias = up_blocks_1_resnets_0_conv1_bias_to_fp16, dilations = hidden_states_501_dilations_0, groups = hidden_states_501_groups_0, pad = hidden_states_501_pad_0, pad_type = hidden_states_501_pad_type_0, strides = hidden_states_501_strides_0, weight = up_blocks_1_resnets_0_conv1_weight_to_fp16_palettized, x = input_725_cast_fp16)[name = tensor<string, []>("hidden_states_501_cast_fp16")];
            tensor<string, []> temb_23_pad_type_0 = const()[name = tensor<string, []>("temb_23_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> temb_23_strides_0 = const()[name = tensor<string, []>("temb_23_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> temb_23_pad_0 = const()[name = tensor<string, []>("temb_23_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> temb_23_dilations_0 = const()[name = tensor<string, []>("temb_23_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> temb_23_groups_0 = const()[name = tensor<string, []>("temb_23_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 1280, 1, 1]> up_blocks_1_resnets_0_time_emb_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [614400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(879813888))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(880428352))), name = tensor<string, []>("up_blocks_1_resnets_0_time_emb_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 1280, 1, 1])];
            tensor<fp16, [640]> 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, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(880428544)))];
            tensor<fp16, [2, 640, 1, 1]> temb_23_cast_fp16 = conv(bias = up_blocks_1_resnets_0_time_emb_proj_bias_to_fp16, dilations = temb_23_dilations_0, groups = temb_23_groups_0, pad = temb_23_pad_0, pad_type = temb_23_pad_type_0, strides = temb_23_strides_0, weight = up_blocks_1_resnets_0_time_emb_proj_weight_to_fp16_palettized, x = cast_2)[name = tensor<string, []>("temb_23_cast_fp16")];
            tensor<fp16, [2, 640, 64, 64]> input_729_cast_fp16 = add(x = hidden_states_501_cast_fp16, y = temb_23_cast_fp16)[name = tensor<string, []>("input_729_cast_fp16")];
            tensor<int32, [5]> reshape_124_shape_0 = const()[name = tensor<string, []>("reshape_124_shape_0"), val = tensor<int32, [5]>([2, 32, 20, 64, 64])];
            tensor<fp16, [2, 32, 20, 64, 64]> reshape_124_cast_fp16 = reshape(shape = reshape_124_shape_0, x = input_729_cast_fp16)[name = tensor<string, []>("reshape_124_cast_fp16")];
            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_fp16 = reduce_mean(axes = reduce_mean_93_axes_0, keep_dims = reduce_mean_93_keep_dims_0, x = reshape_124_cast_fp16)[name = tensor<string, []>("reduce_mean_93_cast_fp16")];
            tensor<fp16, [2, 32, 20, 64, 64]> sub_62_cast_fp16 = sub(x = reshape_124_cast_fp16, y = reduce_mean_93_cast_fp16)[name = tensor<string, []>("sub_62_cast_fp16")];
            tensor<fp16, [2, 32, 20, 64, 64]> square_31_cast_fp16 = square(x = sub_62_cast_fp16)[name = tensor<string, []>("square_31_cast_fp16")];
            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_fp16 = reduce_mean(axes = reduce_mean_95_axes_0, keep_dims = reduce_mean_95_keep_dims_0, x = square_31_cast_fp16)[name = tensor<string, []>("reduce_mean_95_cast_fp16")];
            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_fp16 = add(x = reduce_mean_95_cast_fp16, y = add_62_y_0_to_fp16)[name = tensor<string, []>("add_62_cast_fp16")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_31_cast_fp16 = sqrt(x = add_62_cast_fp16)[name = tensor<string, []>("sqrt_31_cast_fp16")];
            tensor<fp16, [2, 32, 20, 64, 64]> real_div_31_cast_fp16 = real_div(x = sub_62_cast_fp16, y = sqrt_31_cast_fp16)[name = tensor<string, []>("real_div_31_cast_fp16")];
            tensor<int32, [4]> reshape_125_shape_0 = const()[name = tensor<string, []>("reshape_125_shape_0"), val = tensor<int32, [4]>([2, 640, 64, 64])];
            tensor<fp16, [2, 640, 64, 64]> reshape_125_cast_fp16 = reshape(shape = reshape_125_shape_0, x = real_div_31_cast_fp16)[name = tensor<string, []>("reshape_125_cast_fp16")];
            tensor<fp16, [640]> add_63_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_63_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(880429888)))];
            tensor<fp16, [640]> add_63_beta_0_to_fp16 = const()[name = tensor<string, []>("add_63_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(880431232)))];
            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, 640, 64, 64]> add_63_cast_fp16 = batch_norm(beta = add_63_beta_0_to_fp16, epsilon = add_63_epsilon_0_to_fp16, gamma = add_63_gamma_0_to_fp16, mean = add_11_mean_0_to_fp16, variance = add_11_variance_0_to_fp16, x = reshape_125_cast_fp16)[name = tensor<string, []>("add_63_cast_fp16")];
            tensor<fp16, [2, 640, 64, 64]> input_733_cast_fp16 = silu(x = add_63_cast_fp16)[name = tensor<string, []>("input_733_cast_fp16")];
            tensor<string, []> hidden_states_503_pad_type_0 = const()[name = tensor<string, []>("hidden_states_503_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_503_pad_0 = const()[name = tensor<string, []>("hidden_states_503_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> hidden_states_503_strides_0 = const()[name = tensor<string, []>("hidden_states_503_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> hidden_states_503_dilations_0 = const()[name = tensor<string, []>("hidden_states_503_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_503_groups_0 = const()[name = tensor<string, []>("hidden_states_503_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 640, 3, 3]> up_blocks_1_resnets_0_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2764800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(880432576))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(883197440))), name = tensor<string, []>("up_blocks_1_resnets_0_conv2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 3, 3])];
            tensor<fp16, [640]> 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, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(883197632)))];
            tensor<fp16, [2, 640, 64, 64]> hidden_states_503_cast_fp16 = conv(bias = up_blocks_1_resnets_0_conv2_bias_to_fp16, dilations = hidden_states_503_dilations_0, groups = hidden_states_503_groups_0, pad = hidden_states_503_pad_0, pad_type = hidden_states_503_pad_type_0, strides = hidden_states_503_strides_0, weight = up_blocks_1_resnets_0_conv2_weight_to_fp16_palettized, x = input_733_cast_fp16)[name = tensor<string, []>("hidden_states_503_cast_fp16")];
            tensor<string, []> x_11_pad_type_0 = const()[name = tensor<string, []>("x_11_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> x_11_strides_0 = const()[name = tensor<string, []>("x_11_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> x_11_pad_0 = const()[name = tensor<string, []>("x_11_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> x_11_dilations_0 = const()[name = tensor<string, []>("x_11_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> x_11_groups_0 = const()[name = tensor<string, []>("x_11_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 1920, 1, 1]> up_blocks_1_resnets_0_conv_shortcut_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [921600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(883198976))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(884120640))), name = tensor<string, []>("up_blocks_1_resnets_0_conv_shortcut_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 1920, 1, 1])];
            tensor<fp16, [640]> 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, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(884120832)))];
            tensor<fp16, [2, 640, 64, 64]> x_11_cast_fp16 = conv(bias = up_blocks_1_resnets_0_conv_shortcut_bias_to_fp16, dilations = x_11_dilations_0, groups = x_11_groups_0, pad = x_11_pad_0, pad_type = x_11_pad_type_0, strides = x_11_strides_0, weight = up_blocks_1_resnets_0_conv_shortcut_weight_to_fp16_palettized, x = input_721_cast_fp16)[name = tensor<string, []>("x_11_cast_fp16")];
            tensor<fp16, [2, 640, 64, 64]> hidden_states_505_cast_fp16 = add(x = x_11_cast_fp16, y = hidden_states_503_cast_fp16)[name = tensor<string, []>("hidden_states_505_cast_fp16")];
            tensor<int32, [5]> reshape_128_shape_0 = const()[name = tensor<string, []>("reshape_128_shape_0"), val = tensor<int32, [5]>([2, 32, 20, 64, 64])];
            tensor<fp16, [2, 32, 20, 64, 64]> reshape_128_cast_fp16 = reshape(shape = reshape_128_shape_0, x = hidden_states_505_cast_fp16)[name = tensor<string, []>("reshape_128_cast_fp16")];
            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_fp16 = reduce_mean(axes = reduce_mean_96_axes_0, keep_dims = reduce_mean_96_keep_dims_0, x = reshape_128_cast_fp16)[name = tensor<string, []>("reduce_mean_96_cast_fp16")];
            tensor<fp16, [2, 32, 20, 64, 64]> sub_64_cast_fp16 = sub(x = reshape_128_cast_fp16, y = reduce_mean_96_cast_fp16)[name = tensor<string, []>("sub_64_cast_fp16")];
            tensor<fp16, [2, 32, 20, 64, 64]> square_32_cast_fp16 = square(x = sub_64_cast_fp16)[name = tensor<string, []>("square_32_cast_fp16")];
            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_fp16 = reduce_mean(axes = reduce_mean_98_axes_0, keep_dims = reduce_mean_98_keep_dims_0, x = square_32_cast_fp16)[name = tensor<string, []>("reduce_mean_98_cast_fp16")];
            tensor<fp16, []> add_64_y_0_to_fp16 = const()[name = tensor<string, []>("add_64_y_0_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_64_cast_fp16 = add(x = reduce_mean_98_cast_fp16, y = add_64_y_0_to_fp16)[name = tensor<string, []>("add_64_cast_fp16")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_32_cast_fp16 = sqrt(x = add_64_cast_fp16)[name = tensor<string, []>("sqrt_32_cast_fp16")];
            tensor<fp16, [2, 32, 20, 64, 64]> real_div_32_cast_fp16 = real_div(x = sub_64_cast_fp16, y = sqrt_32_cast_fp16)[name = tensor<string, []>("real_div_32_cast_fp16")];
            tensor<int32, [4]> reshape_129_shape_0 = const()[name = tensor<string, []>("reshape_129_shape_0"), val = tensor<int32, [4]>([2, 640, 64, 64])];
            tensor<fp16, [2, 640, 64, 64]> reshape_129_cast_fp16 = reshape(shape = reshape_129_shape_0, x = real_div_32_cast_fp16)[name = tensor<string, []>("reshape_129_cast_fp16")];
            tensor<fp16, [640]> add_65_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_65_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(884122176)))];
            tensor<fp16, [640]> add_65_beta_0_to_fp16 = const()[name = tensor<string, []>("add_65_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(884123520)))];
            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, 640, 64, 64]> add_65_cast_fp16 = 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_11_mean_0_to_fp16, variance = add_11_variance_0_to_fp16, x = reshape_129_cast_fp16)[name = tensor<string, []>("add_65_cast_fp16")];
            tensor<string, []> hidden_states_507_pad_type_0 = const()[name = tensor<string, []>("hidden_states_507_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> hidden_states_507_strides_0 = const()[name = tensor<string, []>("hidden_states_507_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_507_pad_0 = const()[name = tensor<string, []>("hidden_states_507_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_507_dilations_0 = const()[name = tensor<string, []>("hidden_states_507_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_507_groups_0 = const()[name = tensor<string, []>("hidden_states_507_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_0_proj_in_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(884124864))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(884432128))), name = tensor<string, []>("up_blocks_1_attentions_0_proj_in_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            tensor<fp16, [640]> 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, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(884432320)))];
            tensor<fp16, [2, 640, 64, 64]> hidden_states_507_cast_fp16 = conv(bias = up_blocks_1_attentions_0_proj_in_bias_to_fp16, dilations = hidden_states_507_dilations_0, groups = hidden_states_507_groups_0, pad = hidden_states_507_pad_0, pad_type = hidden_states_507_pad_type_0, strides = hidden_states_507_strides_0, weight = up_blocks_1_attentions_0_proj_in_weight_to_fp16_palettized, x = add_65_cast_fp16)[name = tensor<string, []>("hidden_states_507_cast_fp16")];
            tensor<int32, [4]> var_58121 = const()[name = tensor<string, []>("op_58121"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<fp16, [2, 640, 1, 4096]> inputs_385_cast_fp16 = reshape(shape = var_58121, x = hidden_states_507_cast_fp16)[name = tensor<string, []>("inputs_385_cast_fp16")];
            tensor<int32, [1]> hidden_states_509_axes_0 = const()[name = tensor<string, []>("hidden_states_509_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [640]> hidden_states_509_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_509_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(884433664)))];
            tensor<fp16, [640]> hidden_states_509_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_509_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(884435008)))];
            tensor<fp16, []> var_58137_to_fp16 = const()[name = tensor<string, []>("op_58137_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 640, 1, 4096]> hidden_states_509_cast_fp16 = layer_norm(axes = hidden_states_509_axes_0, beta = hidden_states_509_beta_0_to_fp16, epsilon = var_58137_to_fp16, gamma = hidden_states_509_gamma_0_to_fp16, x = inputs_385_cast_fp16)[name = tensor<string, []>("hidden_states_509_cast_fp16")];
            tensor<string, []> q_257_pad_type_0 = const()[name = tensor<string, []>("q_257_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_257_strides_0 = const()[name = tensor<string, []>("q_257_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_257_pad_0 = const()[name = tensor<string, []>("q_257_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_257_dilations_0 = const()[name = tensor<string, []>("q_257_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_257_groups_0 = const()[name = tensor<string, []>("q_257_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(884436352))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(884743616))), name = tensor<string, []>("up_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, 4096]> q_257_cast_fp16 = conv(dilations = q_257_dilations_0, groups = q_257_groups_0, pad = q_257_pad_0, pad_type = q_257_pad_type_0, strides = q_257_strides_0, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_509_cast_fp16)[name = tensor<string, []>("q_257_cast_fp16")];
            tensor<string, []> k_513_pad_type_0 = const()[name = tensor<string, []>("k_513_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_513_strides_0 = const()[name = tensor<string, []>("k_513_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_513_pad_0 = const()[name = tensor<string, []>("k_513_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_513_dilations_0 = const()[name = tensor<string, []>("k_513_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_513_groups_0 = const()[name = tensor<string, []>("k_513_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(884743808))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(885051072))), name = tensor<string, []>("up_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, 4096]> k_513_cast_fp16 = conv(dilations = k_513_dilations_0, groups = k_513_groups_0, pad = k_513_pad_0, pad_type = k_513_pad_type_0, strides = k_513_strides_0, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_509_cast_fp16)[name = tensor<string, []>("k_513_cast_fp16")];
            tensor<string, []> v_257_pad_type_0 = const()[name = tensor<string, []>("v_257_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_257_strides_0 = const()[name = tensor<string, []>("v_257_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_257_pad_0 = const()[name = tensor<string, []>("v_257_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_257_dilations_0 = const()[name = tensor<string, []>("v_257_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_257_groups_0 = const()[name = tensor<string, []>("v_257_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(885051264))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(885358528))), name = tensor<string, []>("up_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, 4096]> v_257_cast_fp16 = conv(dilations = v_257_dilations_0, groups = v_257_groups_0, pad = v_257_pad_0, pad_type = v_257_pad_type_0, strides = v_257_strides_0, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_509_cast_fp16)[name = tensor<string, []>("v_257_cast_fp16")];
            tensor<int32, [4]> var_58170_begin_0 = const()[name = tensor<string, []>("op_58170_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_58170_end_0 = const()[name = tensor<string, []>("op_58170_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_58170_end_mask_0 = const()[name = tensor<string, []>("op_58170_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58170_cast_fp16 = slice_by_index(begin = var_58170_begin_0, end = var_58170_end_0, end_mask = var_58170_end_mask_0, x = q_257_cast_fp16)[name = tensor<string, []>("op_58170_cast_fp16")];
            tensor<int32, [4]> var_58174_begin_0 = const()[name = tensor<string, []>("op_58174_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_58174_end_0 = const()[name = tensor<string, []>("op_58174_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_58174_end_mask_0 = const()[name = tensor<string, []>("op_58174_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58174_cast_fp16 = slice_by_index(begin = var_58174_begin_0, end = var_58174_end_0, end_mask = var_58174_end_mask_0, x = q_257_cast_fp16)[name = tensor<string, []>("op_58174_cast_fp16")];
            tensor<int32, [4]> var_58178_begin_0 = const()[name = tensor<string, []>("op_58178_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_58178_end_0 = const()[name = tensor<string, []>("op_58178_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_58178_end_mask_0 = const()[name = tensor<string, []>("op_58178_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58178_cast_fp16 = slice_by_index(begin = var_58178_begin_0, end = var_58178_end_0, end_mask = var_58178_end_mask_0, x = q_257_cast_fp16)[name = tensor<string, []>("op_58178_cast_fp16")];
            tensor<int32, [4]> var_58182_begin_0 = const()[name = tensor<string, []>("op_58182_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_58182_end_0 = const()[name = tensor<string, []>("op_58182_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_58182_end_mask_0 = const()[name = tensor<string, []>("op_58182_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58182_cast_fp16 = slice_by_index(begin = var_58182_begin_0, end = var_58182_end_0, end_mask = var_58182_end_mask_0, x = q_257_cast_fp16)[name = tensor<string, []>("op_58182_cast_fp16")];
            tensor<int32, [4]> var_58186_begin_0 = const()[name = tensor<string, []>("op_58186_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_58186_end_0 = const()[name = tensor<string, []>("op_58186_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_58186_end_mask_0 = const()[name = tensor<string, []>("op_58186_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58186_cast_fp16 = slice_by_index(begin = var_58186_begin_0, end = var_58186_end_0, end_mask = var_58186_end_mask_0, x = q_257_cast_fp16)[name = tensor<string, []>("op_58186_cast_fp16")];
            tensor<int32, [4]> var_58190_begin_0 = const()[name = tensor<string, []>("op_58190_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_58190_end_0 = const()[name = tensor<string, []>("op_58190_end_0"), val = tensor<int32, [4]>([2, 384, 1, 4096])];
            tensor<bool, [4]> var_58190_end_mask_0 = const()[name = tensor<string, []>("op_58190_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58190_cast_fp16 = slice_by_index(begin = var_58190_begin_0, end = var_58190_end_0, end_mask = var_58190_end_mask_0, x = q_257_cast_fp16)[name = tensor<string, []>("op_58190_cast_fp16")];
            tensor<int32, [4]> var_58194_begin_0 = const()[name = tensor<string, []>("op_58194_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_58194_end_0 = const()[name = tensor<string, []>("op_58194_end_0"), val = tensor<int32, [4]>([2, 448, 1, 4096])];
            tensor<bool, [4]> var_58194_end_mask_0 = const()[name = tensor<string, []>("op_58194_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58194_cast_fp16 = slice_by_index(begin = var_58194_begin_0, end = var_58194_end_0, end_mask = var_58194_end_mask_0, x = q_257_cast_fp16)[name = tensor<string, []>("op_58194_cast_fp16")];
            tensor<int32, [4]> var_58198_begin_0 = const()[name = tensor<string, []>("op_58198_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_58198_end_0 = const()[name = tensor<string, []>("op_58198_end_0"), val = tensor<int32, [4]>([2, 512, 1, 4096])];
            tensor<bool, [4]> var_58198_end_mask_0 = const()[name = tensor<string, []>("op_58198_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58198_cast_fp16 = slice_by_index(begin = var_58198_begin_0, end = var_58198_end_0, end_mask = var_58198_end_mask_0, x = q_257_cast_fp16)[name = tensor<string, []>("op_58198_cast_fp16")];
            tensor<int32, [4]> var_58202_begin_0 = const()[name = tensor<string, []>("op_58202_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_58202_end_0 = const()[name = tensor<string, []>("op_58202_end_0"), val = tensor<int32, [4]>([2, 576, 1, 4096])];
            tensor<bool, [4]> var_58202_end_mask_0 = const()[name = tensor<string, []>("op_58202_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58202_cast_fp16 = slice_by_index(begin = var_58202_begin_0, end = var_58202_end_0, end_mask = var_58202_end_mask_0, x = q_257_cast_fp16)[name = tensor<string, []>("op_58202_cast_fp16")];
            tensor<int32, [4]> var_58206_begin_0 = const()[name = tensor<string, []>("op_58206_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_58206_end_0 = const()[name = tensor<string, []>("op_58206_end_0"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<bool, [4]> var_58206_end_mask_0 = const()[name = tensor<string, []>("op_58206_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58206_cast_fp16 = slice_by_index(begin = var_58206_begin_0, end = var_58206_end_0, end_mask = var_58206_end_mask_0, x = q_257_cast_fp16)[name = tensor<string, []>("op_58206_cast_fp16")];
            tensor<int32, [4]> k_515_perm_0 = const()[name = tensor<string, []>("k_515_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_58213_begin_0 = const()[name = tensor<string, []>("op_58213_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_58213_end_0 = const()[name = tensor<string, []>("op_58213_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 64])];
            tensor<bool, [4]> var_58213_end_mask_0 = const()[name = tensor<string, []>("op_58213_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 640]> k_515_cast_fp16 = transpose(perm = k_515_perm_0, x = k_513_cast_fp16)[name = tensor<string, []>("transpose_11")];
            tensor<fp16, [2, 4096, 1, 64]> var_58213_cast_fp16 = slice_by_index(begin = var_58213_begin_0, end = var_58213_end_0, end_mask = var_58213_end_mask_0, x = k_515_cast_fp16)[name = tensor<string, []>("op_58213_cast_fp16")];
            tensor<int32, [4]> var_58217_begin_0 = const()[name = tensor<string, []>("op_58217_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_58217_end_0 = const()[name = tensor<string, []>("op_58217_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 128])];
            tensor<bool, [4]> var_58217_end_mask_0 = const()[name = tensor<string, []>("op_58217_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_58217_cast_fp16 = slice_by_index(begin = var_58217_begin_0, end = var_58217_end_0, end_mask = var_58217_end_mask_0, x = k_515_cast_fp16)[name = tensor<string, []>("op_58217_cast_fp16")];
            tensor<int32, [4]> var_58221_begin_0 = const()[name = tensor<string, []>("op_58221_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_58221_end_0 = const()[name = tensor<string, []>("op_58221_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 192])];
            tensor<bool, [4]> var_58221_end_mask_0 = const()[name = tensor<string, []>("op_58221_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_58221_cast_fp16 = slice_by_index(begin = var_58221_begin_0, end = var_58221_end_0, end_mask = var_58221_end_mask_0, x = k_515_cast_fp16)[name = tensor<string, []>("op_58221_cast_fp16")];
            tensor<int32, [4]> var_58225_begin_0 = const()[name = tensor<string, []>("op_58225_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_58225_end_0 = const()[name = tensor<string, []>("op_58225_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 256])];
            tensor<bool, [4]> var_58225_end_mask_0 = const()[name = tensor<string, []>("op_58225_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_58225_cast_fp16 = slice_by_index(begin = var_58225_begin_0, end = var_58225_end_0, end_mask = var_58225_end_mask_0, x = k_515_cast_fp16)[name = tensor<string, []>("op_58225_cast_fp16")];
            tensor<int32, [4]> var_58229_begin_0 = const()[name = tensor<string, []>("op_58229_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_58229_end_0 = const()[name = tensor<string, []>("op_58229_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 320])];
            tensor<bool, [4]> var_58229_end_mask_0 = const()[name = tensor<string, []>("op_58229_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_58229_cast_fp16 = slice_by_index(begin = var_58229_begin_0, end = var_58229_end_0, end_mask = var_58229_end_mask_0, x = k_515_cast_fp16)[name = tensor<string, []>("op_58229_cast_fp16")];
            tensor<int32, [4]> var_58233_begin_0 = const()[name = tensor<string, []>("op_58233_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_58233_end_0 = const()[name = tensor<string, []>("op_58233_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 384])];
            tensor<bool, [4]> var_58233_end_mask_0 = const()[name = tensor<string, []>("op_58233_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_58233_cast_fp16 = slice_by_index(begin = var_58233_begin_0, end = var_58233_end_0, end_mask = var_58233_end_mask_0, x = k_515_cast_fp16)[name = tensor<string, []>("op_58233_cast_fp16")];
            tensor<int32, [4]> var_58237_begin_0 = const()[name = tensor<string, []>("op_58237_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_58237_end_0 = const()[name = tensor<string, []>("op_58237_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 448])];
            tensor<bool, [4]> var_58237_end_mask_0 = const()[name = tensor<string, []>("op_58237_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_58237_cast_fp16 = slice_by_index(begin = var_58237_begin_0, end = var_58237_end_0, end_mask = var_58237_end_mask_0, x = k_515_cast_fp16)[name = tensor<string, []>("op_58237_cast_fp16")];
            tensor<int32, [4]> var_58241_begin_0 = const()[name = tensor<string, []>("op_58241_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_58241_end_0 = const()[name = tensor<string, []>("op_58241_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 512])];
            tensor<bool, [4]> var_58241_end_mask_0 = const()[name = tensor<string, []>("op_58241_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_58241_cast_fp16 = slice_by_index(begin = var_58241_begin_0, end = var_58241_end_0, end_mask = var_58241_end_mask_0, x = k_515_cast_fp16)[name = tensor<string, []>("op_58241_cast_fp16")];
            tensor<int32, [4]> var_58245_begin_0 = const()[name = tensor<string, []>("op_58245_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_58245_end_0 = const()[name = tensor<string, []>("op_58245_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 576])];
            tensor<bool, [4]> var_58245_end_mask_0 = const()[name = tensor<string, []>("op_58245_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_58245_cast_fp16 = slice_by_index(begin = var_58245_begin_0, end = var_58245_end_0, end_mask = var_58245_end_mask_0, x = k_515_cast_fp16)[name = tensor<string, []>("op_58245_cast_fp16")];
            tensor<int32, [4]> var_58249_begin_0 = const()[name = tensor<string, []>("op_58249_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_58249_end_0 = const()[name = tensor<string, []>("op_58249_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 640])];
            tensor<bool, [4]> var_58249_end_mask_0 = const()[name = tensor<string, []>("op_58249_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_58249_cast_fp16 = slice_by_index(begin = var_58249_begin_0, end = var_58249_end_0, end_mask = var_58249_end_mask_0, x = k_515_cast_fp16)[name = tensor<string, []>("op_58249_cast_fp16")];
            tensor<int32, [4]> var_58251_begin_0 = const()[name = tensor<string, []>("op_58251_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_58251_end_0 = const()[name = tensor<string, []>("op_58251_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_58251_end_mask_0 = const()[name = tensor<string, []>("op_58251_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58251_cast_fp16 = slice_by_index(begin = var_58251_begin_0, end = var_58251_end_0, end_mask = var_58251_end_mask_0, x = v_257_cast_fp16)[name = tensor<string, []>("op_58251_cast_fp16")];
            tensor<int32, [4]> var_58255_begin_0 = const()[name = tensor<string, []>("op_58255_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_58255_end_0 = const()[name = tensor<string, []>("op_58255_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_58255_end_mask_0 = const()[name = tensor<string, []>("op_58255_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58255_cast_fp16 = slice_by_index(begin = var_58255_begin_0, end = var_58255_end_0, end_mask = var_58255_end_mask_0, x = v_257_cast_fp16)[name = tensor<string, []>("op_58255_cast_fp16")];
            tensor<int32, [4]> var_58259_begin_0 = const()[name = tensor<string, []>("op_58259_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_58259_end_0 = const()[name = tensor<string, []>("op_58259_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_58259_end_mask_0 = const()[name = tensor<string, []>("op_58259_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58259_cast_fp16 = slice_by_index(begin = var_58259_begin_0, end = var_58259_end_0, end_mask = var_58259_end_mask_0, x = v_257_cast_fp16)[name = tensor<string, []>("op_58259_cast_fp16")];
            tensor<int32, [4]> var_58263_begin_0 = const()[name = tensor<string, []>("op_58263_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_58263_end_0 = const()[name = tensor<string, []>("op_58263_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_58263_end_mask_0 = const()[name = tensor<string, []>("op_58263_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58263_cast_fp16 = slice_by_index(begin = var_58263_begin_0, end = var_58263_end_0, end_mask = var_58263_end_mask_0, x = v_257_cast_fp16)[name = tensor<string, []>("op_58263_cast_fp16")];
            tensor<int32, [4]> var_58267_begin_0 = const()[name = tensor<string, []>("op_58267_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_58267_end_0 = const()[name = tensor<string, []>("op_58267_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_58267_end_mask_0 = const()[name = tensor<string, []>("op_58267_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58267_cast_fp16 = slice_by_index(begin = var_58267_begin_0, end = var_58267_end_0, end_mask = var_58267_end_mask_0, x = v_257_cast_fp16)[name = tensor<string, []>("op_58267_cast_fp16")];
            tensor<int32, [4]> var_58271_begin_0 = const()[name = tensor<string, []>("op_58271_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_58271_end_0 = const()[name = tensor<string, []>("op_58271_end_0"), val = tensor<int32, [4]>([2, 384, 1, 4096])];
            tensor<bool, [4]> var_58271_end_mask_0 = const()[name = tensor<string, []>("op_58271_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58271_cast_fp16 = slice_by_index(begin = var_58271_begin_0, end = var_58271_end_0, end_mask = var_58271_end_mask_0, x = v_257_cast_fp16)[name = tensor<string, []>("op_58271_cast_fp16")];
            tensor<int32, [4]> var_58275_begin_0 = const()[name = tensor<string, []>("op_58275_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_58275_end_0 = const()[name = tensor<string, []>("op_58275_end_0"), val = tensor<int32, [4]>([2, 448, 1, 4096])];
            tensor<bool, [4]> var_58275_end_mask_0 = const()[name = tensor<string, []>("op_58275_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58275_cast_fp16 = slice_by_index(begin = var_58275_begin_0, end = var_58275_end_0, end_mask = var_58275_end_mask_0, x = v_257_cast_fp16)[name = tensor<string, []>("op_58275_cast_fp16")];
            tensor<int32, [4]> var_58279_begin_0 = const()[name = tensor<string, []>("op_58279_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_58279_end_0 = const()[name = tensor<string, []>("op_58279_end_0"), val = tensor<int32, [4]>([2, 512, 1, 4096])];
            tensor<bool, [4]> var_58279_end_mask_0 = const()[name = tensor<string, []>("op_58279_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58279_cast_fp16 = slice_by_index(begin = var_58279_begin_0, end = var_58279_end_0, end_mask = var_58279_end_mask_0, x = v_257_cast_fp16)[name = tensor<string, []>("op_58279_cast_fp16")];
            tensor<int32, [4]> var_58283_begin_0 = const()[name = tensor<string, []>("op_58283_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_58283_end_0 = const()[name = tensor<string, []>("op_58283_end_0"), val = tensor<int32, [4]>([2, 576, 1, 4096])];
            tensor<bool, [4]> var_58283_end_mask_0 = const()[name = tensor<string, []>("op_58283_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58283_cast_fp16 = slice_by_index(begin = var_58283_begin_0, end = var_58283_end_0, end_mask = var_58283_end_mask_0, x = v_257_cast_fp16)[name = tensor<string, []>("op_58283_cast_fp16")];
            tensor<int32, [4]> var_58287_begin_0 = const()[name = tensor<string, []>("op_58287_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_58287_end_0 = const()[name = tensor<string, []>("op_58287_end_0"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<bool, [4]> var_58287_end_mask_0 = const()[name = tensor<string, []>("op_58287_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58287_cast_fp16 = slice_by_index(begin = var_58287_begin_0, end = var_58287_end_0, end_mask = var_58287_end_mask_0, x = v_257_cast_fp16)[name = tensor<string, []>("op_58287_cast_fp16")];
            tensor<string, []> var_58291_equation_0 = const()[name = tensor<string, []>("op_58291_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58291_cast_fp16 = einsum(equation = var_58291_equation_0, values = (var_58213_cast_fp16, var_58170_cast_fp16))[name = tensor<string, []>("op_58291_cast_fp16")];
            tensor<fp16, []> var_58292_to_fp16 = const()[name = tensor<string, []>("op_58292_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_4961_cast_fp16 = mul(x = var_58291_cast_fp16, y = var_58292_to_fp16)[name = tensor<string, []>("aw_4961_cast_fp16")];
            tensor<string, []> var_58295_equation_0 = const()[name = tensor<string, []>("op_58295_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58295_cast_fp16 = einsum(equation = var_58295_equation_0, values = (var_58217_cast_fp16, var_58174_cast_fp16))[name = tensor<string, []>("op_58295_cast_fp16")];
            tensor<fp16, []> var_58296_to_fp16 = const()[name = tensor<string, []>("op_58296_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_4963_cast_fp16 = mul(x = var_58295_cast_fp16, y = var_58296_to_fp16)[name = tensor<string, []>("aw_4963_cast_fp16")];
            tensor<string, []> var_58299_equation_0 = const()[name = tensor<string, []>("op_58299_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58299_cast_fp16 = einsum(equation = var_58299_equation_0, values = (var_58221_cast_fp16, var_58178_cast_fp16))[name = tensor<string, []>("op_58299_cast_fp16")];
            tensor<fp16, []> var_58300_to_fp16 = const()[name = tensor<string, []>("op_58300_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_4965_cast_fp16 = mul(x = var_58299_cast_fp16, y = var_58300_to_fp16)[name = tensor<string, []>("aw_4965_cast_fp16")];
            tensor<string, []> var_58303_equation_0 = const()[name = tensor<string, []>("op_58303_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58303_cast_fp16 = einsum(equation = var_58303_equation_0, values = (var_58225_cast_fp16, var_58182_cast_fp16))[name = tensor<string, []>("op_58303_cast_fp16")];
            tensor<fp16, []> var_58304_to_fp16 = const()[name = tensor<string, []>("op_58304_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_4967_cast_fp16 = mul(x = var_58303_cast_fp16, y = var_58304_to_fp16)[name = tensor<string, []>("aw_4967_cast_fp16")];
            tensor<string, []> var_58307_equation_0 = const()[name = tensor<string, []>("op_58307_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58307_cast_fp16 = einsum(equation = var_58307_equation_0, values = (var_58229_cast_fp16, var_58186_cast_fp16))[name = tensor<string, []>("op_58307_cast_fp16")];
            tensor<fp16, []> var_58308_to_fp16 = const()[name = tensor<string, []>("op_58308_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_4969_cast_fp16 = mul(x = var_58307_cast_fp16, y = var_58308_to_fp16)[name = tensor<string, []>("aw_4969_cast_fp16")];
            tensor<string, []> var_58311_equation_0 = const()[name = tensor<string, []>("op_58311_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58311_cast_fp16 = einsum(equation = var_58311_equation_0, values = (var_58233_cast_fp16, var_58190_cast_fp16))[name = tensor<string, []>("op_58311_cast_fp16")];
            tensor<fp16, []> var_58312_to_fp16 = const()[name = tensor<string, []>("op_58312_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_4971_cast_fp16 = mul(x = var_58311_cast_fp16, y = var_58312_to_fp16)[name = tensor<string, []>("aw_4971_cast_fp16")];
            tensor<string, []> var_58315_equation_0 = const()[name = tensor<string, []>("op_58315_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58315_cast_fp16 = einsum(equation = var_58315_equation_0, values = (var_58237_cast_fp16, var_58194_cast_fp16))[name = tensor<string, []>("op_58315_cast_fp16")];
            tensor<fp16, []> var_58316_to_fp16 = const()[name = tensor<string, []>("op_58316_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_4973_cast_fp16 = mul(x = var_58315_cast_fp16, y = var_58316_to_fp16)[name = tensor<string, []>("aw_4973_cast_fp16")];
            tensor<string, []> var_58319_equation_0 = const()[name = tensor<string, []>("op_58319_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58319_cast_fp16 = einsum(equation = var_58319_equation_0, values = (var_58241_cast_fp16, var_58198_cast_fp16))[name = tensor<string, []>("op_58319_cast_fp16")];
            tensor<fp16, []> var_58320_to_fp16 = const()[name = tensor<string, []>("op_58320_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_4975_cast_fp16 = mul(x = var_58319_cast_fp16, y = var_58320_to_fp16)[name = tensor<string, []>("aw_4975_cast_fp16")];
            tensor<string, []> var_58323_equation_0 = const()[name = tensor<string, []>("op_58323_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58323_cast_fp16 = einsum(equation = var_58323_equation_0, values = (var_58245_cast_fp16, var_58202_cast_fp16))[name = tensor<string, []>("op_58323_cast_fp16")];
            tensor<fp16, []> var_58324_to_fp16 = const()[name = tensor<string, []>("op_58324_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_4977_cast_fp16 = mul(x = var_58323_cast_fp16, y = var_58324_to_fp16)[name = tensor<string, []>("aw_4977_cast_fp16")];
            tensor<string, []> var_58327_equation_0 = const()[name = tensor<string, []>("op_58327_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58327_cast_fp16 = einsum(equation = var_58327_equation_0, values = (var_58249_cast_fp16, var_58206_cast_fp16))[name = tensor<string, []>("op_58327_cast_fp16")];
            tensor<fp16, []> var_58328_to_fp16 = const()[name = tensor<string, []>("op_58328_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_4979_cast_fp16 = mul(x = var_58327_cast_fp16, y = var_58328_to_fp16)[name = tensor<string, []>("aw_4979_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58330_cast_fp16 = softmax(axis = var_58034, x = aw_4961_cast_fp16)[name = tensor<string, []>("op_58330_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58331_cast_fp16 = softmax(axis = var_58034, x = aw_4963_cast_fp16)[name = tensor<string, []>("op_58331_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58332_cast_fp16 = softmax(axis = var_58034, x = aw_4965_cast_fp16)[name = tensor<string, []>("op_58332_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58333_cast_fp16 = softmax(axis = var_58034, x = aw_4967_cast_fp16)[name = tensor<string, []>("op_58333_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58334_cast_fp16 = softmax(axis = var_58034, x = aw_4969_cast_fp16)[name = tensor<string, []>("op_58334_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58335_cast_fp16 = softmax(axis = var_58034, x = aw_4971_cast_fp16)[name = tensor<string, []>("op_58335_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58336_cast_fp16 = softmax(axis = var_58034, x = aw_4973_cast_fp16)[name = tensor<string, []>("op_58336_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58337_cast_fp16 = softmax(axis = var_58034, x = aw_4975_cast_fp16)[name = tensor<string, []>("op_58337_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58338_cast_fp16 = softmax(axis = var_58034, x = aw_4977_cast_fp16)[name = tensor<string, []>("op_58338_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58339_cast_fp16 = softmax(axis = var_58034, x = aw_4979_cast_fp16)[name = tensor<string, []>("op_58339_cast_fp16")];
            tensor<string, []> var_58341_equation_0 = const()[name = tensor<string, []>("op_58341_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58341_cast_fp16 = einsum(equation = var_58341_equation_0, values = (var_58251_cast_fp16, var_58330_cast_fp16))[name = tensor<string, []>("op_58341_cast_fp16")];
            tensor<string, []> var_58343_equation_0 = const()[name = tensor<string, []>("op_58343_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58343_cast_fp16 = einsum(equation = var_58343_equation_0, values = (var_58255_cast_fp16, var_58331_cast_fp16))[name = tensor<string, []>("op_58343_cast_fp16")];
            tensor<string, []> var_58345_equation_0 = const()[name = tensor<string, []>("op_58345_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58345_cast_fp16 = einsum(equation = var_58345_equation_0, values = (var_58259_cast_fp16, var_58332_cast_fp16))[name = tensor<string, []>("op_58345_cast_fp16")];
            tensor<string, []> var_58347_equation_0 = const()[name = tensor<string, []>("op_58347_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58347_cast_fp16 = einsum(equation = var_58347_equation_0, values = (var_58263_cast_fp16, var_58333_cast_fp16))[name = tensor<string, []>("op_58347_cast_fp16")];
            tensor<string, []> var_58349_equation_0 = const()[name = tensor<string, []>("op_58349_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58349_cast_fp16 = einsum(equation = var_58349_equation_0, values = (var_58267_cast_fp16, var_58334_cast_fp16))[name = tensor<string, []>("op_58349_cast_fp16")];
            tensor<string, []> var_58351_equation_0 = const()[name = tensor<string, []>("op_58351_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58351_cast_fp16 = einsum(equation = var_58351_equation_0, values = (var_58271_cast_fp16, var_58335_cast_fp16))[name = tensor<string, []>("op_58351_cast_fp16")];
            tensor<string, []> var_58353_equation_0 = const()[name = tensor<string, []>("op_58353_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58353_cast_fp16 = einsum(equation = var_58353_equation_0, values = (var_58275_cast_fp16, var_58336_cast_fp16))[name = tensor<string, []>("op_58353_cast_fp16")];
            tensor<string, []> var_58355_equation_0 = const()[name = tensor<string, []>("op_58355_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58355_cast_fp16 = einsum(equation = var_58355_equation_0, values = (var_58279_cast_fp16, var_58337_cast_fp16))[name = tensor<string, []>("op_58355_cast_fp16")];
            tensor<string, []> var_58357_equation_0 = const()[name = tensor<string, []>("op_58357_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58357_cast_fp16 = einsum(equation = var_58357_equation_0, values = (var_58283_cast_fp16, var_58338_cast_fp16))[name = tensor<string, []>("op_58357_cast_fp16")];
            tensor<string, []> var_58359_equation_0 = const()[name = tensor<string, []>("op_58359_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58359_cast_fp16 = einsum(equation = var_58359_equation_0, values = (var_58287_cast_fp16, var_58339_cast_fp16))[name = tensor<string, []>("op_58359_cast_fp16")];
            tensor<bool, []> input_737_interleave_0 = const()[name = tensor<string, []>("input_737_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 640, 1, 4096]> input_737_cast_fp16 = concat(axis = var_58034, interleave = input_737_interleave_0, values = (var_58341_cast_fp16, var_58343_cast_fp16, var_58345_cast_fp16, var_58347_cast_fp16, var_58349_cast_fp16, var_58351_cast_fp16, var_58353_cast_fp16, var_58355_cast_fp16, var_58357_cast_fp16, var_58359_cast_fp16))[name = tensor<string, []>("input_737_cast_fp16")];
            tensor<string, []> var_58369_pad_type_0 = const()[name = tensor<string, []>("op_58369_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_58369_strides_0 = const()[name = tensor<string, []>("op_58369_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_58369_pad_0 = const()[name = tensor<string, []>("op_58369_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_58369_dilations_0 = const()[name = tensor<string, []>("op_58369_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_58369_groups_0 = const()[name = tensor<string, []>("op_58369_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 640, 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, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(885358720))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(885665984))), name = tensor<string, []>("up_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]> 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, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(885666176)))];
            tensor<fp16, [2, 640, 1, 4096]> var_58369_cast_fp16 = conv(bias = up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_58369_dilations_0, groups = var_58369_groups_0, pad = var_58369_pad_0, pad_type = var_58369_pad_type_0, strides = var_58369_strides_0, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized, x = input_737_cast_fp16)[name = tensor<string, []>("op_58369_cast_fp16")];
            tensor<fp16, [2, 640, 1, 4096]> inputs_387_cast_fp16 = add(x = var_58369_cast_fp16, y = inputs_385_cast_fp16)[name = tensor<string, []>("inputs_387_cast_fp16")];
            tensor<int32, [1]> hidden_states_511_axes_0 = const()[name = tensor<string, []>("hidden_states_511_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [640]> hidden_states_511_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_511_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(885667520)))];
            tensor<fp16, [640]> hidden_states_511_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_511_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(885668864)))];
            tensor<fp16, []> var_58379_to_fp16 = const()[name = tensor<string, []>("op_58379_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 640, 1, 4096]> hidden_states_511_cast_fp16 = layer_norm(axes = hidden_states_511_axes_0, beta = hidden_states_511_beta_0_to_fp16, epsilon = var_58379_to_fp16, gamma = hidden_states_511_gamma_0_to_fp16, x = inputs_387_cast_fp16)[name = tensor<string, []>("hidden_states_511_cast_fp16")];
            tensor<string, []> q_259_pad_type_0 = const()[name = tensor<string, []>("q_259_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_259_strides_0 = const()[name = tensor<string, []>("q_259_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_259_pad_0 = const()[name = tensor<string, []>("q_259_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_259_dilations_0 = const()[name = tensor<string, []>("q_259_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_259_groups_0 = const()[name = tensor<string, []>("q_259_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(885670208))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(885977472))), name = tensor<string, []>("up_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, 4096]> q_259_cast_fp16 = conv(dilations = q_259_dilations_0, groups = q_259_groups_0, pad = q_259_pad_0, pad_type = q_259_pad_type_0, strides = q_259_strides_0, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_511_cast_fp16)[name = tensor<string, []>("q_259_cast_fp16")];
            tensor<string, []> k_517_pad_type_0 = const()[name = tensor<string, []>("k_517_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_517_strides_0 = const()[name = tensor<string, []>("k_517_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_517_pad_0 = const()[name = tensor<string, []>("k_517_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_517_dilations_0 = const()[name = tensor<string, []>("k_517_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_517_groups_0 = const()[name = tensor<string, []>("k_517_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 2048, 1, 1]> up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [983040]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(885977664))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(886960768))), name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 2048, 1, 1])];
            tensor<fp16, [2, 640, 1, 77]> k_517_cast_fp16 = conv(dilations = k_517_dilations_0, groups = k_517_groups_0, pad = k_517_pad_0, pad_type = k_517_pad_type_0, strides = k_517_strides_0, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("k_517_cast_fp16")];
            tensor<string, []> v_259_pad_type_0 = const()[name = tensor<string, []>("v_259_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_259_strides_0 = const()[name = tensor<string, []>("v_259_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_259_pad_0 = const()[name = tensor<string, []>("v_259_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_259_dilations_0 = const()[name = tensor<string, []>("v_259_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_259_groups_0 = const()[name = tensor<string, []>("v_259_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 2048, 1, 1]> up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [983040]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(886960960))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(887944064))), name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 2048, 1, 1])];
            tensor<fp16, [2, 640, 1, 77]> v_259_cast_fp16 = conv(dilations = v_259_dilations_0, groups = v_259_groups_0, pad = v_259_pad_0, pad_type = v_259_pad_type_0, strides = v_259_strides_0, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("v_259_cast_fp16")];
            tensor<int32, [4]> var_58412_begin_0 = const()[name = tensor<string, []>("op_58412_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_58412_end_0 = const()[name = tensor<string, []>("op_58412_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_58412_end_mask_0 = const()[name = tensor<string, []>("op_58412_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58412_cast_fp16 = slice_by_index(begin = var_58412_begin_0, end = var_58412_end_0, end_mask = var_58412_end_mask_0, x = q_259_cast_fp16)[name = tensor<string, []>("op_58412_cast_fp16")];
            tensor<int32, [4]> var_58416_begin_0 = const()[name = tensor<string, []>("op_58416_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_58416_end_0 = const()[name = tensor<string, []>("op_58416_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_58416_end_mask_0 = const()[name = tensor<string, []>("op_58416_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58416_cast_fp16 = slice_by_index(begin = var_58416_begin_0, end = var_58416_end_0, end_mask = var_58416_end_mask_0, x = q_259_cast_fp16)[name = tensor<string, []>("op_58416_cast_fp16")];
            tensor<int32, [4]> var_58420_begin_0 = const()[name = tensor<string, []>("op_58420_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_58420_end_0 = const()[name = tensor<string, []>("op_58420_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_58420_end_mask_0 = const()[name = tensor<string, []>("op_58420_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58420_cast_fp16 = slice_by_index(begin = var_58420_begin_0, end = var_58420_end_0, end_mask = var_58420_end_mask_0, x = q_259_cast_fp16)[name = tensor<string, []>("op_58420_cast_fp16")];
            tensor<int32, [4]> var_58424_begin_0 = const()[name = tensor<string, []>("op_58424_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_58424_end_0 = const()[name = tensor<string, []>("op_58424_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_58424_end_mask_0 = const()[name = tensor<string, []>("op_58424_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58424_cast_fp16 = slice_by_index(begin = var_58424_begin_0, end = var_58424_end_0, end_mask = var_58424_end_mask_0, x = q_259_cast_fp16)[name = tensor<string, []>("op_58424_cast_fp16")];
            tensor<int32, [4]> var_58428_begin_0 = const()[name = tensor<string, []>("op_58428_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_58428_end_0 = const()[name = tensor<string, []>("op_58428_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_58428_end_mask_0 = const()[name = tensor<string, []>("op_58428_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58428_cast_fp16 = slice_by_index(begin = var_58428_begin_0, end = var_58428_end_0, end_mask = var_58428_end_mask_0, x = q_259_cast_fp16)[name = tensor<string, []>("op_58428_cast_fp16")];
            tensor<int32, [4]> var_58432_begin_0 = const()[name = tensor<string, []>("op_58432_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_58432_end_0 = const()[name = tensor<string, []>("op_58432_end_0"), val = tensor<int32, [4]>([2, 384, 1, 4096])];
            tensor<bool, [4]> var_58432_end_mask_0 = const()[name = tensor<string, []>("op_58432_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58432_cast_fp16 = slice_by_index(begin = var_58432_begin_0, end = var_58432_end_0, end_mask = var_58432_end_mask_0, x = q_259_cast_fp16)[name = tensor<string, []>("op_58432_cast_fp16")];
            tensor<int32, [4]> var_58436_begin_0 = const()[name = tensor<string, []>("op_58436_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_58436_end_0 = const()[name = tensor<string, []>("op_58436_end_0"), val = tensor<int32, [4]>([2, 448, 1, 4096])];
            tensor<bool, [4]> var_58436_end_mask_0 = const()[name = tensor<string, []>("op_58436_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58436_cast_fp16 = slice_by_index(begin = var_58436_begin_0, end = var_58436_end_0, end_mask = var_58436_end_mask_0, x = q_259_cast_fp16)[name = tensor<string, []>("op_58436_cast_fp16")];
            tensor<int32, [4]> var_58440_begin_0 = const()[name = tensor<string, []>("op_58440_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_58440_end_0 = const()[name = tensor<string, []>("op_58440_end_0"), val = tensor<int32, [4]>([2, 512, 1, 4096])];
            tensor<bool, [4]> var_58440_end_mask_0 = const()[name = tensor<string, []>("op_58440_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58440_cast_fp16 = slice_by_index(begin = var_58440_begin_0, end = var_58440_end_0, end_mask = var_58440_end_mask_0, x = q_259_cast_fp16)[name = tensor<string, []>("op_58440_cast_fp16")];
            tensor<int32, [4]> var_58444_begin_0 = const()[name = tensor<string, []>("op_58444_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_58444_end_0 = const()[name = tensor<string, []>("op_58444_end_0"), val = tensor<int32, [4]>([2, 576, 1, 4096])];
            tensor<bool, [4]> var_58444_end_mask_0 = const()[name = tensor<string, []>("op_58444_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58444_cast_fp16 = slice_by_index(begin = var_58444_begin_0, end = var_58444_end_0, end_mask = var_58444_end_mask_0, x = q_259_cast_fp16)[name = tensor<string, []>("op_58444_cast_fp16")];
            tensor<int32, [4]> var_58448_begin_0 = const()[name = tensor<string, []>("op_58448_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_58448_end_0 = const()[name = tensor<string, []>("op_58448_end_0"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<bool, [4]> var_58448_end_mask_0 = const()[name = tensor<string, []>("op_58448_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58448_cast_fp16 = slice_by_index(begin = var_58448_begin_0, end = var_58448_end_0, end_mask = var_58448_end_mask_0, x = q_259_cast_fp16)[name = tensor<string, []>("op_58448_cast_fp16")];
            tensor<int32, [4]> k_519_perm_0 = const()[name = tensor<string, []>("k_519_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_58455_begin_0 = const()[name = tensor<string, []>("op_58455_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_58455_end_0 = const()[name = tensor<string, []>("op_58455_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_58455_end_mask_0 = const()[name = tensor<string, []>("op_58455_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 640]> k_519_cast_fp16 = transpose(perm = k_519_perm_0, x = k_517_cast_fp16)[name = tensor<string, []>("transpose_10")];
            tensor<fp16, [2, 77, 1, 64]> var_58455_cast_fp16 = slice_by_index(begin = var_58455_begin_0, end = var_58455_end_0, end_mask = var_58455_end_mask_0, x = k_519_cast_fp16)[name = tensor<string, []>("op_58455_cast_fp16")];
            tensor<int32, [4]> var_58459_begin_0 = const()[name = tensor<string, []>("op_58459_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_58459_end_0 = const()[name = tensor<string, []>("op_58459_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_58459_end_mask_0 = const()[name = tensor<string, []>("op_58459_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_58459_cast_fp16 = slice_by_index(begin = var_58459_begin_0, end = var_58459_end_0, end_mask = var_58459_end_mask_0, x = k_519_cast_fp16)[name = tensor<string, []>("op_58459_cast_fp16")];
            tensor<int32, [4]> var_58463_begin_0 = const()[name = tensor<string, []>("op_58463_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_58463_end_0 = const()[name = tensor<string, []>("op_58463_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_58463_end_mask_0 = const()[name = tensor<string, []>("op_58463_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_58463_cast_fp16 = slice_by_index(begin = var_58463_begin_0, end = var_58463_end_0, end_mask = var_58463_end_mask_0, x = k_519_cast_fp16)[name = tensor<string, []>("op_58463_cast_fp16")];
            tensor<int32, [4]> var_58467_begin_0 = const()[name = tensor<string, []>("op_58467_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_58467_end_0 = const()[name = tensor<string, []>("op_58467_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_58467_end_mask_0 = const()[name = tensor<string, []>("op_58467_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_58467_cast_fp16 = slice_by_index(begin = var_58467_begin_0, end = var_58467_end_0, end_mask = var_58467_end_mask_0, x = k_519_cast_fp16)[name = tensor<string, []>("op_58467_cast_fp16")];
            tensor<int32, [4]> var_58471_begin_0 = const()[name = tensor<string, []>("op_58471_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_58471_end_0 = const()[name = tensor<string, []>("op_58471_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_58471_end_mask_0 = const()[name = tensor<string, []>("op_58471_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_58471_cast_fp16 = slice_by_index(begin = var_58471_begin_0, end = var_58471_end_0, end_mask = var_58471_end_mask_0, x = k_519_cast_fp16)[name = tensor<string, []>("op_58471_cast_fp16")];
            tensor<int32, [4]> var_58475_begin_0 = const()[name = tensor<string, []>("op_58475_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_58475_end_0 = const()[name = tensor<string, []>("op_58475_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_58475_end_mask_0 = const()[name = tensor<string, []>("op_58475_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_58475_cast_fp16 = slice_by_index(begin = var_58475_begin_0, end = var_58475_end_0, end_mask = var_58475_end_mask_0, x = k_519_cast_fp16)[name = tensor<string, []>("op_58475_cast_fp16")];
            tensor<int32, [4]> var_58479_begin_0 = const()[name = tensor<string, []>("op_58479_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_58479_end_0 = const()[name = tensor<string, []>("op_58479_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_58479_end_mask_0 = const()[name = tensor<string, []>("op_58479_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_58479_cast_fp16 = slice_by_index(begin = var_58479_begin_0, end = var_58479_end_0, end_mask = var_58479_end_mask_0, x = k_519_cast_fp16)[name = tensor<string, []>("op_58479_cast_fp16")];
            tensor<int32, [4]> var_58483_begin_0 = const()[name = tensor<string, []>("op_58483_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_58483_end_0 = const()[name = tensor<string, []>("op_58483_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_58483_end_mask_0 = const()[name = tensor<string, []>("op_58483_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_58483_cast_fp16 = slice_by_index(begin = var_58483_begin_0, end = var_58483_end_0, end_mask = var_58483_end_mask_0, x = k_519_cast_fp16)[name = tensor<string, []>("op_58483_cast_fp16")];
            tensor<int32, [4]> var_58487_begin_0 = const()[name = tensor<string, []>("op_58487_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_58487_end_0 = const()[name = tensor<string, []>("op_58487_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_58487_end_mask_0 = const()[name = tensor<string, []>("op_58487_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_58487_cast_fp16 = slice_by_index(begin = var_58487_begin_0, end = var_58487_end_0, end_mask = var_58487_end_mask_0, x = k_519_cast_fp16)[name = tensor<string, []>("op_58487_cast_fp16")];
            tensor<int32, [4]> var_58491_begin_0 = const()[name = tensor<string, []>("op_58491_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_58491_end_0 = const()[name = tensor<string, []>("op_58491_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_58491_end_mask_0 = const()[name = tensor<string, []>("op_58491_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_58491_cast_fp16 = slice_by_index(begin = var_58491_begin_0, end = var_58491_end_0, end_mask = var_58491_end_mask_0, x = k_519_cast_fp16)[name = tensor<string, []>("op_58491_cast_fp16")];
            tensor<int32, [4]> var_58493_begin_0 = const()[name = tensor<string, []>("op_58493_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_58493_end_0 = const()[name = tensor<string, []>("op_58493_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_58493_end_mask_0 = const()[name = tensor<string, []>("op_58493_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_58493_cast_fp16 = slice_by_index(begin = var_58493_begin_0, end = var_58493_end_0, end_mask = var_58493_end_mask_0, x = v_259_cast_fp16)[name = tensor<string, []>("op_58493_cast_fp16")];
            tensor<int32, [4]> var_58497_begin_0 = const()[name = tensor<string, []>("op_58497_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_58497_end_0 = const()[name = tensor<string, []>("op_58497_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_58497_end_mask_0 = const()[name = tensor<string, []>("op_58497_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_58497_cast_fp16 = slice_by_index(begin = var_58497_begin_0, end = var_58497_end_0, end_mask = var_58497_end_mask_0, x = v_259_cast_fp16)[name = tensor<string, []>("op_58497_cast_fp16")];
            tensor<int32, [4]> var_58501_begin_0 = const()[name = tensor<string, []>("op_58501_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_58501_end_0 = const()[name = tensor<string, []>("op_58501_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_58501_end_mask_0 = const()[name = tensor<string, []>("op_58501_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_58501_cast_fp16 = slice_by_index(begin = var_58501_begin_0, end = var_58501_end_0, end_mask = var_58501_end_mask_0, x = v_259_cast_fp16)[name = tensor<string, []>("op_58501_cast_fp16")];
            tensor<int32, [4]> var_58505_begin_0 = const()[name = tensor<string, []>("op_58505_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_58505_end_0 = const()[name = tensor<string, []>("op_58505_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_58505_end_mask_0 = const()[name = tensor<string, []>("op_58505_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_58505_cast_fp16 = slice_by_index(begin = var_58505_begin_0, end = var_58505_end_0, end_mask = var_58505_end_mask_0, x = v_259_cast_fp16)[name = tensor<string, []>("op_58505_cast_fp16")];
            tensor<int32, [4]> var_58509_begin_0 = const()[name = tensor<string, []>("op_58509_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_58509_end_0 = const()[name = tensor<string, []>("op_58509_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_58509_end_mask_0 = const()[name = tensor<string, []>("op_58509_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_58509_cast_fp16 = slice_by_index(begin = var_58509_begin_0, end = var_58509_end_0, end_mask = var_58509_end_mask_0, x = v_259_cast_fp16)[name = tensor<string, []>("op_58509_cast_fp16")];
            tensor<int32, [4]> var_58513_begin_0 = const()[name = tensor<string, []>("op_58513_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_58513_end_0 = const()[name = tensor<string, []>("op_58513_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_58513_end_mask_0 = const()[name = tensor<string, []>("op_58513_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_58513_cast_fp16 = slice_by_index(begin = var_58513_begin_0, end = var_58513_end_0, end_mask = var_58513_end_mask_0, x = v_259_cast_fp16)[name = tensor<string, []>("op_58513_cast_fp16")];
            tensor<int32, [4]> var_58517_begin_0 = const()[name = tensor<string, []>("op_58517_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_58517_end_0 = const()[name = tensor<string, []>("op_58517_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_58517_end_mask_0 = const()[name = tensor<string, []>("op_58517_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_58517_cast_fp16 = slice_by_index(begin = var_58517_begin_0, end = var_58517_end_0, end_mask = var_58517_end_mask_0, x = v_259_cast_fp16)[name = tensor<string, []>("op_58517_cast_fp16")];
            tensor<int32, [4]> var_58521_begin_0 = const()[name = tensor<string, []>("op_58521_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_58521_end_0 = const()[name = tensor<string, []>("op_58521_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_58521_end_mask_0 = const()[name = tensor<string, []>("op_58521_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_58521_cast_fp16 = slice_by_index(begin = var_58521_begin_0, end = var_58521_end_0, end_mask = var_58521_end_mask_0, x = v_259_cast_fp16)[name = tensor<string, []>("op_58521_cast_fp16")];
            tensor<int32, [4]> var_58525_begin_0 = const()[name = tensor<string, []>("op_58525_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_58525_end_0 = const()[name = tensor<string, []>("op_58525_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_58525_end_mask_0 = const()[name = tensor<string, []>("op_58525_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_58525_cast_fp16 = slice_by_index(begin = var_58525_begin_0, end = var_58525_end_0, end_mask = var_58525_end_mask_0, x = v_259_cast_fp16)[name = tensor<string, []>("op_58525_cast_fp16")];
            tensor<int32, [4]> var_58529_begin_0 = const()[name = tensor<string, []>("op_58529_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_58529_end_0 = const()[name = tensor<string, []>("op_58529_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_58529_end_mask_0 = const()[name = tensor<string, []>("op_58529_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_58529_cast_fp16 = slice_by_index(begin = var_58529_begin_0, end = var_58529_end_0, end_mask = var_58529_end_mask_0, x = v_259_cast_fp16)[name = tensor<string, []>("op_58529_cast_fp16")];
            tensor<string, []> var_58533_equation_0 = const()[name = tensor<string, []>("op_58533_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_58533_cast_fp16 = einsum(equation = var_58533_equation_0, values = (var_58455_cast_fp16, var_58412_cast_fp16))[name = tensor<string, []>("op_58533_cast_fp16")];
            tensor<fp16, []> var_58534_to_fp16 = const()[name = tensor<string, []>("op_58534_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_4981_cast_fp16 = mul(x = var_58533_cast_fp16, y = var_58534_to_fp16)[name = tensor<string, []>("aw_4981_cast_fp16")];
            tensor<string, []> var_58537_equation_0 = const()[name = tensor<string, []>("op_58537_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_58537_cast_fp16 = einsum(equation = var_58537_equation_0, values = (var_58459_cast_fp16, var_58416_cast_fp16))[name = tensor<string, []>("op_58537_cast_fp16")];
            tensor<fp16, []> var_58538_to_fp16 = const()[name = tensor<string, []>("op_58538_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_4983_cast_fp16 = mul(x = var_58537_cast_fp16, y = var_58538_to_fp16)[name = tensor<string, []>("aw_4983_cast_fp16")];
            tensor<string, []> var_58541_equation_0 = const()[name = tensor<string, []>("op_58541_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_58541_cast_fp16 = einsum(equation = var_58541_equation_0, values = (var_58463_cast_fp16, var_58420_cast_fp16))[name = tensor<string, []>("op_58541_cast_fp16")];
            tensor<fp16, []> var_58542_to_fp16 = const()[name = tensor<string, []>("op_58542_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_4985_cast_fp16 = mul(x = var_58541_cast_fp16, y = var_58542_to_fp16)[name = tensor<string, []>("aw_4985_cast_fp16")];
            tensor<string, []> var_58545_equation_0 = const()[name = tensor<string, []>("op_58545_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_58545_cast_fp16 = einsum(equation = var_58545_equation_0, values = (var_58467_cast_fp16, var_58424_cast_fp16))[name = tensor<string, []>("op_58545_cast_fp16")];
            tensor<fp16, []> var_58546_to_fp16 = const()[name = tensor<string, []>("op_58546_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_4987_cast_fp16 = mul(x = var_58545_cast_fp16, y = var_58546_to_fp16)[name = tensor<string, []>("aw_4987_cast_fp16")];
            tensor<string, []> var_58549_equation_0 = const()[name = tensor<string, []>("op_58549_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_58549_cast_fp16 = einsum(equation = var_58549_equation_0, values = (var_58471_cast_fp16, var_58428_cast_fp16))[name = tensor<string, []>("op_58549_cast_fp16")];
            tensor<fp16, []> var_58550_to_fp16 = const()[name = tensor<string, []>("op_58550_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_4989_cast_fp16 = mul(x = var_58549_cast_fp16, y = var_58550_to_fp16)[name = tensor<string, []>("aw_4989_cast_fp16")];
            tensor<string, []> var_58553_equation_0 = const()[name = tensor<string, []>("op_58553_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_58553_cast_fp16 = einsum(equation = var_58553_equation_0, values = (var_58475_cast_fp16, var_58432_cast_fp16))[name = tensor<string, []>("op_58553_cast_fp16")];
            tensor<fp16, []> var_58554_to_fp16 = const()[name = tensor<string, []>("op_58554_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_4991_cast_fp16 = mul(x = var_58553_cast_fp16, y = var_58554_to_fp16)[name = tensor<string, []>("aw_4991_cast_fp16")];
            tensor<string, []> var_58557_equation_0 = const()[name = tensor<string, []>("op_58557_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_58557_cast_fp16 = einsum(equation = var_58557_equation_0, values = (var_58479_cast_fp16, var_58436_cast_fp16))[name = tensor<string, []>("op_58557_cast_fp16")];
            tensor<fp16, []> var_58558_to_fp16 = const()[name = tensor<string, []>("op_58558_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_4993_cast_fp16 = mul(x = var_58557_cast_fp16, y = var_58558_to_fp16)[name = tensor<string, []>("aw_4993_cast_fp16")];
            tensor<string, []> var_58561_equation_0 = const()[name = tensor<string, []>("op_58561_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_58561_cast_fp16 = einsum(equation = var_58561_equation_0, values = (var_58483_cast_fp16, var_58440_cast_fp16))[name = tensor<string, []>("op_58561_cast_fp16")];
            tensor<fp16, []> var_58562_to_fp16 = const()[name = tensor<string, []>("op_58562_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_4995_cast_fp16 = mul(x = var_58561_cast_fp16, y = var_58562_to_fp16)[name = tensor<string, []>("aw_4995_cast_fp16")];
            tensor<string, []> var_58565_equation_0 = const()[name = tensor<string, []>("op_58565_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_58565_cast_fp16 = einsum(equation = var_58565_equation_0, values = (var_58487_cast_fp16, var_58444_cast_fp16))[name = tensor<string, []>("op_58565_cast_fp16")];
            tensor<fp16, []> var_58566_to_fp16 = const()[name = tensor<string, []>("op_58566_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_4997_cast_fp16 = mul(x = var_58565_cast_fp16, y = var_58566_to_fp16)[name = tensor<string, []>("aw_4997_cast_fp16")];
            tensor<string, []> var_58569_equation_0 = const()[name = tensor<string, []>("op_58569_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_58569_cast_fp16 = einsum(equation = var_58569_equation_0, values = (var_58491_cast_fp16, var_58448_cast_fp16))[name = tensor<string, []>("op_58569_cast_fp16")];
            tensor<fp16, []> var_58570_to_fp16 = const()[name = tensor<string, []>("op_58570_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_4999_cast_fp16 = mul(x = var_58569_cast_fp16, y = var_58570_to_fp16)[name = tensor<string, []>("aw_4999_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_58572_cast_fp16 = softmax(axis = var_58034, x = aw_4981_cast_fp16)[name = tensor<string, []>("op_58572_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_58573_cast_fp16 = softmax(axis = var_58034, x = aw_4983_cast_fp16)[name = tensor<string, []>("op_58573_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_58574_cast_fp16 = softmax(axis = var_58034, x = aw_4985_cast_fp16)[name = tensor<string, []>("op_58574_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_58575_cast_fp16 = softmax(axis = var_58034, x = aw_4987_cast_fp16)[name = tensor<string, []>("op_58575_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_58576_cast_fp16 = softmax(axis = var_58034, x = aw_4989_cast_fp16)[name = tensor<string, []>("op_58576_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_58577_cast_fp16 = softmax(axis = var_58034, x = aw_4991_cast_fp16)[name = tensor<string, []>("op_58577_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_58578_cast_fp16 = softmax(axis = var_58034, x = aw_4993_cast_fp16)[name = tensor<string, []>("op_58578_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_58579_cast_fp16 = softmax(axis = var_58034, x = aw_4995_cast_fp16)[name = tensor<string, []>("op_58579_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_58580_cast_fp16 = softmax(axis = var_58034, x = aw_4997_cast_fp16)[name = tensor<string, []>("op_58580_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_58581_cast_fp16 = softmax(axis = var_58034, x = aw_4999_cast_fp16)[name = tensor<string, []>("op_58581_cast_fp16")];
            tensor<string, []> var_58583_equation_0 = const()[name = tensor<string, []>("op_58583_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58583_cast_fp16 = einsum(equation = var_58583_equation_0, values = (var_58493_cast_fp16, var_58572_cast_fp16))[name = tensor<string, []>("op_58583_cast_fp16")];
            tensor<string, []> var_58585_equation_0 = const()[name = tensor<string, []>("op_58585_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58585_cast_fp16 = einsum(equation = var_58585_equation_0, values = (var_58497_cast_fp16, var_58573_cast_fp16))[name = tensor<string, []>("op_58585_cast_fp16")];
            tensor<string, []> var_58587_equation_0 = const()[name = tensor<string, []>("op_58587_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58587_cast_fp16 = einsum(equation = var_58587_equation_0, values = (var_58501_cast_fp16, var_58574_cast_fp16))[name = tensor<string, []>("op_58587_cast_fp16")];
            tensor<string, []> var_58589_equation_0 = const()[name = tensor<string, []>("op_58589_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58589_cast_fp16 = einsum(equation = var_58589_equation_0, values = (var_58505_cast_fp16, var_58575_cast_fp16))[name = tensor<string, []>("op_58589_cast_fp16")];
            tensor<string, []> var_58591_equation_0 = const()[name = tensor<string, []>("op_58591_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58591_cast_fp16 = einsum(equation = var_58591_equation_0, values = (var_58509_cast_fp16, var_58576_cast_fp16))[name = tensor<string, []>("op_58591_cast_fp16")];
            tensor<string, []> var_58593_equation_0 = const()[name = tensor<string, []>("op_58593_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58593_cast_fp16 = einsum(equation = var_58593_equation_0, values = (var_58513_cast_fp16, var_58577_cast_fp16))[name = tensor<string, []>("op_58593_cast_fp16")];
            tensor<string, []> var_58595_equation_0 = const()[name = tensor<string, []>("op_58595_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58595_cast_fp16 = einsum(equation = var_58595_equation_0, values = (var_58517_cast_fp16, var_58578_cast_fp16))[name = tensor<string, []>("op_58595_cast_fp16")];
            tensor<string, []> var_58597_equation_0 = const()[name = tensor<string, []>("op_58597_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58597_cast_fp16 = einsum(equation = var_58597_equation_0, values = (var_58521_cast_fp16, var_58579_cast_fp16))[name = tensor<string, []>("op_58597_cast_fp16")];
            tensor<string, []> var_58599_equation_0 = const()[name = tensor<string, []>("op_58599_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58599_cast_fp16 = einsum(equation = var_58599_equation_0, values = (var_58525_cast_fp16, var_58580_cast_fp16))[name = tensor<string, []>("op_58599_cast_fp16")];
            tensor<string, []> var_58601_equation_0 = const()[name = tensor<string, []>("op_58601_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58601_cast_fp16 = einsum(equation = var_58601_equation_0, values = (var_58529_cast_fp16, var_58581_cast_fp16))[name = tensor<string, []>("op_58601_cast_fp16")];
            tensor<bool, []> input_739_interleave_0 = const()[name = tensor<string, []>("input_739_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 640, 1, 4096]> input_739_cast_fp16 = concat(axis = var_58034, interleave = input_739_interleave_0, values = (var_58583_cast_fp16, var_58585_cast_fp16, var_58587_cast_fp16, var_58589_cast_fp16, var_58591_cast_fp16, var_58593_cast_fp16, var_58595_cast_fp16, var_58597_cast_fp16, var_58599_cast_fp16, var_58601_cast_fp16))[name = tensor<string, []>("input_739_cast_fp16")];
            tensor<string, []> var_58611_pad_type_0 = const()[name = tensor<string, []>("op_58611_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_58611_strides_0 = const()[name = tensor<string, []>("op_58611_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_58611_pad_0 = const()[name = tensor<string, []>("op_58611_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_58611_dilations_0 = const()[name = tensor<string, []>("op_58611_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_58611_groups_0 = const()[name = tensor<string, []>("op_58611_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 640, 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, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(887944256))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(888251520))), name = tensor<string, []>("up_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]> 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, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(888251712)))];
            tensor<fp16, [2, 640, 1, 4096]> var_58611_cast_fp16 = conv(bias = up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_58611_dilations_0, groups = var_58611_groups_0, pad = var_58611_pad_0, pad_type = var_58611_pad_type_0, strides = var_58611_strides_0, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized, x = input_739_cast_fp16)[name = tensor<string, []>("op_58611_cast_fp16")];
            tensor<fp16, [2, 640, 1, 4096]> inputs_389_cast_fp16 = add(x = var_58611_cast_fp16, y = inputs_387_cast_fp16)[name = tensor<string, []>("inputs_389_cast_fp16")];
            tensor<int32, [1]> input_741_axes_0 = const()[name = tensor<string, []>("input_741_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [640]> input_741_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_741_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(888253056)))];
            tensor<fp16, [640]> input_741_beta_0_to_fp16 = const()[name = tensor<string, []>("input_741_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(888254400)))];
            tensor<fp16, []> var_58621_to_fp16 = const()[name = tensor<string, []>("op_58621_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 640, 1, 4096]> input_741_cast_fp16 = layer_norm(axes = input_741_axes_0, beta = input_741_beta_0_to_fp16, epsilon = var_58621_to_fp16, gamma = input_741_gamma_0_to_fp16, x = inputs_389_cast_fp16)[name = tensor<string, []>("input_741_cast_fp16")];
            tensor<string, []> var_58641_pad_type_0 = const()[name = tensor<string, []>("op_58641_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_58641_strides_0 = const()[name = tensor<string, []>("op_58641_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_58641_pad_0 = const()[name = tensor<string, []>("op_58641_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_58641_dilations_0 = const()[name = tensor<string, []>("op_58641_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_58641_groups_0 = const()[name = tensor<string, []>("op_58641_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [5120, 640, 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, [2457600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(888255744))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(890713408))), name = tensor<string, []>("up_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]> up_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(890713600)))];
            tensor<fp16, [2, 5120, 1, 4096]> var_58641_cast_fp16 = conv(bias = up_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_58641_dilations_0, groups = var_58641_groups_0, pad = var_58641_pad_0, pad_type = var_58641_pad_type_0, strides = var_58641_strides_0, weight = up_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized, x = input_741_cast_fp16)[name = tensor<string, []>("op_58641_cast_fp16")];
            tensor<int32, [2]> var_58642_split_sizes_0 = const()[name = tensor<string, []>("op_58642_split_sizes_0"), val = tensor<int32, [2]>([2560, 2560])];
            tensor<int32, []> var_58642_axis_0 = const()[name = tensor<string, []>("op_58642_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 2560, 1, 4096]> var_58642_cast_fp16_0, tensor<fp16, [2, 2560, 1, 4096]> var_58642_cast_fp16_1 = split(axis = var_58642_axis_0, split_sizes = var_58642_split_sizes_0, x = var_58641_cast_fp16)[name = tensor<string, []>("op_58642_cast_fp16")];
            tensor<string, []> var_58644_mode_0 = const()[name = tensor<string, []>("op_58644_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 2560, 1, 4096]> var_58644_cast_fp16 = gelu(mode = var_58644_mode_0, x = var_58642_cast_fp16_1)[name = tensor<string, []>("op_58644_cast_fp16")];
            tensor<fp16, [2, 2560, 1, 4096]> input_743_cast_fp16 = mul(x = var_58642_cast_fp16_0, y = var_58644_cast_fp16)[name = tensor<string, []>("input_743_cast_fp16")];
            tensor<string, []> var_58652_pad_type_0 = const()[name = tensor<string, []>("op_58652_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_58652_strides_0 = const()[name = tensor<string, []>("op_58652_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_58652_pad_0 = const()[name = tensor<string, []>("op_58652_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_58652_dilations_0 = const()[name = tensor<string, []>("op_58652_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_58652_groups_0 = const()[name = tensor<string, []>("op_58652_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 2560, 1, 1]> up_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(890723904))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(891952768))), name = tensor<string, []>("up_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]> 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, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(891952960)))];
            tensor<fp16, [2, 640, 1, 4096]> var_58652_cast_fp16 = conv(bias = up_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_58652_dilations_0, groups = var_58652_groups_0, pad = var_58652_pad_0, pad_type = var_58652_pad_type_0, strides = var_58652_strides_0, weight = up_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_743_cast_fp16)[name = tensor<string, []>("op_58652_cast_fp16")];
            tensor<fp16, [2, 640, 1, 4096]> inputs_391_cast_fp16 = add(x = var_58652_cast_fp16, y = inputs_389_cast_fp16)[name = tensor<string, []>("inputs_391_cast_fp16")];
            tensor<int32, [1]> hidden_states_515_axes_0 = const()[name = tensor<string, []>("hidden_states_515_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [640]> hidden_states_515_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_515_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(891954304)))];
            tensor<fp16, [640]> hidden_states_515_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_515_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(891955648)))];
            tensor<fp16, []> var_58668_to_fp16 = const()[name = tensor<string, []>("op_58668_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 640, 1, 4096]> hidden_states_515_cast_fp16 = layer_norm(axes = hidden_states_515_axes_0, beta = hidden_states_515_beta_0_to_fp16, epsilon = var_58668_to_fp16, gamma = hidden_states_515_gamma_0_to_fp16, x = inputs_391_cast_fp16)[name = tensor<string, []>("hidden_states_515_cast_fp16")];
            tensor<string, []> q_261_pad_type_0 = const()[name = tensor<string, []>("q_261_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_261_strides_0 = const()[name = tensor<string, []>("q_261_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_261_pad_0 = const()[name = tensor<string, []>("q_261_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_261_dilations_0 = const()[name = tensor<string, []>("q_261_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_261_groups_0 = const()[name = tensor<string, []>("q_261_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_0_transformer_blocks_1_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(891956992))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(892264256))), name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_1_attn1_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            tensor<fp16, [2, 640, 1, 4096]> q_261_cast_fp16 = conv(dilations = q_261_dilations_0, groups = q_261_groups_0, pad = q_261_pad_0, pad_type = q_261_pad_type_0, strides = q_261_strides_0, weight = up_blocks_1_attentions_0_transformer_blocks_1_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_515_cast_fp16)[name = tensor<string, []>("q_261_cast_fp16")];
            tensor<string, []> k_521_pad_type_0 = const()[name = tensor<string, []>("k_521_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_521_strides_0 = const()[name = tensor<string, []>("k_521_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_521_pad_0 = const()[name = tensor<string, []>("k_521_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_521_dilations_0 = const()[name = tensor<string, []>("k_521_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_521_groups_0 = const()[name = tensor<string, []>("k_521_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_0_transformer_blocks_1_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(892264448))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(892571712))), name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_1_attn1_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            tensor<fp16, [2, 640, 1, 4096]> k_521_cast_fp16 = conv(dilations = k_521_dilations_0, groups = k_521_groups_0, pad = k_521_pad_0, pad_type = k_521_pad_type_0, strides = k_521_strides_0, weight = up_blocks_1_attentions_0_transformer_blocks_1_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_515_cast_fp16)[name = tensor<string, []>("k_521_cast_fp16")];
            tensor<string, []> v_261_pad_type_0 = const()[name = tensor<string, []>("v_261_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_261_strides_0 = const()[name = tensor<string, []>("v_261_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_261_pad_0 = const()[name = tensor<string, []>("v_261_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_261_dilations_0 = const()[name = tensor<string, []>("v_261_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_261_groups_0 = const()[name = tensor<string, []>("v_261_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_0_transformer_blocks_1_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(892571904))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(892879168))), name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_1_attn1_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            tensor<fp16, [2, 640, 1, 4096]> v_261_cast_fp16 = conv(dilations = v_261_dilations_0, groups = v_261_groups_0, pad = v_261_pad_0, pad_type = v_261_pad_type_0, strides = v_261_strides_0, weight = up_blocks_1_attentions_0_transformer_blocks_1_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_515_cast_fp16)[name = tensor<string, []>("v_261_cast_fp16")];
            tensor<int32, [4]> var_58701_begin_0 = const()[name = tensor<string, []>("op_58701_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_58701_end_0 = const()[name = tensor<string, []>("op_58701_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_58701_end_mask_0 = const()[name = tensor<string, []>("op_58701_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58701_cast_fp16 = slice_by_index(begin = var_58701_begin_0, end = var_58701_end_0, end_mask = var_58701_end_mask_0, x = q_261_cast_fp16)[name = tensor<string, []>("op_58701_cast_fp16")];
            tensor<int32, [4]> var_58705_begin_0 = const()[name = tensor<string, []>("op_58705_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_58705_end_0 = const()[name = tensor<string, []>("op_58705_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_58705_end_mask_0 = const()[name = tensor<string, []>("op_58705_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58705_cast_fp16 = slice_by_index(begin = var_58705_begin_0, end = var_58705_end_0, end_mask = var_58705_end_mask_0, x = q_261_cast_fp16)[name = tensor<string, []>("op_58705_cast_fp16")];
            tensor<int32, [4]> var_58709_begin_0 = const()[name = tensor<string, []>("op_58709_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_58709_end_0 = const()[name = tensor<string, []>("op_58709_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_58709_end_mask_0 = const()[name = tensor<string, []>("op_58709_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58709_cast_fp16 = slice_by_index(begin = var_58709_begin_0, end = var_58709_end_0, end_mask = var_58709_end_mask_0, x = q_261_cast_fp16)[name = tensor<string, []>("op_58709_cast_fp16")];
            tensor<int32, [4]> var_58713_begin_0 = const()[name = tensor<string, []>("op_58713_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_58713_end_0 = const()[name = tensor<string, []>("op_58713_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_58713_end_mask_0 = const()[name = tensor<string, []>("op_58713_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58713_cast_fp16 = slice_by_index(begin = var_58713_begin_0, end = var_58713_end_0, end_mask = var_58713_end_mask_0, x = q_261_cast_fp16)[name = tensor<string, []>("op_58713_cast_fp16")];
            tensor<int32, [4]> var_58717_begin_0 = const()[name = tensor<string, []>("op_58717_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_58717_end_0 = const()[name = tensor<string, []>("op_58717_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_58717_end_mask_0 = const()[name = tensor<string, []>("op_58717_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58717_cast_fp16 = slice_by_index(begin = var_58717_begin_0, end = var_58717_end_0, end_mask = var_58717_end_mask_0, x = q_261_cast_fp16)[name = tensor<string, []>("op_58717_cast_fp16")];
            tensor<int32, [4]> var_58721_begin_0 = const()[name = tensor<string, []>("op_58721_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_58721_end_0 = const()[name = tensor<string, []>("op_58721_end_0"), val = tensor<int32, [4]>([2, 384, 1, 4096])];
            tensor<bool, [4]> var_58721_end_mask_0 = const()[name = tensor<string, []>("op_58721_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58721_cast_fp16 = slice_by_index(begin = var_58721_begin_0, end = var_58721_end_0, end_mask = var_58721_end_mask_0, x = q_261_cast_fp16)[name = tensor<string, []>("op_58721_cast_fp16")];
            tensor<int32, [4]> var_58725_begin_0 = const()[name = tensor<string, []>("op_58725_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_58725_end_0 = const()[name = tensor<string, []>("op_58725_end_0"), val = tensor<int32, [4]>([2, 448, 1, 4096])];
            tensor<bool, [4]> var_58725_end_mask_0 = const()[name = tensor<string, []>("op_58725_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58725_cast_fp16 = slice_by_index(begin = var_58725_begin_0, end = var_58725_end_0, end_mask = var_58725_end_mask_0, x = q_261_cast_fp16)[name = tensor<string, []>("op_58725_cast_fp16")];
            tensor<int32, [4]> var_58729_begin_0 = const()[name = tensor<string, []>("op_58729_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_58729_end_0 = const()[name = tensor<string, []>("op_58729_end_0"), val = tensor<int32, [4]>([2, 512, 1, 4096])];
            tensor<bool, [4]> var_58729_end_mask_0 = const()[name = tensor<string, []>("op_58729_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58729_cast_fp16 = slice_by_index(begin = var_58729_begin_0, end = var_58729_end_0, end_mask = var_58729_end_mask_0, x = q_261_cast_fp16)[name = tensor<string, []>("op_58729_cast_fp16")];
            tensor<int32, [4]> var_58733_begin_0 = const()[name = tensor<string, []>("op_58733_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_58733_end_0 = const()[name = tensor<string, []>("op_58733_end_0"), val = tensor<int32, [4]>([2, 576, 1, 4096])];
            tensor<bool, [4]> var_58733_end_mask_0 = const()[name = tensor<string, []>("op_58733_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58733_cast_fp16 = slice_by_index(begin = var_58733_begin_0, end = var_58733_end_0, end_mask = var_58733_end_mask_0, x = q_261_cast_fp16)[name = tensor<string, []>("op_58733_cast_fp16")];
            tensor<int32, [4]> var_58737_begin_0 = const()[name = tensor<string, []>("op_58737_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_58737_end_0 = const()[name = tensor<string, []>("op_58737_end_0"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<bool, [4]> var_58737_end_mask_0 = const()[name = tensor<string, []>("op_58737_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58737_cast_fp16 = slice_by_index(begin = var_58737_begin_0, end = var_58737_end_0, end_mask = var_58737_end_mask_0, x = q_261_cast_fp16)[name = tensor<string, []>("op_58737_cast_fp16")];
            tensor<int32, [4]> k_523_perm_0 = const()[name = tensor<string, []>("k_523_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_58744_begin_0 = const()[name = tensor<string, []>("op_58744_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_58744_end_0 = const()[name = tensor<string, []>("op_58744_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 64])];
            tensor<bool, [4]> var_58744_end_mask_0 = const()[name = tensor<string, []>("op_58744_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 640]> k_523_cast_fp16 = transpose(perm = k_523_perm_0, x = k_521_cast_fp16)[name = tensor<string, []>("transpose_9")];
            tensor<fp16, [2, 4096, 1, 64]> var_58744_cast_fp16 = slice_by_index(begin = var_58744_begin_0, end = var_58744_end_0, end_mask = var_58744_end_mask_0, x = k_523_cast_fp16)[name = tensor<string, []>("op_58744_cast_fp16")];
            tensor<int32, [4]> var_58748_begin_0 = const()[name = tensor<string, []>("op_58748_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_58748_end_0 = const()[name = tensor<string, []>("op_58748_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 128])];
            tensor<bool, [4]> var_58748_end_mask_0 = const()[name = tensor<string, []>("op_58748_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_58748_cast_fp16 = slice_by_index(begin = var_58748_begin_0, end = var_58748_end_0, end_mask = var_58748_end_mask_0, x = k_523_cast_fp16)[name = tensor<string, []>("op_58748_cast_fp16")];
            tensor<int32, [4]> var_58752_begin_0 = const()[name = tensor<string, []>("op_58752_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_58752_end_0 = const()[name = tensor<string, []>("op_58752_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 192])];
            tensor<bool, [4]> var_58752_end_mask_0 = const()[name = tensor<string, []>("op_58752_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_58752_cast_fp16 = slice_by_index(begin = var_58752_begin_0, end = var_58752_end_0, end_mask = var_58752_end_mask_0, x = k_523_cast_fp16)[name = tensor<string, []>("op_58752_cast_fp16")];
            tensor<int32, [4]> var_58756_begin_0 = const()[name = tensor<string, []>("op_58756_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_58756_end_0 = const()[name = tensor<string, []>("op_58756_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 256])];
            tensor<bool, [4]> var_58756_end_mask_0 = const()[name = tensor<string, []>("op_58756_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_58756_cast_fp16 = slice_by_index(begin = var_58756_begin_0, end = var_58756_end_0, end_mask = var_58756_end_mask_0, x = k_523_cast_fp16)[name = tensor<string, []>("op_58756_cast_fp16")];
            tensor<int32, [4]> var_58760_begin_0 = const()[name = tensor<string, []>("op_58760_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_58760_end_0 = const()[name = tensor<string, []>("op_58760_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 320])];
            tensor<bool, [4]> var_58760_end_mask_0 = const()[name = tensor<string, []>("op_58760_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_58760_cast_fp16 = slice_by_index(begin = var_58760_begin_0, end = var_58760_end_0, end_mask = var_58760_end_mask_0, x = k_523_cast_fp16)[name = tensor<string, []>("op_58760_cast_fp16")];
            tensor<int32, [4]> var_58764_begin_0 = const()[name = tensor<string, []>("op_58764_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_58764_end_0 = const()[name = tensor<string, []>("op_58764_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 384])];
            tensor<bool, [4]> var_58764_end_mask_0 = const()[name = tensor<string, []>("op_58764_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_58764_cast_fp16 = slice_by_index(begin = var_58764_begin_0, end = var_58764_end_0, end_mask = var_58764_end_mask_0, x = k_523_cast_fp16)[name = tensor<string, []>("op_58764_cast_fp16")];
            tensor<int32, [4]> var_58768_begin_0 = const()[name = tensor<string, []>("op_58768_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_58768_end_0 = const()[name = tensor<string, []>("op_58768_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 448])];
            tensor<bool, [4]> var_58768_end_mask_0 = const()[name = tensor<string, []>("op_58768_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_58768_cast_fp16 = slice_by_index(begin = var_58768_begin_0, end = var_58768_end_0, end_mask = var_58768_end_mask_0, x = k_523_cast_fp16)[name = tensor<string, []>("op_58768_cast_fp16")];
            tensor<int32, [4]> var_58772_begin_0 = const()[name = tensor<string, []>("op_58772_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_58772_end_0 = const()[name = tensor<string, []>("op_58772_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 512])];
            tensor<bool, [4]> var_58772_end_mask_0 = const()[name = tensor<string, []>("op_58772_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_58772_cast_fp16 = slice_by_index(begin = var_58772_begin_0, end = var_58772_end_0, end_mask = var_58772_end_mask_0, x = k_523_cast_fp16)[name = tensor<string, []>("op_58772_cast_fp16")];
            tensor<int32, [4]> var_58776_begin_0 = const()[name = tensor<string, []>("op_58776_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_58776_end_0 = const()[name = tensor<string, []>("op_58776_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 576])];
            tensor<bool, [4]> var_58776_end_mask_0 = const()[name = tensor<string, []>("op_58776_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_58776_cast_fp16 = slice_by_index(begin = var_58776_begin_0, end = var_58776_end_0, end_mask = var_58776_end_mask_0, x = k_523_cast_fp16)[name = tensor<string, []>("op_58776_cast_fp16")];
            tensor<int32, [4]> var_58780_begin_0 = const()[name = tensor<string, []>("op_58780_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_58780_end_0 = const()[name = tensor<string, []>("op_58780_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 640])];
            tensor<bool, [4]> var_58780_end_mask_0 = const()[name = tensor<string, []>("op_58780_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_58780_cast_fp16 = slice_by_index(begin = var_58780_begin_0, end = var_58780_end_0, end_mask = var_58780_end_mask_0, x = k_523_cast_fp16)[name = tensor<string, []>("op_58780_cast_fp16")];
            tensor<int32, [4]> var_58782_begin_0 = const()[name = tensor<string, []>("op_58782_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_58782_end_0 = const()[name = tensor<string, []>("op_58782_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_58782_end_mask_0 = const()[name = tensor<string, []>("op_58782_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58782_cast_fp16 = slice_by_index(begin = var_58782_begin_0, end = var_58782_end_0, end_mask = var_58782_end_mask_0, x = v_261_cast_fp16)[name = tensor<string, []>("op_58782_cast_fp16")];
            tensor<int32, [4]> var_58786_begin_0 = const()[name = tensor<string, []>("op_58786_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_58786_end_0 = const()[name = tensor<string, []>("op_58786_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_58786_end_mask_0 = const()[name = tensor<string, []>("op_58786_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58786_cast_fp16 = slice_by_index(begin = var_58786_begin_0, end = var_58786_end_0, end_mask = var_58786_end_mask_0, x = v_261_cast_fp16)[name = tensor<string, []>("op_58786_cast_fp16")];
            tensor<int32, [4]> var_58790_begin_0 = const()[name = tensor<string, []>("op_58790_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_58790_end_0 = const()[name = tensor<string, []>("op_58790_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_58790_end_mask_0 = const()[name = tensor<string, []>("op_58790_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58790_cast_fp16 = slice_by_index(begin = var_58790_begin_0, end = var_58790_end_0, end_mask = var_58790_end_mask_0, x = v_261_cast_fp16)[name = tensor<string, []>("op_58790_cast_fp16")];
            tensor<int32, [4]> var_58794_begin_0 = const()[name = tensor<string, []>("op_58794_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_58794_end_0 = const()[name = tensor<string, []>("op_58794_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_58794_end_mask_0 = const()[name = tensor<string, []>("op_58794_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58794_cast_fp16 = slice_by_index(begin = var_58794_begin_0, end = var_58794_end_0, end_mask = var_58794_end_mask_0, x = v_261_cast_fp16)[name = tensor<string, []>("op_58794_cast_fp16")];
            tensor<int32, [4]> var_58798_begin_0 = const()[name = tensor<string, []>("op_58798_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_58798_end_0 = const()[name = tensor<string, []>("op_58798_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_58798_end_mask_0 = const()[name = tensor<string, []>("op_58798_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58798_cast_fp16 = slice_by_index(begin = var_58798_begin_0, end = var_58798_end_0, end_mask = var_58798_end_mask_0, x = v_261_cast_fp16)[name = tensor<string, []>("op_58798_cast_fp16")];
            tensor<int32, [4]> var_58802_begin_0 = const()[name = tensor<string, []>("op_58802_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_58802_end_0 = const()[name = tensor<string, []>("op_58802_end_0"), val = tensor<int32, [4]>([2, 384, 1, 4096])];
            tensor<bool, [4]> var_58802_end_mask_0 = const()[name = tensor<string, []>("op_58802_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58802_cast_fp16 = slice_by_index(begin = var_58802_begin_0, end = var_58802_end_0, end_mask = var_58802_end_mask_0, x = v_261_cast_fp16)[name = tensor<string, []>("op_58802_cast_fp16")];
            tensor<int32, [4]> var_58806_begin_0 = const()[name = tensor<string, []>("op_58806_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_58806_end_0 = const()[name = tensor<string, []>("op_58806_end_0"), val = tensor<int32, [4]>([2, 448, 1, 4096])];
            tensor<bool, [4]> var_58806_end_mask_0 = const()[name = tensor<string, []>("op_58806_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58806_cast_fp16 = slice_by_index(begin = var_58806_begin_0, end = var_58806_end_0, end_mask = var_58806_end_mask_0, x = v_261_cast_fp16)[name = tensor<string, []>("op_58806_cast_fp16")];
            tensor<int32, [4]> var_58810_begin_0 = const()[name = tensor<string, []>("op_58810_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_58810_end_0 = const()[name = tensor<string, []>("op_58810_end_0"), val = tensor<int32, [4]>([2, 512, 1, 4096])];
            tensor<bool, [4]> var_58810_end_mask_0 = const()[name = tensor<string, []>("op_58810_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58810_cast_fp16 = slice_by_index(begin = var_58810_begin_0, end = var_58810_end_0, end_mask = var_58810_end_mask_0, x = v_261_cast_fp16)[name = tensor<string, []>("op_58810_cast_fp16")];
            tensor<int32, [4]> var_58814_begin_0 = const()[name = tensor<string, []>("op_58814_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_58814_end_0 = const()[name = tensor<string, []>("op_58814_end_0"), val = tensor<int32, [4]>([2, 576, 1, 4096])];
            tensor<bool, [4]> var_58814_end_mask_0 = const()[name = tensor<string, []>("op_58814_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58814_cast_fp16 = slice_by_index(begin = var_58814_begin_0, end = var_58814_end_0, end_mask = var_58814_end_mask_0, x = v_261_cast_fp16)[name = tensor<string, []>("op_58814_cast_fp16")];
            tensor<int32, [4]> var_58818_begin_0 = const()[name = tensor<string, []>("op_58818_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_58818_end_0 = const()[name = tensor<string, []>("op_58818_end_0"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<bool, [4]> var_58818_end_mask_0 = const()[name = tensor<string, []>("op_58818_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58818_cast_fp16 = slice_by_index(begin = var_58818_begin_0, end = var_58818_end_0, end_mask = var_58818_end_mask_0, x = v_261_cast_fp16)[name = tensor<string, []>("op_58818_cast_fp16")];
            tensor<string, []> var_58822_equation_0 = const()[name = tensor<string, []>("op_58822_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58822_cast_fp16 = einsum(equation = var_58822_equation_0, values = (var_58744_cast_fp16, var_58701_cast_fp16))[name = tensor<string, []>("op_58822_cast_fp16")];
            tensor<fp16, []> var_58823_to_fp16 = const()[name = tensor<string, []>("op_58823_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5001_cast_fp16 = mul(x = var_58822_cast_fp16, y = var_58823_to_fp16)[name = tensor<string, []>("aw_5001_cast_fp16")];
            tensor<string, []> var_58826_equation_0 = const()[name = tensor<string, []>("op_58826_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58826_cast_fp16 = einsum(equation = var_58826_equation_0, values = (var_58748_cast_fp16, var_58705_cast_fp16))[name = tensor<string, []>("op_58826_cast_fp16")];
            tensor<fp16, []> var_58827_to_fp16 = const()[name = tensor<string, []>("op_58827_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5003_cast_fp16 = mul(x = var_58826_cast_fp16, y = var_58827_to_fp16)[name = tensor<string, []>("aw_5003_cast_fp16")];
            tensor<string, []> var_58830_equation_0 = const()[name = tensor<string, []>("op_58830_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58830_cast_fp16 = einsum(equation = var_58830_equation_0, values = (var_58752_cast_fp16, var_58709_cast_fp16))[name = tensor<string, []>("op_58830_cast_fp16")];
            tensor<fp16, []> var_58831_to_fp16 = const()[name = tensor<string, []>("op_58831_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5005_cast_fp16 = mul(x = var_58830_cast_fp16, y = var_58831_to_fp16)[name = tensor<string, []>("aw_5005_cast_fp16")];
            tensor<string, []> var_58834_equation_0 = const()[name = tensor<string, []>("op_58834_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58834_cast_fp16 = einsum(equation = var_58834_equation_0, values = (var_58756_cast_fp16, var_58713_cast_fp16))[name = tensor<string, []>("op_58834_cast_fp16")];
            tensor<fp16, []> var_58835_to_fp16 = const()[name = tensor<string, []>("op_58835_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5007_cast_fp16 = mul(x = var_58834_cast_fp16, y = var_58835_to_fp16)[name = tensor<string, []>("aw_5007_cast_fp16")];
            tensor<string, []> var_58838_equation_0 = const()[name = tensor<string, []>("op_58838_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58838_cast_fp16 = einsum(equation = var_58838_equation_0, values = (var_58760_cast_fp16, var_58717_cast_fp16))[name = tensor<string, []>("op_58838_cast_fp16")];
            tensor<fp16, []> var_58839_to_fp16 = const()[name = tensor<string, []>("op_58839_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5009_cast_fp16 = mul(x = var_58838_cast_fp16, y = var_58839_to_fp16)[name = tensor<string, []>("aw_5009_cast_fp16")];
            tensor<string, []> var_58842_equation_0 = const()[name = tensor<string, []>("op_58842_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58842_cast_fp16 = einsum(equation = var_58842_equation_0, values = (var_58764_cast_fp16, var_58721_cast_fp16))[name = tensor<string, []>("op_58842_cast_fp16")];
            tensor<fp16, []> var_58843_to_fp16 = const()[name = tensor<string, []>("op_58843_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5011_cast_fp16 = mul(x = var_58842_cast_fp16, y = var_58843_to_fp16)[name = tensor<string, []>("aw_5011_cast_fp16")];
            tensor<string, []> var_58846_equation_0 = const()[name = tensor<string, []>("op_58846_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58846_cast_fp16 = einsum(equation = var_58846_equation_0, values = (var_58768_cast_fp16, var_58725_cast_fp16))[name = tensor<string, []>("op_58846_cast_fp16")];
            tensor<fp16, []> var_58847_to_fp16 = const()[name = tensor<string, []>("op_58847_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5013_cast_fp16 = mul(x = var_58846_cast_fp16, y = var_58847_to_fp16)[name = tensor<string, []>("aw_5013_cast_fp16")];
            tensor<string, []> var_58850_equation_0 = const()[name = tensor<string, []>("op_58850_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58850_cast_fp16 = einsum(equation = var_58850_equation_0, values = (var_58772_cast_fp16, var_58729_cast_fp16))[name = tensor<string, []>("op_58850_cast_fp16")];
            tensor<fp16, []> var_58851_to_fp16 = const()[name = tensor<string, []>("op_58851_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5015_cast_fp16 = mul(x = var_58850_cast_fp16, y = var_58851_to_fp16)[name = tensor<string, []>("aw_5015_cast_fp16")];
            tensor<string, []> var_58854_equation_0 = const()[name = tensor<string, []>("op_58854_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58854_cast_fp16 = einsum(equation = var_58854_equation_0, values = (var_58776_cast_fp16, var_58733_cast_fp16))[name = tensor<string, []>("op_58854_cast_fp16")];
            tensor<fp16, []> var_58855_to_fp16 = const()[name = tensor<string, []>("op_58855_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5017_cast_fp16 = mul(x = var_58854_cast_fp16, y = var_58855_to_fp16)[name = tensor<string, []>("aw_5017_cast_fp16")];
            tensor<string, []> var_58858_equation_0 = const()[name = tensor<string, []>("op_58858_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58858_cast_fp16 = einsum(equation = var_58858_equation_0, values = (var_58780_cast_fp16, var_58737_cast_fp16))[name = tensor<string, []>("op_58858_cast_fp16")];
            tensor<fp16, []> var_58859_to_fp16 = const()[name = tensor<string, []>("op_58859_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5019_cast_fp16 = mul(x = var_58858_cast_fp16, y = var_58859_to_fp16)[name = tensor<string, []>("aw_5019_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58861_cast_fp16 = softmax(axis = var_58034, x = aw_5001_cast_fp16)[name = tensor<string, []>("op_58861_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58862_cast_fp16 = softmax(axis = var_58034, x = aw_5003_cast_fp16)[name = tensor<string, []>("op_58862_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58863_cast_fp16 = softmax(axis = var_58034, x = aw_5005_cast_fp16)[name = tensor<string, []>("op_58863_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58864_cast_fp16 = softmax(axis = var_58034, x = aw_5007_cast_fp16)[name = tensor<string, []>("op_58864_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58865_cast_fp16 = softmax(axis = var_58034, x = aw_5009_cast_fp16)[name = tensor<string, []>("op_58865_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58866_cast_fp16 = softmax(axis = var_58034, x = aw_5011_cast_fp16)[name = tensor<string, []>("op_58866_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58867_cast_fp16 = softmax(axis = var_58034, x = aw_5013_cast_fp16)[name = tensor<string, []>("op_58867_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58868_cast_fp16 = softmax(axis = var_58034, x = aw_5015_cast_fp16)[name = tensor<string, []>("op_58868_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58869_cast_fp16 = softmax(axis = var_58034, x = aw_5017_cast_fp16)[name = tensor<string, []>("op_58869_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58870_cast_fp16 = softmax(axis = var_58034, x = aw_5019_cast_fp16)[name = tensor<string, []>("op_58870_cast_fp16")];
            tensor<string, []> var_58872_equation_0 = const()[name = tensor<string, []>("op_58872_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58872_cast_fp16 = einsum(equation = var_58872_equation_0, values = (var_58782_cast_fp16, var_58861_cast_fp16))[name = tensor<string, []>("op_58872_cast_fp16")];
            tensor<string, []> var_58874_equation_0 = const()[name = tensor<string, []>("op_58874_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58874_cast_fp16 = einsum(equation = var_58874_equation_0, values = (var_58786_cast_fp16, var_58862_cast_fp16))[name = tensor<string, []>("op_58874_cast_fp16")];
            tensor<string, []> var_58876_equation_0 = const()[name = tensor<string, []>("op_58876_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58876_cast_fp16 = einsum(equation = var_58876_equation_0, values = (var_58790_cast_fp16, var_58863_cast_fp16))[name = tensor<string, []>("op_58876_cast_fp16")];
            tensor<string, []> var_58878_equation_0 = const()[name = tensor<string, []>("op_58878_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58878_cast_fp16 = einsum(equation = var_58878_equation_0, values = (var_58794_cast_fp16, var_58864_cast_fp16))[name = tensor<string, []>("op_58878_cast_fp16")];
            tensor<string, []> var_58880_equation_0 = const()[name = tensor<string, []>("op_58880_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58880_cast_fp16 = einsum(equation = var_58880_equation_0, values = (var_58798_cast_fp16, var_58865_cast_fp16))[name = tensor<string, []>("op_58880_cast_fp16")];
            tensor<string, []> var_58882_equation_0 = const()[name = tensor<string, []>("op_58882_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58882_cast_fp16 = einsum(equation = var_58882_equation_0, values = (var_58802_cast_fp16, var_58866_cast_fp16))[name = tensor<string, []>("op_58882_cast_fp16")];
            tensor<string, []> var_58884_equation_0 = const()[name = tensor<string, []>("op_58884_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58884_cast_fp16 = einsum(equation = var_58884_equation_0, values = (var_58806_cast_fp16, var_58867_cast_fp16))[name = tensor<string, []>("op_58884_cast_fp16")];
            tensor<string, []> var_58886_equation_0 = const()[name = tensor<string, []>("op_58886_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58886_cast_fp16 = einsum(equation = var_58886_equation_0, values = (var_58810_cast_fp16, var_58868_cast_fp16))[name = tensor<string, []>("op_58886_cast_fp16")];
            tensor<string, []> var_58888_equation_0 = const()[name = tensor<string, []>("op_58888_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58888_cast_fp16 = einsum(equation = var_58888_equation_0, values = (var_58814_cast_fp16, var_58869_cast_fp16))[name = tensor<string, []>("op_58888_cast_fp16")];
            tensor<string, []> var_58890_equation_0 = const()[name = tensor<string, []>("op_58890_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58890_cast_fp16 = einsum(equation = var_58890_equation_0, values = (var_58818_cast_fp16, var_58870_cast_fp16))[name = tensor<string, []>("op_58890_cast_fp16")];
            tensor<bool, []> input_745_interleave_0 = const()[name = tensor<string, []>("input_745_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 640, 1, 4096]> input_745_cast_fp16 = concat(axis = var_58034, interleave = input_745_interleave_0, values = (var_58872_cast_fp16, var_58874_cast_fp16, var_58876_cast_fp16, var_58878_cast_fp16, var_58880_cast_fp16, var_58882_cast_fp16, var_58884_cast_fp16, var_58886_cast_fp16, var_58888_cast_fp16, var_58890_cast_fp16))[name = tensor<string, []>("input_745_cast_fp16")];
            tensor<string, []> var_58900_pad_type_0 = const()[name = tensor<string, []>("op_58900_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_58900_strides_0 = const()[name = tensor<string, []>("op_58900_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_58900_pad_0 = const()[name = tensor<string, []>("op_58900_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_58900_dilations_0 = const()[name = tensor<string, []>("op_58900_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_58900_groups_0 = const()[name = tensor<string, []>("op_58900_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_0_transformer_blocks_1_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(892879360))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(893186624))), name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_1_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            tensor<fp16, [640]> up_blocks_1_attentions_0_transformer_blocks_1_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_1_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(893186816)))];
            tensor<fp16, [2, 640, 1, 4096]> var_58900_cast_fp16 = conv(bias = up_blocks_1_attentions_0_transformer_blocks_1_attn1_to_out_0_bias_to_fp16, dilations = var_58900_dilations_0, groups = var_58900_groups_0, pad = var_58900_pad_0, pad_type = var_58900_pad_type_0, strides = var_58900_strides_0, weight = up_blocks_1_attentions_0_transformer_blocks_1_attn1_to_out_0_weight_to_fp16_palettized, x = input_745_cast_fp16)[name = tensor<string, []>("op_58900_cast_fp16")];
            tensor<fp16, [2, 640, 1, 4096]> inputs_393_cast_fp16 = add(x = var_58900_cast_fp16, y = inputs_391_cast_fp16)[name = tensor<string, []>("inputs_393_cast_fp16")];
            tensor<int32, [1]> hidden_states_517_axes_0 = const()[name = tensor<string, []>("hidden_states_517_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [640]> hidden_states_517_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_517_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(893188160)))];
            tensor<fp16, [640]> hidden_states_517_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_517_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(893189504)))];
            tensor<fp16, []> var_58910_to_fp16 = const()[name = tensor<string, []>("op_58910_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 640, 1, 4096]> hidden_states_517_cast_fp16 = layer_norm(axes = hidden_states_517_axes_0, beta = hidden_states_517_beta_0_to_fp16, epsilon = var_58910_to_fp16, gamma = hidden_states_517_gamma_0_to_fp16, x = inputs_393_cast_fp16)[name = tensor<string, []>("hidden_states_517_cast_fp16")];
            tensor<string, []> q_263_pad_type_0 = const()[name = tensor<string, []>("q_263_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_263_strides_0 = const()[name = tensor<string, []>("q_263_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_263_pad_0 = const()[name = tensor<string, []>("q_263_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_263_dilations_0 = const()[name = tensor<string, []>("q_263_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_263_groups_0 = const()[name = tensor<string, []>("q_263_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_0_transformer_blocks_1_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(893190848))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(893498112))), name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_1_attn2_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            tensor<fp16, [2, 640, 1, 4096]> q_263_cast_fp16 = conv(dilations = q_263_dilations_0, groups = q_263_groups_0, pad = q_263_pad_0, pad_type = q_263_pad_type_0, strides = q_263_strides_0, weight = up_blocks_1_attentions_0_transformer_blocks_1_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_517_cast_fp16)[name = tensor<string, []>("q_263_cast_fp16")];
            tensor<string, []> k_525_pad_type_0 = const()[name = tensor<string, []>("k_525_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_525_strides_0 = const()[name = tensor<string, []>("k_525_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_525_pad_0 = const()[name = tensor<string, []>("k_525_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_525_dilations_0 = const()[name = tensor<string, []>("k_525_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_525_groups_0 = const()[name = tensor<string, []>("k_525_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 2048, 1, 1]> up_blocks_1_attentions_0_transformer_blocks_1_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [983040]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(893498304))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(894481408))), name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_1_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 2048, 1, 1])];
            tensor<fp16, [2, 640, 1, 77]> k_525_cast_fp16 = conv(dilations = k_525_dilations_0, groups = k_525_groups_0, pad = k_525_pad_0, pad_type = k_525_pad_type_0, strides = k_525_strides_0, weight = up_blocks_1_attentions_0_transformer_blocks_1_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("k_525_cast_fp16")];
            tensor<string, []> v_263_pad_type_0 = const()[name = tensor<string, []>("v_263_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_263_strides_0 = const()[name = tensor<string, []>("v_263_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_263_pad_0 = const()[name = tensor<string, []>("v_263_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_263_dilations_0 = const()[name = tensor<string, []>("v_263_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_263_groups_0 = const()[name = tensor<string, []>("v_263_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 2048, 1, 1]> up_blocks_1_attentions_0_transformer_blocks_1_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [983040]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(894481600))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(895464704))), name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_1_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 2048, 1, 1])];
            tensor<fp16, [2, 640, 1, 77]> v_263_cast_fp16 = conv(dilations = v_263_dilations_0, groups = v_263_groups_0, pad = v_263_pad_0, pad_type = v_263_pad_type_0, strides = v_263_strides_0, weight = up_blocks_1_attentions_0_transformer_blocks_1_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("v_263_cast_fp16")];
            tensor<int32, [4]> var_58943_begin_0 = const()[name = tensor<string, []>("op_58943_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_58943_end_0 = const()[name = tensor<string, []>("op_58943_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_58943_end_mask_0 = const()[name = tensor<string, []>("op_58943_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58943_cast_fp16 = slice_by_index(begin = var_58943_begin_0, end = var_58943_end_0, end_mask = var_58943_end_mask_0, x = q_263_cast_fp16)[name = tensor<string, []>("op_58943_cast_fp16")];
            tensor<int32, [4]> var_58947_begin_0 = const()[name = tensor<string, []>("op_58947_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_58947_end_0 = const()[name = tensor<string, []>("op_58947_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_58947_end_mask_0 = const()[name = tensor<string, []>("op_58947_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58947_cast_fp16 = slice_by_index(begin = var_58947_begin_0, end = var_58947_end_0, end_mask = var_58947_end_mask_0, x = q_263_cast_fp16)[name = tensor<string, []>("op_58947_cast_fp16")];
            tensor<int32, [4]> var_58951_begin_0 = const()[name = tensor<string, []>("op_58951_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_58951_end_0 = const()[name = tensor<string, []>("op_58951_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_58951_end_mask_0 = const()[name = tensor<string, []>("op_58951_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58951_cast_fp16 = slice_by_index(begin = var_58951_begin_0, end = var_58951_end_0, end_mask = var_58951_end_mask_0, x = q_263_cast_fp16)[name = tensor<string, []>("op_58951_cast_fp16")];
            tensor<int32, [4]> var_58955_begin_0 = const()[name = tensor<string, []>("op_58955_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_58955_end_0 = const()[name = tensor<string, []>("op_58955_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_58955_end_mask_0 = const()[name = tensor<string, []>("op_58955_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58955_cast_fp16 = slice_by_index(begin = var_58955_begin_0, end = var_58955_end_0, end_mask = var_58955_end_mask_0, x = q_263_cast_fp16)[name = tensor<string, []>("op_58955_cast_fp16")];
            tensor<int32, [4]> var_58959_begin_0 = const()[name = tensor<string, []>("op_58959_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_58959_end_0 = const()[name = tensor<string, []>("op_58959_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_58959_end_mask_0 = const()[name = tensor<string, []>("op_58959_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58959_cast_fp16 = slice_by_index(begin = var_58959_begin_0, end = var_58959_end_0, end_mask = var_58959_end_mask_0, x = q_263_cast_fp16)[name = tensor<string, []>("op_58959_cast_fp16")];
            tensor<int32, [4]> var_58963_begin_0 = const()[name = tensor<string, []>("op_58963_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_58963_end_0 = const()[name = tensor<string, []>("op_58963_end_0"), val = tensor<int32, [4]>([2, 384, 1, 4096])];
            tensor<bool, [4]> var_58963_end_mask_0 = const()[name = tensor<string, []>("op_58963_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58963_cast_fp16 = slice_by_index(begin = var_58963_begin_0, end = var_58963_end_0, end_mask = var_58963_end_mask_0, x = q_263_cast_fp16)[name = tensor<string, []>("op_58963_cast_fp16")];
            tensor<int32, [4]> var_58967_begin_0 = const()[name = tensor<string, []>("op_58967_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_58967_end_0 = const()[name = tensor<string, []>("op_58967_end_0"), val = tensor<int32, [4]>([2, 448, 1, 4096])];
            tensor<bool, [4]> var_58967_end_mask_0 = const()[name = tensor<string, []>("op_58967_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58967_cast_fp16 = slice_by_index(begin = var_58967_begin_0, end = var_58967_end_0, end_mask = var_58967_end_mask_0, x = q_263_cast_fp16)[name = tensor<string, []>("op_58967_cast_fp16")];
            tensor<int32, [4]> var_58971_begin_0 = const()[name = tensor<string, []>("op_58971_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_58971_end_0 = const()[name = tensor<string, []>("op_58971_end_0"), val = tensor<int32, [4]>([2, 512, 1, 4096])];
            tensor<bool, [4]> var_58971_end_mask_0 = const()[name = tensor<string, []>("op_58971_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58971_cast_fp16 = slice_by_index(begin = var_58971_begin_0, end = var_58971_end_0, end_mask = var_58971_end_mask_0, x = q_263_cast_fp16)[name = tensor<string, []>("op_58971_cast_fp16")];
            tensor<int32, [4]> var_58975_begin_0 = const()[name = tensor<string, []>("op_58975_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_58975_end_0 = const()[name = tensor<string, []>("op_58975_end_0"), val = tensor<int32, [4]>([2, 576, 1, 4096])];
            tensor<bool, [4]> var_58975_end_mask_0 = const()[name = tensor<string, []>("op_58975_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58975_cast_fp16 = slice_by_index(begin = var_58975_begin_0, end = var_58975_end_0, end_mask = var_58975_end_mask_0, x = q_263_cast_fp16)[name = tensor<string, []>("op_58975_cast_fp16")];
            tensor<int32, [4]> var_58979_begin_0 = const()[name = tensor<string, []>("op_58979_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_58979_end_0 = const()[name = tensor<string, []>("op_58979_end_0"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<bool, [4]> var_58979_end_mask_0 = const()[name = tensor<string, []>("op_58979_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58979_cast_fp16 = slice_by_index(begin = var_58979_begin_0, end = var_58979_end_0, end_mask = var_58979_end_mask_0, x = q_263_cast_fp16)[name = tensor<string, []>("op_58979_cast_fp16")];
            tensor<int32, [4]> k_527_perm_0 = const()[name = tensor<string, []>("k_527_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_58986_begin_0 = const()[name = tensor<string, []>("op_58986_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_58986_end_0 = const()[name = tensor<string, []>("op_58986_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_58986_end_mask_0 = const()[name = tensor<string, []>("op_58986_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 640]> k_527_cast_fp16 = transpose(perm = k_527_perm_0, x = k_525_cast_fp16)[name = tensor<string, []>("transpose_8")];
            tensor<fp16, [2, 77, 1, 64]> var_58986_cast_fp16 = slice_by_index(begin = var_58986_begin_0, end = var_58986_end_0, end_mask = var_58986_end_mask_0, x = k_527_cast_fp16)[name = tensor<string, []>("op_58986_cast_fp16")];
            tensor<int32, [4]> var_58990_begin_0 = const()[name = tensor<string, []>("op_58990_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_58990_end_0 = const()[name = tensor<string, []>("op_58990_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_58990_end_mask_0 = const()[name = tensor<string, []>("op_58990_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_58990_cast_fp16 = slice_by_index(begin = var_58990_begin_0, end = var_58990_end_0, end_mask = var_58990_end_mask_0, x = k_527_cast_fp16)[name = tensor<string, []>("op_58990_cast_fp16")];
            tensor<int32, [4]> var_58994_begin_0 = const()[name = tensor<string, []>("op_58994_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_58994_end_0 = const()[name = tensor<string, []>("op_58994_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_58994_end_mask_0 = const()[name = tensor<string, []>("op_58994_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_58994_cast_fp16 = slice_by_index(begin = var_58994_begin_0, end = var_58994_end_0, end_mask = var_58994_end_mask_0, x = k_527_cast_fp16)[name = tensor<string, []>("op_58994_cast_fp16")];
            tensor<int32, [4]> var_58998_begin_0 = const()[name = tensor<string, []>("op_58998_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_58998_end_0 = const()[name = tensor<string, []>("op_58998_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_58998_end_mask_0 = const()[name = tensor<string, []>("op_58998_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_58998_cast_fp16 = slice_by_index(begin = var_58998_begin_0, end = var_58998_end_0, end_mask = var_58998_end_mask_0, x = k_527_cast_fp16)[name = tensor<string, []>("op_58998_cast_fp16")];
            tensor<int32, [4]> var_59002_begin_0 = const()[name = tensor<string, []>("op_59002_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_59002_end_0 = const()[name = tensor<string, []>("op_59002_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_59002_end_mask_0 = const()[name = tensor<string, []>("op_59002_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_59002_cast_fp16 = slice_by_index(begin = var_59002_begin_0, end = var_59002_end_0, end_mask = var_59002_end_mask_0, x = k_527_cast_fp16)[name = tensor<string, []>("op_59002_cast_fp16")];
            tensor<int32, [4]> var_59006_begin_0 = const()[name = tensor<string, []>("op_59006_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_59006_end_0 = const()[name = tensor<string, []>("op_59006_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_59006_end_mask_0 = const()[name = tensor<string, []>("op_59006_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_59006_cast_fp16 = slice_by_index(begin = var_59006_begin_0, end = var_59006_end_0, end_mask = var_59006_end_mask_0, x = k_527_cast_fp16)[name = tensor<string, []>("op_59006_cast_fp16")];
            tensor<int32, [4]> var_59010_begin_0 = const()[name = tensor<string, []>("op_59010_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_59010_end_0 = const()[name = tensor<string, []>("op_59010_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_59010_end_mask_0 = const()[name = tensor<string, []>("op_59010_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_59010_cast_fp16 = slice_by_index(begin = var_59010_begin_0, end = var_59010_end_0, end_mask = var_59010_end_mask_0, x = k_527_cast_fp16)[name = tensor<string, []>("op_59010_cast_fp16")];
            tensor<int32, [4]> var_59014_begin_0 = const()[name = tensor<string, []>("op_59014_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_59014_end_0 = const()[name = tensor<string, []>("op_59014_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_59014_end_mask_0 = const()[name = tensor<string, []>("op_59014_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_59014_cast_fp16 = slice_by_index(begin = var_59014_begin_0, end = var_59014_end_0, end_mask = var_59014_end_mask_0, x = k_527_cast_fp16)[name = tensor<string, []>("op_59014_cast_fp16")];
            tensor<int32, [4]> var_59018_begin_0 = const()[name = tensor<string, []>("op_59018_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_59018_end_0 = const()[name = tensor<string, []>("op_59018_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_59018_end_mask_0 = const()[name = tensor<string, []>("op_59018_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_59018_cast_fp16 = slice_by_index(begin = var_59018_begin_0, end = var_59018_end_0, end_mask = var_59018_end_mask_0, x = k_527_cast_fp16)[name = tensor<string, []>("op_59018_cast_fp16")];
            tensor<int32, [4]> var_59022_begin_0 = const()[name = tensor<string, []>("op_59022_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_59022_end_0 = const()[name = tensor<string, []>("op_59022_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_59022_end_mask_0 = const()[name = tensor<string, []>("op_59022_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_59022_cast_fp16 = slice_by_index(begin = var_59022_begin_0, end = var_59022_end_0, end_mask = var_59022_end_mask_0, x = k_527_cast_fp16)[name = tensor<string, []>("op_59022_cast_fp16")];
            tensor<int32, [4]> var_59024_begin_0 = const()[name = tensor<string, []>("op_59024_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_59024_end_0 = const()[name = tensor<string, []>("op_59024_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_59024_end_mask_0 = const()[name = tensor<string, []>("op_59024_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_59024_cast_fp16 = slice_by_index(begin = var_59024_begin_0, end = var_59024_end_0, end_mask = var_59024_end_mask_0, x = v_263_cast_fp16)[name = tensor<string, []>("op_59024_cast_fp16")];
            tensor<int32, [4]> var_59028_begin_0 = const()[name = tensor<string, []>("op_59028_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_59028_end_0 = const()[name = tensor<string, []>("op_59028_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_59028_end_mask_0 = const()[name = tensor<string, []>("op_59028_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_59028_cast_fp16 = slice_by_index(begin = var_59028_begin_0, end = var_59028_end_0, end_mask = var_59028_end_mask_0, x = v_263_cast_fp16)[name = tensor<string, []>("op_59028_cast_fp16")];
            tensor<int32, [4]> var_59032_begin_0 = const()[name = tensor<string, []>("op_59032_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_59032_end_0 = const()[name = tensor<string, []>("op_59032_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_59032_end_mask_0 = const()[name = tensor<string, []>("op_59032_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_59032_cast_fp16 = slice_by_index(begin = var_59032_begin_0, end = var_59032_end_0, end_mask = var_59032_end_mask_0, x = v_263_cast_fp16)[name = tensor<string, []>("op_59032_cast_fp16")];
            tensor<int32, [4]> var_59036_begin_0 = const()[name = tensor<string, []>("op_59036_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_59036_end_0 = const()[name = tensor<string, []>("op_59036_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_59036_end_mask_0 = const()[name = tensor<string, []>("op_59036_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_59036_cast_fp16 = slice_by_index(begin = var_59036_begin_0, end = var_59036_end_0, end_mask = var_59036_end_mask_0, x = v_263_cast_fp16)[name = tensor<string, []>("op_59036_cast_fp16")];
            tensor<int32, [4]> var_59040_begin_0 = const()[name = tensor<string, []>("op_59040_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_59040_end_0 = const()[name = tensor<string, []>("op_59040_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_59040_end_mask_0 = const()[name = tensor<string, []>("op_59040_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_59040_cast_fp16 = slice_by_index(begin = var_59040_begin_0, end = var_59040_end_0, end_mask = var_59040_end_mask_0, x = v_263_cast_fp16)[name = tensor<string, []>("op_59040_cast_fp16")];
            tensor<int32, [4]> var_59044_begin_0 = const()[name = tensor<string, []>("op_59044_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_59044_end_0 = const()[name = tensor<string, []>("op_59044_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_59044_end_mask_0 = const()[name = tensor<string, []>("op_59044_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_59044_cast_fp16 = slice_by_index(begin = var_59044_begin_0, end = var_59044_end_0, end_mask = var_59044_end_mask_0, x = v_263_cast_fp16)[name = tensor<string, []>("op_59044_cast_fp16")];
            tensor<int32, [4]> var_59048_begin_0 = const()[name = tensor<string, []>("op_59048_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_59048_end_0 = const()[name = tensor<string, []>("op_59048_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_59048_end_mask_0 = const()[name = tensor<string, []>("op_59048_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_59048_cast_fp16 = slice_by_index(begin = var_59048_begin_0, end = var_59048_end_0, end_mask = var_59048_end_mask_0, x = v_263_cast_fp16)[name = tensor<string, []>("op_59048_cast_fp16")];
            tensor<int32, [4]> var_59052_begin_0 = const()[name = tensor<string, []>("op_59052_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_59052_end_0 = const()[name = tensor<string, []>("op_59052_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_59052_end_mask_0 = const()[name = tensor<string, []>("op_59052_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_59052_cast_fp16 = slice_by_index(begin = var_59052_begin_0, end = var_59052_end_0, end_mask = var_59052_end_mask_0, x = v_263_cast_fp16)[name = tensor<string, []>("op_59052_cast_fp16")];
            tensor<int32, [4]> var_59056_begin_0 = const()[name = tensor<string, []>("op_59056_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_59056_end_0 = const()[name = tensor<string, []>("op_59056_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_59056_end_mask_0 = const()[name = tensor<string, []>("op_59056_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_59056_cast_fp16 = slice_by_index(begin = var_59056_begin_0, end = var_59056_end_0, end_mask = var_59056_end_mask_0, x = v_263_cast_fp16)[name = tensor<string, []>("op_59056_cast_fp16")];
            tensor<int32, [4]> var_59060_begin_0 = const()[name = tensor<string, []>("op_59060_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_59060_end_0 = const()[name = tensor<string, []>("op_59060_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_59060_end_mask_0 = const()[name = tensor<string, []>("op_59060_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_59060_cast_fp16 = slice_by_index(begin = var_59060_begin_0, end = var_59060_end_0, end_mask = var_59060_end_mask_0, x = v_263_cast_fp16)[name = tensor<string, []>("op_59060_cast_fp16")];
            tensor<string, []> var_59064_equation_0 = const()[name = tensor<string, []>("op_59064_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_59064_cast_fp16 = einsum(equation = var_59064_equation_0, values = (var_58986_cast_fp16, var_58943_cast_fp16))[name = tensor<string, []>("op_59064_cast_fp16")];
            tensor<fp16, []> var_59065_to_fp16 = const()[name = tensor<string, []>("op_59065_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5021_cast_fp16 = mul(x = var_59064_cast_fp16, y = var_59065_to_fp16)[name = tensor<string, []>("aw_5021_cast_fp16")];
            tensor<string, []> var_59068_equation_0 = const()[name = tensor<string, []>("op_59068_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_59068_cast_fp16 = einsum(equation = var_59068_equation_0, values = (var_58990_cast_fp16, var_58947_cast_fp16))[name = tensor<string, []>("op_59068_cast_fp16")];
            tensor<fp16, []> var_59069_to_fp16 = const()[name = tensor<string, []>("op_59069_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5023_cast_fp16 = mul(x = var_59068_cast_fp16, y = var_59069_to_fp16)[name = tensor<string, []>("aw_5023_cast_fp16")];
            tensor<string, []> var_59072_equation_0 = const()[name = tensor<string, []>("op_59072_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_59072_cast_fp16 = einsum(equation = var_59072_equation_0, values = (var_58994_cast_fp16, var_58951_cast_fp16))[name = tensor<string, []>("op_59072_cast_fp16")];
            tensor<fp16, []> var_59073_to_fp16 = const()[name = tensor<string, []>("op_59073_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5025_cast_fp16 = mul(x = var_59072_cast_fp16, y = var_59073_to_fp16)[name = tensor<string, []>("aw_5025_cast_fp16")];
            tensor<string, []> var_59076_equation_0 = const()[name = tensor<string, []>("op_59076_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_59076_cast_fp16 = einsum(equation = var_59076_equation_0, values = (var_58998_cast_fp16, var_58955_cast_fp16))[name = tensor<string, []>("op_59076_cast_fp16")];
            tensor<fp16, []> var_59077_to_fp16 = const()[name = tensor<string, []>("op_59077_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5027_cast_fp16 = mul(x = var_59076_cast_fp16, y = var_59077_to_fp16)[name = tensor<string, []>("aw_5027_cast_fp16")];
            tensor<string, []> var_59080_equation_0 = const()[name = tensor<string, []>("op_59080_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_59080_cast_fp16 = einsum(equation = var_59080_equation_0, values = (var_59002_cast_fp16, var_58959_cast_fp16))[name = tensor<string, []>("op_59080_cast_fp16")];
            tensor<fp16, []> var_59081_to_fp16 = const()[name = tensor<string, []>("op_59081_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5029_cast_fp16 = mul(x = var_59080_cast_fp16, y = var_59081_to_fp16)[name = tensor<string, []>("aw_5029_cast_fp16")];
            tensor<string, []> var_59084_equation_0 = const()[name = tensor<string, []>("op_59084_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_59084_cast_fp16 = einsum(equation = var_59084_equation_0, values = (var_59006_cast_fp16, var_58963_cast_fp16))[name = tensor<string, []>("op_59084_cast_fp16")];
            tensor<fp16, []> var_59085_to_fp16 = const()[name = tensor<string, []>("op_59085_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5031_cast_fp16 = mul(x = var_59084_cast_fp16, y = var_59085_to_fp16)[name = tensor<string, []>("aw_5031_cast_fp16")];
            tensor<string, []> var_59088_equation_0 = const()[name = tensor<string, []>("op_59088_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_59088_cast_fp16 = einsum(equation = var_59088_equation_0, values = (var_59010_cast_fp16, var_58967_cast_fp16))[name = tensor<string, []>("op_59088_cast_fp16")];
            tensor<fp16, []> var_59089_to_fp16 = const()[name = tensor<string, []>("op_59089_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5033_cast_fp16 = mul(x = var_59088_cast_fp16, y = var_59089_to_fp16)[name = tensor<string, []>("aw_5033_cast_fp16")];
            tensor<string, []> var_59092_equation_0 = const()[name = tensor<string, []>("op_59092_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_59092_cast_fp16 = einsum(equation = var_59092_equation_0, values = (var_59014_cast_fp16, var_58971_cast_fp16))[name = tensor<string, []>("op_59092_cast_fp16")];
            tensor<fp16, []> var_59093_to_fp16 = const()[name = tensor<string, []>("op_59093_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5035_cast_fp16 = mul(x = var_59092_cast_fp16, y = var_59093_to_fp16)[name = tensor<string, []>("aw_5035_cast_fp16")];
            tensor<string, []> var_59096_equation_0 = const()[name = tensor<string, []>("op_59096_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_59096_cast_fp16 = einsum(equation = var_59096_equation_0, values = (var_59018_cast_fp16, var_58975_cast_fp16))[name = tensor<string, []>("op_59096_cast_fp16")];
            tensor<fp16, []> var_59097_to_fp16 = const()[name = tensor<string, []>("op_59097_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5037_cast_fp16 = mul(x = var_59096_cast_fp16, y = var_59097_to_fp16)[name = tensor<string, []>("aw_5037_cast_fp16")];
            tensor<string, []> var_59100_equation_0 = const()[name = tensor<string, []>("op_59100_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_59100_cast_fp16 = einsum(equation = var_59100_equation_0, values = (var_59022_cast_fp16, var_58979_cast_fp16))[name = tensor<string, []>("op_59100_cast_fp16")];
            tensor<fp16, []> var_59101_to_fp16 = const()[name = tensor<string, []>("op_59101_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5039_cast_fp16 = mul(x = var_59100_cast_fp16, y = var_59101_to_fp16)[name = tensor<string, []>("aw_5039_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_59103_cast_fp16 = softmax(axis = var_58034, x = aw_5021_cast_fp16)[name = tensor<string, []>("op_59103_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_59104_cast_fp16 = softmax(axis = var_58034, x = aw_5023_cast_fp16)[name = tensor<string, []>("op_59104_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_59105_cast_fp16 = softmax(axis = var_58034, x = aw_5025_cast_fp16)[name = tensor<string, []>("op_59105_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_59106_cast_fp16 = softmax(axis = var_58034, x = aw_5027_cast_fp16)[name = tensor<string, []>("op_59106_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_59107_cast_fp16 = softmax(axis = var_58034, x = aw_5029_cast_fp16)[name = tensor<string, []>("op_59107_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_59108_cast_fp16 = softmax(axis = var_58034, x = aw_5031_cast_fp16)[name = tensor<string, []>("op_59108_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_59109_cast_fp16 = softmax(axis = var_58034, x = aw_5033_cast_fp16)[name = tensor<string, []>("op_59109_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_59110_cast_fp16 = softmax(axis = var_58034, x = aw_5035_cast_fp16)[name = tensor<string, []>("op_59110_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_59111_cast_fp16 = softmax(axis = var_58034, x = aw_5037_cast_fp16)[name = tensor<string, []>("op_59111_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_59112_cast_fp16 = softmax(axis = var_58034, x = aw_5039_cast_fp16)[name = tensor<string, []>("op_59112_cast_fp16")];
            tensor<string, []> var_59114_equation_0 = const()[name = tensor<string, []>("op_59114_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59114_cast_fp16 = einsum(equation = var_59114_equation_0, values = (var_59024_cast_fp16, var_59103_cast_fp16))[name = tensor<string, []>("op_59114_cast_fp16")];
            tensor<string, []> var_59116_equation_0 = const()[name = tensor<string, []>("op_59116_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59116_cast_fp16 = einsum(equation = var_59116_equation_0, values = (var_59028_cast_fp16, var_59104_cast_fp16))[name = tensor<string, []>("op_59116_cast_fp16")];
            tensor<string, []> var_59118_equation_0 = const()[name = tensor<string, []>("op_59118_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59118_cast_fp16 = einsum(equation = var_59118_equation_0, values = (var_59032_cast_fp16, var_59105_cast_fp16))[name = tensor<string, []>("op_59118_cast_fp16")];
            tensor<string, []> var_59120_equation_0 = const()[name = tensor<string, []>("op_59120_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59120_cast_fp16 = einsum(equation = var_59120_equation_0, values = (var_59036_cast_fp16, var_59106_cast_fp16))[name = tensor<string, []>("op_59120_cast_fp16")];
            tensor<string, []> var_59122_equation_0 = const()[name = tensor<string, []>("op_59122_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59122_cast_fp16 = einsum(equation = var_59122_equation_0, values = (var_59040_cast_fp16, var_59107_cast_fp16))[name = tensor<string, []>("op_59122_cast_fp16")];
            tensor<string, []> var_59124_equation_0 = const()[name = tensor<string, []>("op_59124_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59124_cast_fp16 = einsum(equation = var_59124_equation_0, values = (var_59044_cast_fp16, var_59108_cast_fp16))[name = tensor<string, []>("op_59124_cast_fp16")];
            tensor<string, []> var_59126_equation_0 = const()[name = tensor<string, []>("op_59126_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59126_cast_fp16 = einsum(equation = var_59126_equation_0, values = (var_59048_cast_fp16, var_59109_cast_fp16))[name = tensor<string, []>("op_59126_cast_fp16")];
            tensor<string, []> var_59128_equation_0 = const()[name = tensor<string, []>("op_59128_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59128_cast_fp16 = einsum(equation = var_59128_equation_0, values = (var_59052_cast_fp16, var_59110_cast_fp16))[name = tensor<string, []>("op_59128_cast_fp16")];
            tensor<string, []> var_59130_equation_0 = const()[name = tensor<string, []>("op_59130_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59130_cast_fp16 = einsum(equation = var_59130_equation_0, values = (var_59056_cast_fp16, var_59111_cast_fp16))[name = tensor<string, []>("op_59130_cast_fp16")];
            tensor<string, []> var_59132_equation_0 = const()[name = tensor<string, []>("op_59132_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59132_cast_fp16 = einsum(equation = var_59132_equation_0, values = (var_59060_cast_fp16, var_59112_cast_fp16))[name = tensor<string, []>("op_59132_cast_fp16")];
            tensor<bool, []> input_747_interleave_0 = const()[name = tensor<string, []>("input_747_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 640, 1, 4096]> input_747_cast_fp16 = concat(axis = var_58034, interleave = input_747_interleave_0, values = (var_59114_cast_fp16, var_59116_cast_fp16, var_59118_cast_fp16, var_59120_cast_fp16, var_59122_cast_fp16, var_59124_cast_fp16, var_59126_cast_fp16, var_59128_cast_fp16, var_59130_cast_fp16, var_59132_cast_fp16))[name = tensor<string, []>("input_747_cast_fp16")];
            tensor<string, []> var_59142_pad_type_0 = const()[name = tensor<string, []>("op_59142_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_59142_strides_0 = const()[name = tensor<string, []>("op_59142_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_59142_pad_0 = const()[name = tensor<string, []>("op_59142_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_59142_dilations_0 = const()[name = tensor<string, []>("op_59142_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_59142_groups_0 = const()[name = tensor<string, []>("op_59142_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_0_transformer_blocks_1_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(895464896))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(895772160))), name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_1_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            tensor<fp16, [640]> up_blocks_1_attentions_0_transformer_blocks_1_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_1_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(895772352)))];
            tensor<fp16, [2, 640, 1, 4096]> var_59142_cast_fp16 = conv(bias = up_blocks_1_attentions_0_transformer_blocks_1_attn2_to_out_0_bias_to_fp16, dilations = var_59142_dilations_0, groups = var_59142_groups_0, pad = var_59142_pad_0, pad_type = var_59142_pad_type_0, strides = var_59142_strides_0, weight = up_blocks_1_attentions_0_transformer_blocks_1_attn2_to_out_0_weight_to_fp16_palettized, x = input_747_cast_fp16)[name = tensor<string, []>("op_59142_cast_fp16")];
            tensor<fp16, [2, 640, 1, 4096]> inputs_395_cast_fp16 = add(x = var_59142_cast_fp16, y = inputs_393_cast_fp16)[name = tensor<string, []>("inputs_395_cast_fp16")];
            tensor<int32, [1]> input_749_axes_0 = const()[name = tensor<string, []>("input_749_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [640]> input_749_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_749_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(895773696)))];
            tensor<fp16, [640]> input_749_beta_0_to_fp16 = const()[name = tensor<string, []>("input_749_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(895775040)))];
            tensor<fp16, []> var_59152_to_fp16 = const()[name = tensor<string, []>("op_59152_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 640, 1, 4096]> input_749_cast_fp16 = layer_norm(axes = input_749_axes_0, beta = input_749_beta_0_to_fp16, epsilon = var_59152_to_fp16, gamma = input_749_gamma_0_to_fp16, x = inputs_395_cast_fp16)[name = tensor<string, []>("input_749_cast_fp16")];
            tensor<string, []> var_59172_pad_type_0 = const()[name = tensor<string, []>("op_59172_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_59172_strides_0 = const()[name = tensor<string, []>("op_59172_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_59172_pad_0 = const()[name = tensor<string, []>("op_59172_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_59172_dilations_0 = const()[name = tensor<string, []>("op_59172_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_59172_groups_0 = const()[name = tensor<string, []>("op_59172_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [5120, 640, 1, 1]> up_blocks_1_attentions_0_transformer_blocks_1_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2457600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(895776384))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(898234048))), name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_1_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([5120, 640, 1, 1])];
            tensor<fp16, [5120]> up_blocks_1_attentions_0_transformer_blocks_1_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_1_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(898234240)))];
            tensor<fp16, [2, 5120, 1, 4096]> var_59172_cast_fp16 = conv(bias = up_blocks_1_attentions_0_transformer_blocks_1_ff_net_0_proj_bias_to_fp16, dilations = var_59172_dilations_0, groups = var_59172_groups_0, pad = var_59172_pad_0, pad_type = var_59172_pad_type_0, strides = var_59172_strides_0, weight = up_blocks_1_attentions_0_transformer_blocks_1_ff_net_0_proj_weight_to_fp16_palettized, x = input_749_cast_fp16)[name = tensor<string, []>("op_59172_cast_fp16")];
            tensor<int32, [2]> var_59173_split_sizes_0 = const()[name = tensor<string, []>("op_59173_split_sizes_0"), val = tensor<int32, [2]>([2560, 2560])];
            tensor<int32, []> var_59173_axis_0 = const()[name = tensor<string, []>("op_59173_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 2560, 1, 4096]> var_59173_cast_fp16_0, tensor<fp16, [2, 2560, 1, 4096]> var_59173_cast_fp16_1 = split(axis = var_59173_axis_0, split_sizes = var_59173_split_sizes_0, x = var_59172_cast_fp16)[name = tensor<string, []>("op_59173_cast_fp16")];
            tensor<string, []> var_59175_mode_0 = const()[name = tensor<string, []>("op_59175_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 2560, 1, 4096]> var_59175_cast_fp16 = gelu(mode = var_59175_mode_0, x = var_59173_cast_fp16_1)[name = tensor<string, []>("op_59175_cast_fp16")];
            tensor<fp16, [2, 2560, 1, 4096]> input_751_cast_fp16 = mul(x = var_59173_cast_fp16_0, y = var_59175_cast_fp16)[name = tensor<string, []>("input_751_cast_fp16")];
            tensor<string, []> var_59183_pad_type_0 = const()[name = tensor<string, []>("op_59183_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_59183_strides_0 = const()[name = tensor<string, []>("op_59183_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_59183_pad_0 = const()[name = tensor<string, []>("op_59183_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_59183_dilations_0 = const()[name = tensor<string, []>("op_59183_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_59183_groups_0 = const()[name = tensor<string, []>("op_59183_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 2560, 1, 1]> up_blocks_1_attentions_0_transformer_blocks_1_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(898244544))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(899473408))), name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_1_ff_net_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 2560, 1, 1])];
            tensor<fp16, [640]> up_blocks_1_attentions_0_transformer_blocks_1_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_1_ff_net_2_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(899473600)))];
            tensor<fp16, [2, 640, 1, 4096]> var_59183_cast_fp16 = conv(bias = up_blocks_1_attentions_0_transformer_blocks_1_ff_net_2_bias_to_fp16, dilations = var_59183_dilations_0, groups = var_59183_groups_0, pad = var_59183_pad_0, pad_type = var_59183_pad_type_0, strides = var_59183_strides_0, weight = up_blocks_1_attentions_0_transformer_blocks_1_ff_net_2_weight_to_fp16_palettized, x = input_751_cast_fp16)[name = tensor<string, []>("op_59183_cast_fp16")];
            tensor<fp16, [2, 640, 1, 4096]> hidden_states_521_cast_fp16 = add(x = var_59183_cast_fp16, y = inputs_395_cast_fp16)[name = tensor<string, []>("hidden_states_521_cast_fp16")];
            tensor<int32, [4]> var_59185 = const()[name = tensor<string, []>("op_59185"), val = tensor<int32, [4]>([2, 640, 64, 64])];
            tensor<fp16, [2, 640, 64, 64]> input_753_cast_fp16 = reshape(shape = var_59185, x = hidden_states_521_cast_fp16)[name = tensor<string, []>("input_753_cast_fp16")];
            tensor<string, []> hidden_states_523_pad_type_0 = const()[name = tensor<string, []>("hidden_states_523_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> hidden_states_523_strides_0 = const()[name = tensor<string, []>("hidden_states_523_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_523_pad_0 = const()[name = tensor<string, []>("hidden_states_523_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_523_dilations_0 = const()[name = tensor<string, []>("hidden_states_523_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_523_groups_0 = const()[name = tensor<string, []>("hidden_states_523_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_0_proj_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(899474944))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(899782208))), name = tensor<string, []>("up_blocks_1_attentions_0_proj_out_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            tensor<fp16, [640]> 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, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(899782400)))];
            tensor<fp16, [2, 640, 64, 64]> hidden_states_523_cast_fp16 = conv(bias = up_blocks_1_attentions_0_proj_out_bias_to_fp16, dilations = hidden_states_523_dilations_0, groups = hidden_states_523_groups_0, pad = hidden_states_523_pad_0, pad_type = hidden_states_523_pad_type_0, strides = hidden_states_523_strides_0, weight = up_blocks_1_attentions_0_proj_out_weight_to_fp16_palettized, x = input_753_cast_fp16)[name = tensor<string, []>("hidden_states_523_cast_fp16")];
            tensor<fp16, [2, 640, 64, 64]> hidden_states_525_cast_fp16 = add(x = hidden_states_523_cast_fp16, y = hidden_states_505_cast_fp16)[name = tensor<string, []>("hidden_states_525_cast_fp16")];
            tensor<bool, []> input_755_interleave_0 = const()[name = tensor<string, []>("input_755_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 640, 64, 64]> cast_8 = cast(dtype = cast_8_dtype_0, x = input_79_cast_fp16)[name = tensor<string, []>("cast_6")];
            tensor<fp16, [2, 1280, 64, 64]> input_755_cast_fp16 = concat(axis = var_58034, interleave = input_755_interleave_0, values = (hidden_states_525_cast_fp16, cast_8))[name = tensor<string, []>("input_755_cast_fp16")];
            tensor<int32, [5]> reshape_132_shape_0 = const()[name = tensor<string, []>("reshape_132_shape_0"), val = tensor<int32, [5]>([2, 32, 40, 64, 64])];
            tensor<fp16, [2, 32, 40, 64, 64]> reshape_132_cast_fp16 = reshape(shape = reshape_132_shape_0, x = input_755_cast_fp16)[name = tensor<string, []>("reshape_132_cast_fp16")];
            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_fp16 = reduce_mean(axes = reduce_mean_99_axes_0, keep_dims = reduce_mean_99_keep_dims_0, x = reshape_132_cast_fp16)[name = tensor<string, []>("reduce_mean_99_cast_fp16")];
            tensor<fp16, [2, 32, 40, 64, 64]> sub_66_cast_fp16 = sub(x = reshape_132_cast_fp16, y = reduce_mean_99_cast_fp16)[name = tensor<string, []>("sub_66_cast_fp16")];
            tensor<fp16, [2, 32, 40, 64, 64]> square_33_cast_fp16 = square(x = sub_66_cast_fp16)[name = tensor<string, []>("square_33_cast_fp16")];
            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_fp16 = reduce_mean(axes = reduce_mean_101_axes_0, keep_dims = reduce_mean_101_keep_dims_0, x = square_33_cast_fp16)[name = tensor<string, []>("reduce_mean_101_cast_fp16")];
            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_fp16 = add(x = reduce_mean_101_cast_fp16, y = add_66_y_0_to_fp16)[name = tensor<string, []>("add_66_cast_fp16")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_33_cast_fp16 = sqrt(x = add_66_cast_fp16)[name = tensor<string, []>("sqrt_33_cast_fp16")];
            tensor<fp16, [2, 32, 40, 64, 64]> real_div_33_cast_fp16 = real_div(x = sub_66_cast_fp16, y = sqrt_33_cast_fp16)[name = tensor<string, []>("real_div_33_cast_fp16")];
            tensor<int32, [4]> reshape_133_shape_0 = const()[name = tensor<string, []>("reshape_133_shape_0"), val = tensor<int32, [4]>([2, 1280, 64, 64])];
            tensor<fp16, [2, 1280, 64, 64]> reshape_133_cast_fp16 = reshape(shape = reshape_133_shape_0, x = real_div_33_cast_fp16)[name = tensor<string, []>("reshape_133_cast_fp16")];
            tensor<fp16, [1280]> add_67_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_67_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(899783744)))];
            tensor<fp16, [1280]> add_67_beta_0_to_fp16 = const()[name = tensor<string, []>("add_67_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(899786368)))];
            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, 1280, 64, 64]> add_67_cast_fp16 = batch_norm(beta = add_67_beta_0_to_fp16, epsilon = add_67_epsilon_0_to_fp16, gamma = add_67_gamma_0_to_fp16, mean = add_23_mean_0_to_fp16, variance = add_23_variance_0_to_fp16, x = reshape_133_cast_fp16)[name = tensor<string, []>("add_67_cast_fp16")];
            tensor<fp16, [2, 1280, 64, 64]> input_759_cast_fp16 = silu(x = add_67_cast_fp16)[name = tensor<string, []>("input_759_cast_fp16")];
            tensor<string, []> hidden_states_527_pad_type_0 = const()[name = tensor<string, []>("hidden_states_527_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_527_pad_0 = const()[name = tensor<string, []>("hidden_states_527_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> hidden_states_527_strides_0 = const()[name = tensor<string, []>("hidden_states_527_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> hidden_states_527_dilations_0 = const()[name = tensor<string, []>("hidden_states_527_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_527_groups_0 = const()[name = tensor<string, []>("hidden_states_527_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 1280, 3, 3]> up_blocks_1_resnets_1_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5529600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(899788992))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(905318656))), name = tensor<string, []>("up_blocks_1_resnets_1_conv1_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 1280, 3, 3])];
            tensor<fp16, [640]> 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, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(905318848)))];
            tensor<fp16, [2, 640, 64, 64]> hidden_states_527_cast_fp16 = conv(bias = up_blocks_1_resnets_1_conv1_bias_to_fp16, dilations = hidden_states_527_dilations_0, groups = hidden_states_527_groups_0, pad = hidden_states_527_pad_0, pad_type = hidden_states_527_pad_type_0, strides = hidden_states_527_strides_0, weight = up_blocks_1_resnets_1_conv1_weight_to_fp16_palettized, x = input_759_cast_fp16)[name = tensor<string, []>("hidden_states_527_cast_fp16")];
            tensor<string, []> temb_25_pad_type_0 = const()[name = tensor<string, []>("temb_25_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> temb_25_strides_0 = const()[name = tensor<string, []>("temb_25_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> temb_25_pad_0 = const()[name = tensor<string, []>("temb_25_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> temb_25_dilations_0 = const()[name = tensor<string, []>("temb_25_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> temb_25_groups_0 = const()[name = tensor<string, []>("temb_25_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 1280, 1, 1]> up_blocks_1_resnets_1_time_emb_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [614400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(905320192))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(905934656))), name = tensor<string, []>("up_blocks_1_resnets_1_time_emb_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 1280, 1, 1])];
            tensor<fp16, [640]> 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, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(905934848)))];
            tensor<fp16, [2, 640, 1, 1]> temb_25_cast_fp16 = conv(bias = up_blocks_1_resnets_1_time_emb_proj_bias_to_fp16, dilations = temb_25_dilations_0, groups = temb_25_groups_0, pad = temb_25_pad_0, pad_type = temb_25_pad_type_0, strides = temb_25_strides_0, weight = up_blocks_1_resnets_1_time_emb_proj_weight_to_fp16_palettized, x = cast_2)[name = tensor<string, []>("temb_25_cast_fp16")];
            tensor<fp16, [2, 640, 64, 64]> input_763_cast_fp16 = add(x = hidden_states_527_cast_fp16, y = temb_25_cast_fp16)[name = tensor<string, []>("input_763_cast_fp16")];
            tensor<int32, [5]> reshape_136_shape_0 = const()[name = tensor<string, []>("reshape_136_shape_0"), val = tensor<int32, [5]>([2, 32, 20, 64, 64])];
            tensor<fp16, [2, 32, 20, 64, 64]> reshape_136_cast_fp16 = reshape(shape = reshape_136_shape_0, x = input_763_cast_fp16)[name = tensor<string, []>("reshape_136_cast_fp16")];
            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_fp16 = reduce_mean(axes = reduce_mean_102_axes_0, keep_dims = reduce_mean_102_keep_dims_0, x = reshape_136_cast_fp16)[name = tensor<string, []>("reduce_mean_102_cast_fp16")];
            tensor<fp16, [2, 32, 20, 64, 64]> sub_68_cast_fp16 = sub(x = reshape_136_cast_fp16, y = reduce_mean_102_cast_fp16)[name = tensor<string, []>("sub_68_cast_fp16")];
            tensor<fp16, [2, 32, 20, 64, 64]> square_34_cast_fp16 = square(x = sub_68_cast_fp16)[name = tensor<string, []>("square_34_cast_fp16")];
            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_fp16 = reduce_mean(axes = reduce_mean_104_axes_0, keep_dims = reduce_mean_104_keep_dims_0, x = square_34_cast_fp16)[name = tensor<string, []>("reduce_mean_104_cast_fp16")];
            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_fp16 = add(x = reduce_mean_104_cast_fp16, y = add_68_y_0_to_fp16)[name = tensor<string, []>("add_68_cast_fp16")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_34_cast_fp16 = sqrt(x = add_68_cast_fp16)[name = tensor<string, []>("sqrt_34_cast_fp16")];
            tensor<fp16, [2, 32, 20, 64, 64]> real_div_34_cast_fp16 = real_div(x = sub_68_cast_fp16, y = sqrt_34_cast_fp16)[name = tensor<string, []>("real_div_34_cast_fp16")];
            tensor<int32, [4]> reshape_137_shape_0 = const()[name = tensor<string, []>("reshape_137_shape_0"), val = tensor<int32, [4]>([2, 640, 64, 64])];
            tensor<fp16, [2, 640, 64, 64]> reshape_137_cast_fp16 = reshape(shape = reshape_137_shape_0, x = real_div_34_cast_fp16)[name = tensor<string, []>("reshape_137_cast_fp16")];
            tensor<fp16, [640]> add_69_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_69_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(905936192)))];
            tensor<fp16, [640]> add_69_beta_0_to_fp16 = const()[name = tensor<string, []>("add_69_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(905937536)))];
            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, 640, 64, 64]> add_69_cast_fp16 = 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_11_mean_0_to_fp16, variance = add_11_variance_0_to_fp16, x = reshape_137_cast_fp16)[name = tensor<string, []>("add_69_cast_fp16")];
            tensor<fp16, [2, 640, 64, 64]> input_767_cast_fp16 = silu(x = add_69_cast_fp16)[name = tensor<string, []>("input_767_cast_fp16")];
            tensor<string, []> hidden_states_529_pad_type_0 = const()[name = tensor<string, []>("hidden_states_529_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_529_pad_0 = const()[name = tensor<string, []>("hidden_states_529_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> hidden_states_529_strides_0 = const()[name = tensor<string, []>("hidden_states_529_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> hidden_states_529_dilations_0 = const()[name = tensor<string, []>("hidden_states_529_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_529_groups_0 = const()[name = tensor<string, []>("hidden_states_529_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 640, 3, 3]> up_blocks_1_resnets_1_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2764800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(905938880))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(908703744))), name = tensor<string, []>("up_blocks_1_resnets_1_conv2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 3, 3])];
            tensor<fp16, [640]> 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, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(908703936)))];
            tensor<fp16, [2, 640, 64, 64]> hidden_states_529_cast_fp16 = conv(bias = up_blocks_1_resnets_1_conv2_bias_to_fp16, dilations = hidden_states_529_dilations_0, groups = hidden_states_529_groups_0, pad = hidden_states_529_pad_0, pad_type = hidden_states_529_pad_type_0, strides = hidden_states_529_strides_0, weight = up_blocks_1_resnets_1_conv2_weight_to_fp16_palettized, x = input_767_cast_fp16)[name = tensor<string, []>("hidden_states_529_cast_fp16")];
            tensor<string, []> x_13_pad_type_0 = const()[name = tensor<string, []>("x_13_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> x_13_strides_0 = const()[name = tensor<string, []>("x_13_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> x_13_pad_0 = const()[name = tensor<string, []>("x_13_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> x_13_dilations_0 = const()[name = tensor<string, []>("x_13_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> x_13_groups_0 = const()[name = tensor<string, []>("x_13_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 1280, 1, 1]> up_blocks_1_resnets_1_conv_shortcut_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [614400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(908705280))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(909319744))), name = tensor<string, []>("up_blocks_1_resnets_1_conv_shortcut_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 1280, 1, 1])];
            tensor<fp16, [640]> 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, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(909319936)))];
            tensor<fp16, [2, 640, 64, 64]> x_13_cast_fp16 = conv(bias = up_blocks_1_resnets_1_conv_shortcut_bias_to_fp16, dilations = x_13_dilations_0, groups = x_13_groups_0, pad = x_13_pad_0, pad_type = x_13_pad_type_0, strides = x_13_strides_0, weight = up_blocks_1_resnets_1_conv_shortcut_weight_to_fp16_palettized, x = input_755_cast_fp16)[name = tensor<string, []>("x_13_cast_fp16")];
            tensor<fp16, [2, 640, 64, 64]> hidden_states_531_cast_fp16 = add(x = x_13_cast_fp16, y = hidden_states_529_cast_fp16)[name = tensor<string, []>("hidden_states_531_cast_fp16")];
            tensor<int32, [5]> reshape_140_shape_0 = const()[name = tensor<string, []>("reshape_140_shape_0"), val = tensor<int32, [5]>([2, 32, 20, 64, 64])];
            tensor<fp16, [2, 32, 20, 64, 64]> reshape_140_cast_fp16 = reshape(shape = reshape_140_shape_0, x = hidden_states_531_cast_fp16)[name = tensor<string, []>("reshape_140_cast_fp16")];
            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_fp16 = reduce_mean(axes = reduce_mean_105_axes_0, keep_dims = reduce_mean_105_keep_dims_0, x = reshape_140_cast_fp16)[name = tensor<string, []>("reduce_mean_105_cast_fp16")];
            tensor<fp16, [2, 32, 20, 64, 64]> sub_70_cast_fp16 = sub(x = reshape_140_cast_fp16, y = reduce_mean_105_cast_fp16)[name = tensor<string, []>("sub_70_cast_fp16")];
            tensor<fp16, [2, 32, 20, 64, 64]> square_35_cast_fp16 = square(x = sub_70_cast_fp16)[name = tensor<string, []>("square_35_cast_fp16")];
            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_fp16 = reduce_mean(axes = reduce_mean_107_axes_0, keep_dims = reduce_mean_107_keep_dims_0, x = square_35_cast_fp16)[name = tensor<string, []>("reduce_mean_107_cast_fp16")];
            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_fp16 = add(x = reduce_mean_107_cast_fp16, y = add_70_y_0_to_fp16)[name = tensor<string, []>("add_70_cast_fp16")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_35_cast_fp16 = sqrt(x = add_70_cast_fp16)[name = tensor<string, []>("sqrt_35_cast_fp16")];
            tensor<fp16, [2, 32, 20, 64, 64]> real_div_35_cast_fp16 = real_div(x = sub_70_cast_fp16, y = sqrt_35_cast_fp16)[name = tensor<string, []>("real_div_35_cast_fp16")];
            tensor<int32, [4]> reshape_141_shape_0 = const()[name = tensor<string, []>("reshape_141_shape_0"), val = tensor<int32, [4]>([2, 640, 64, 64])];
            tensor<fp16, [2, 640, 64, 64]> reshape_141_cast_fp16 = reshape(shape = reshape_141_shape_0, x = real_div_35_cast_fp16)[name = tensor<string, []>("reshape_141_cast_fp16")];
            tensor<fp16, [640]> add_71_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_71_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(909321280)))];
            tensor<fp16, [640]> add_71_beta_0_to_fp16 = const()[name = tensor<string, []>("add_71_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(909322624)))];
            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, 640, 64, 64]> add_71_cast_fp16 = 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_11_mean_0_to_fp16, variance = add_11_variance_0_to_fp16, x = reshape_141_cast_fp16)[name = tensor<string, []>("add_71_cast_fp16")];
            tensor<string, []> hidden_states_533_pad_type_0 = const()[name = tensor<string, []>("hidden_states_533_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> hidden_states_533_strides_0 = const()[name = tensor<string, []>("hidden_states_533_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_533_pad_0 = const()[name = tensor<string, []>("hidden_states_533_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_533_dilations_0 = const()[name = tensor<string, []>("hidden_states_533_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_533_groups_0 = const()[name = tensor<string, []>("hidden_states_533_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_1_proj_in_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(909323968))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(909631232))), name = tensor<string, []>("up_blocks_1_attentions_1_proj_in_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            tensor<fp16, [640]> 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, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(909631424)))];
            tensor<fp16, [2, 640, 64, 64]> hidden_states_533_cast_fp16 = conv(bias = up_blocks_1_attentions_1_proj_in_bias_to_fp16, dilations = hidden_states_533_dilations_0, groups = hidden_states_533_groups_0, pad = hidden_states_533_pad_0, pad_type = hidden_states_533_pad_type_0, strides = hidden_states_533_strides_0, weight = up_blocks_1_attentions_1_proj_in_weight_to_fp16_palettized, x = add_71_cast_fp16)[name = tensor<string, []>("hidden_states_533_cast_fp16")];
            tensor<int32, [4]> var_59267 = const()[name = tensor<string, []>("op_59267"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<fp16, [2, 640, 1, 4096]> inputs_397_cast_fp16 = reshape(shape = var_59267, x = hidden_states_533_cast_fp16)[name = tensor<string, []>("inputs_397_cast_fp16")];
            tensor<int32, [1]> hidden_states_535_axes_0 = const()[name = tensor<string, []>("hidden_states_535_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [640]> hidden_states_535_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_535_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(909632768)))];
            tensor<fp16, [640]> hidden_states_535_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_535_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(909634112)))];
            tensor<fp16, []> var_59283_to_fp16 = const()[name = tensor<string, []>("op_59283_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 640, 1, 4096]> hidden_states_535_cast_fp16 = layer_norm(axes = hidden_states_535_axes_0, beta = hidden_states_535_beta_0_to_fp16, epsilon = var_59283_to_fp16, gamma = hidden_states_535_gamma_0_to_fp16, x = inputs_397_cast_fp16)[name = tensor<string, []>("hidden_states_535_cast_fp16")];
            tensor<string, []> q_265_pad_type_0 = const()[name = tensor<string, []>("q_265_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_265_strides_0 = const()[name = tensor<string, []>("q_265_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_265_pad_0 = const()[name = tensor<string, []>("q_265_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_265_dilations_0 = const()[name = tensor<string, []>("q_265_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_265_groups_0 = const()[name = tensor<string, []>("q_265_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(909635456))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(909942720))), name = tensor<string, []>("up_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, 4096]> q_265_cast_fp16 = conv(dilations = q_265_dilations_0, groups = q_265_groups_0, pad = q_265_pad_0, pad_type = q_265_pad_type_0, strides = q_265_strides_0, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_535_cast_fp16)[name = tensor<string, []>("q_265_cast_fp16")];
            tensor<string, []> k_529_pad_type_0 = const()[name = tensor<string, []>("k_529_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_529_strides_0 = const()[name = tensor<string, []>("k_529_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_529_pad_0 = const()[name = tensor<string, []>("k_529_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_529_dilations_0 = const()[name = tensor<string, []>("k_529_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_529_groups_0 = const()[name = tensor<string, []>("k_529_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(909942912))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(910250176))), name = tensor<string, []>("up_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, 4096]> k_529_cast_fp16 = conv(dilations = k_529_dilations_0, groups = k_529_groups_0, pad = k_529_pad_0, pad_type = k_529_pad_type_0, strides = k_529_strides_0, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_535_cast_fp16)[name = tensor<string, []>("k_529_cast_fp16")];
            tensor<string, []> v_265_pad_type_0 = const()[name = tensor<string, []>("v_265_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_265_strides_0 = const()[name = tensor<string, []>("v_265_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_265_pad_0 = const()[name = tensor<string, []>("v_265_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_265_dilations_0 = const()[name = tensor<string, []>("v_265_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_265_groups_0 = const()[name = tensor<string, []>("v_265_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(910250368))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(910557632))), name = tensor<string, []>("up_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, 4096]> v_265_cast_fp16 = conv(dilations = v_265_dilations_0, groups = v_265_groups_0, pad = v_265_pad_0, pad_type = v_265_pad_type_0, strides = v_265_strides_0, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_535_cast_fp16)[name = tensor<string, []>("v_265_cast_fp16")];
            tensor<int32, [4]> var_59316_begin_0 = const()[name = tensor<string, []>("op_59316_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_59316_end_0 = const()[name = tensor<string, []>("op_59316_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_59316_end_mask_0 = const()[name = tensor<string, []>("op_59316_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59316_cast_fp16 = slice_by_index(begin = var_59316_begin_0, end = var_59316_end_0, end_mask = var_59316_end_mask_0, x = q_265_cast_fp16)[name = tensor<string, []>("op_59316_cast_fp16")];
            tensor<int32, [4]> var_59320_begin_0 = const()[name = tensor<string, []>("op_59320_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_59320_end_0 = const()[name = tensor<string, []>("op_59320_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_59320_end_mask_0 = const()[name = tensor<string, []>("op_59320_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59320_cast_fp16 = slice_by_index(begin = var_59320_begin_0, end = var_59320_end_0, end_mask = var_59320_end_mask_0, x = q_265_cast_fp16)[name = tensor<string, []>("op_59320_cast_fp16")];
            tensor<int32, [4]> var_59324_begin_0 = const()[name = tensor<string, []>("op_59324_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_59324_end_0 = const()[name = tensor<string, []>("op_59324_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_59324_end_mask_0 = const()[name = tensor<string, []>("op_59324_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59324_cast_fp16 = slice_by_index(begin = var_59324_begin_0, end = var_59324_end_0, end_mask = var_59324_end_mask_0, x = q_265_cast_fp16)[name = tensor<string, []>("op_59324_cast_fp16")];
            tensor<int32, [4]> var_59328_begin_0 = const()[name = tensor<string, []>("op_59328_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_59328_end_0 = const()[name = tensor<string, []>("op_59328_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_59328_end_mask_0 = const()[name = tensor<string, []>("op_59328_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59328_cast_fp16 = slice_by_index(begin = var_59328_begin_0, end = var_59328_end_0, end_mask = var_59328_end_mask_0, x = q_265_cast_fp16)[name = tensor<string, []>("op_59328_cast_fp16")];
            tensor<int32, [4]> var_59332_begin_0 = const()[name = tensor<string, []>("op_59332_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_59332_end_0 = const()[name = tensor<string, []>("op_59332_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_59332_end_mask_0 = const()[name = tensor<string, []>("op_59332_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59332_cast_fp16 = slice_by_index(begin = var_59332_begin_0, end = var_59332_end_0, end_mask = var_59332_end_mask_0, x = q_265_cast_fp16)[name = tensor<string, []>("op_59332_cast_fp16")];
            tensor<int32, [4]> var_59336_begin_0 = const()[name = tensor<string, []>("op_59336_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_59336_end_0 = const()[name = tensor<string, []>("op_59336_end_0"), val = tensor<int32, [4]>([2, 384, 1, 4096])];
            tensor<bool, [4]> var_59336_end_mask_0 = const()[name = tensor<string, []>("op_59336_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59336_cast_fp16 = slice_by_index(begin = var_59336_begin_0, end = var_59336_end_0, end_mask = var_59336_end_mask_0, x = q_265_cast_fp16)[name = tensor<string, []>("op_59336_cast_fp16")];
            tensor<int32, [4]> var_59340_begin_0 = const()[name = tensor<string, []>("op_59340_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_59340_end_0 = const()[name = tensor<string, []>("op_59340_end_0"), val = tensor<int32, [4]>([2, 448, 1, 4096])];
            tensor<bool, [4]> var_59340_end_mask_0 = const()[name = tensor<string, []>("op_59340_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59340_cast_fp16 = slice_by_index(begin = var_59340_begin_0, end = var_59340_end_0, end_mask = var_59340_end_mask_0, x = q_265_cast_fp16)[name = tensor<string, []>("op_59340_cast_fp16")];
            tensor<int32, [4]> var_59344_begin_0 = const()[name = tensor<string, []>("op_59344_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_59344_end_0 = const()[name = tensor<string, []>("op_59344_end_0"), val = tensor<int32, [4]>([2, 512, 1, 4096])];
            tensor<bool, [4]> var_59344_end_mask_0 = const()[name = tensor<string, []>("op_59344_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59344_cast_fp16 = slice_by_index(begin = var_59344_begin_0, end = var_59344_end_0, end_mask = var_59344_end_mask_0, x = q_265_cast_fp16)[name = tensor<string, []>("op_59344_cast_fp16")];
            tensor<int32, [4]> var_59348_begin_0 = const()[name = tensor<string, []>("op_59348_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_59348_end_0 = const()[name = tensor<string, []>("op_59348_end_0"), val = tensor<int32, [4]>([2, 576, 1, 4096])];
            tensor<bool, [4]> var_59348_end_mask_0 = const()[name = tensor<string, []>("op_59348_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59348_cast_fp16 = slice_by_index(begin = var_59348_begin_0, end = var_59348_end_0, end_mask = var_59348_end_mask_0, x = q_265_cast_fp16)[name = tensor<string, []>("op_59348_cast_fp16")];
            tensor<int32, [4]> var_59352_begin_0 = const()[name = tensor<string, []>("op_59352_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_59352_end_0 = const()[name = tensor<string, []>("op_59352_end_0"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<bool, [4]> var_59352_end_mask_0 = const()[name = tensor<string, []>("op_59352_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59352_cast_fp16 = slice_by_index(begin = var_59352_begin_0, end = var_59352_end_0, end_mask = var_59352_end_mask_0, x = q_265_cast_fp16)[name = tensor<string, []>("op_59352_cast_fp16")];
            tensor<int32, [4]> k_531_perm_0 = const()[name = tensor<string, []>("k_531_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_59359_begin_0 = const()[name = tensor<string, []>("op_59359_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_59359_end_0 = const()[name = tensor<string, []>("op_59359_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 64])];
            tensor<bool, [4]> var_59359_end_mask_0 = const()[name = tensor<string, []>("op_59359_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 640]> k_531_cast_fp16 = transpose(perm = k_531_perm_0, x = k_529_cast_fp16)[name = tensor<string, []>("transpose_7")];
            tensor<fp16, [2, 4096, 1, 64]> var_59359_cast_fp16 = slice_by_index(begin = var_59359_begin_0, end = var_59359_end_0, end_mask = var_59359_end_mask_0, x = k_531_cast_fp16)[name = tensor<string, []>("op_59359_cast_fp16")];
            tensor<int32, [4]> var_59363_begin_0 = const()[name = tensor<string, []>("op_59363_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_59363_end_0 = const()[name = tensor<string, []>("op_59363_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 128])];
            tensor<bool, [4]> var_59363_end_mask_0 = const()[name = tensor<string, []>("op_59363_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_59363_cast_fp16 = slice_by_index(begin = var_59363_begin_0, end = var_59363_end_0, end_mask = var_59363_end_mask_0, x = k_531_cast_fp16)[name = tensor<string, []>("op_59363_cast_fp16")];
            tensor<int32, [4]> var_59367_begin_0 = const()[name = tensor<string, []>("op_59367_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_59367_end_0 = const()[name = tensor<string, []>("op_59367_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 192])];
            tensor<bool, [4]> var_59367_end_mask_0 = const()[name = tensor<string, []>("op_59367_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_59367_cast_fp16 = slice_by_index(begin = var_59367_begin_0, end = var_59367_end_0, end_mask = var_59367_end_mask_0, x = k_531_cast_fp16)[name = tensor<string, []>("op_59367_cast_fp16")];
            tensor<int32, [4]> var_59371_begin_0 = const()[name = tensor<string, []>("op_59371_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_59371_end_0 = const()[name = tensor<string, []>("op_59371_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 256])];
            tensor<bool, [4]> var_59371_end_mask_0 = const()[name = tensor<string, []>("op_59371_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_59371_cast_fp16 = slice_by_index(begin = var_59371_begin_0, end = var_59371_end_0, end_mask = var_59371_end_mask_0, x = k_531_cast_fp16)[name = tensor<string, []>("op_59371_cast_fp16")];
            tensor<int32, [4]> var_59375_begin_0 = const()[name = tensor<string, []>("op_59375_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_59375_end_0 = const()[name = tensor<string, []>("op_59375_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 320])];
            tensor<bool, [4]> var_59375_end_mask_0 = const()[name = tensor<string, []>("op_59375_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_59375_cast_fp16 = slice_by_index(begin = var_59375_begin_0, end = var_59375_end_0, end_mask = var_59375_end_mask_0, x = k_531_cast_fp16)[name = tensor<string, []>("op_59375_cast_fp16")];
            tensor<int32, [4]> var_59379_begin_0 = const()[name = tensor<string, []>("op_59379_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_59379_end_0 = const()[name = tensor<string, []>("op_59379_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 384])];
            tensor<bool, [4]> var_59379_end_mask_0 = const()[name = tensor<string, []>("op_59379_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_59379_cast_fp16 = slice_by_index(begin = var_59379_begin_0, end = var_59379_end_0, end_mask = var_59379_end_mask_0, x = k_531_cast_fp16)[name = tensor<string, []>("op_59379_cast_fp16")];
            tensor<int32, [4]> var_59383_begin_0 = const()[name = tensor<string, []>("op_59383_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_59383_end_0 = const()[name = tensor<string, []>("op_59383_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 448])];
            tensor<bool, [4]> var_59383_end_mask_0 = const()[name = tensor<string, []>("op_59383_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_59383_cast_fp16 = slice_by_index(begin = var_59383_begin_0, end = var_59383_end_0, end_mask = var_59383_end_mask_0, x = k_531_cast_fp16)[name = tensor<string, []>("op_59383_cast_fp16")];
            tensor<int32, [4]> var_59387_begin_0 = const()[name = tensor<string, []>("op_59387_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_59387_end_0 = const()[name = tensor<string, []>("op_59387_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 512])];
            tensor<bool, [4]> var_59387_end_mask_0 = const()[name = tensor<string, []>("op_59387_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_59387_cast_fp16 = slice_by_index(begin = var_59387_begin_0, end = var_59387_end_0, end_mask = var_59387_end_mask_0, x = k_531_cast_fp16)[name = tensor<string, []>("op_59387_cast_fp16")];
            tensor<int32, [4]> var_59391_begin_0 = const()[name = tensor<string, []>("op_59391_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_59391_end_0 = const()[name = tensor<string, []>("op_59391_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 576])];
            tensor<bool, [4]> var_59391_end_mask_0 = const()[name = tensor<string, []>("op_59391_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_59391_cast_fp16 = slice_by_index(begin = var_59391_begin_0, end = var_59391_end_0, end_mask = var_59391_end_mask_0, x = k_531_cast_fp16)[name = tensor<string, []>("op_59391_cast_fp16")];
            tensor<int32, [4]> var_59395_begin_0 = const()[name = tensor<string, []>("op_59395_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_59395_end_0 = const()[name = tensor<string, []>("op_59395_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 640])];
            tensor<bool, [4]> var_59395_end_mask_0 = const()[name = tensor<string, []>("op_59395_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_59395_cast_fp16 = slice_by_index(begin = var_59395_begin_0, end = var_59395_end_0, end_mask = var_59395_end_mask_0, x = k_531_cast_fp16)[name = tensor<string, []>("op_59395_cast_fp16")];
            tensor<int32, [4]> var_59397_begin_0 = const()[name = tensor<string, []>("op_59397_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_59397_end_0 = const()[name = tensor<string, []>("op_59397_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_59397_end_mask_0 = const()[name = tensor<string, []>("op_59397_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59397_cast_fp16 = slice_by_index(begin = var_59397_begin_0, end = var_59397_end_0, end_mask = var_59397_end_mask_0, x = v_265_cast_fp16)[name = tensor<string, []>("op_59397_cast_fp16")];
            tensor<int32, [4]> var_59401_begin_0 = const()[name = tensor<string, []>("op_59401_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_59401_end_0 = const()[name = tensor<string, []>("op_59401_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_59401_end_mask_0 = const()[name = tensor<string, []>("op_59401_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59401_cast_fp16 = slice_by_index(begin = var_59401_begin_0, end = var_59401_end_0, end_mask = var_59401_end_mask_0, x = v_265_cast_fp16)[name = tensor<string, []>("op_59401_cast_fp16")];
            tensor<int32, [4]> var_59405_begin_0 = const()[name = tensor<string, []>("op_59405_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_59405_end_0 = const()[name = tensor<string, []>("op_59405_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_59405_end_mask_0 = const()[name = tensor<string, []>("op_59405_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59405_cast_fp16 = slice_by_index(begin = var_59405_begin_0, end = var_59405_end_0, end_mask = var_59405_end_mask_0, x = v_265_cast_fp16)[name = tensor<string, []>("op_59405_cast_fp16")];
            tensor<int32, [4]> var_59409_begin_0 = const()[name = tensor<string, []>("op_59409_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_59409_end_0 = const()[name = tensor<string, []>("op_59409_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_59409_end_mask_0 = const()[name = tensor<string, []>("op_59409_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59409_cast_fp16 = slice_by_index(begin = var_59409_begin_0, end = var_59409_end_0, end_mask = var_59409_end_mask_0, x = v_265_cast_fp16)[name = tensor<string, []>("op_59409_cast_fp16")];
            tensor<int32, [4]> var_59413_begin_0 = const()[name = tensor<string, []>("op_59413_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_59413_end_0 = const()[name = tensor<string, []>("op_59413_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_59413_end_mask_0 = const()[name = tensor<string, []>("op_59413_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59413_cast_fp16 = slice_by_index(begin = var_59413_begin_0, end = var_59413_end_0, end_mask = var_59413_end_mask_0, x = v_265_cast_fp16)[name = tensor<string, []>("op_59413_cast_fp16")];
            tensor<int32, [4]> var_59417_begin_0 = const()[name = tensor<string, []>("op_59417_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_59417_end_0 = const()[name = tensor<string, []>("op_59417_end_0"), val = tensor<int32, [4]>([2, 384, 1, 4096])];
            tensor<bool, [4]> var_59417_end_mask_0 = const()[name = tensor<string, []>("op_59417_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59417_cast_fp16 = slice_by_index(begin = var_59417_begin_0, end = var_59417_end_0, end_mask = var_59417_end_mask_0, x = v_265_cast_fp16)[name = tensor<string, []>("op_59417_cast_fp16")];
            tensor<int32, [4]> var_59421_begin_0 = const()[name = tensor<string, []>("op_59421_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_59421_end_0 = const()[name = tensor<string, []>("op_59421_end_0"), val = tensor<int32, [4]>([2, 448, 1, 4096])];
            tensor<bool, [4]> var_59421_end_mask_0 = const()[name = tensor<string, []>("op_59421_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59421_cast_fp16 = slice_by_index(begin = var_59421_begin_0, end = var_59421_end_0, end_mask = var_59421_end_mask_0, x = v_265_cast_fp16)[name = tensor<string, []>("op_59421_cast_fp16")];
            tensor<int32, [4]> var_59425_begin_0 = const()[name = tensor<string, []>("op_59425_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_59425_end_0 = const()[name = tensor<string, []>("op_59425_end_0"), val = tensor<int32, [4]>([2, 512, 1, 4096])];
            tensor<bool, [4]> var_59425_end_mask_0 = const()[name = tensor<string, []>("op_59425_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59425_cast_fp16 = slice_by_index(begin = var_59425_begin_0, end = var_59425_end_0, end_mask = var_59425_end_mask_0, x = v_265_cast_fp16)[name = tensor<string, []>("op_59425_cast_fp16")];
            tensor<int32, [4]> var_59429_begin_0 = const()[name = tensor<string, []>("op_59429_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_59429_end_0 = const()[name = tensor<string, []>("op_59429_end_0"), val = tensor<int32, [4]>([2, 576, 1, 4096])];
            tensor<bool, [4]> var_59429_end_mask_0 = const()[name = tensor<string, []>("op_59429_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59429_cast_fp16 = slice_by_index(begin = var_59429_begin_0, end = var_59429_end_0, end_mask = var_59429_end_mask_0, x = v_265_cast_fp16)[name = tensor<string, []>("op_59429_cast_fp16")];
            tensor<int32, [4]> var_59433_begin_0 = const()[name = tensor<string, []>("op_59433_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_59433_end_0 = const()[name = tensor<string, []>("op_59433_end_0"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<bool, [4]> var_59433_end_mask_0 = const()[name = tensor<string, []>("op_59433_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59433_cast_fp16 = slice_by_index(begin = var_59433_begin_0, end = var_59433_end_0, end_mask = var_59433_end_mask_0, x = v_265_cast_fp16)[name = tensor<string, []>("op_59433_cast_fp16")];
            tensor<string, []> var_59437_equation_0 = const()[name = tensor<string, []>("op_59437_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_59437_cast_fp16 = einsum(equation = var_59437_equation_0, values = (var_59359_cast_fp16, var_59316_cast_fp16))[name = tensor<string, []>("op_59437_cast_fp16")];
            tensor<fp16, []> var_59438_to_fp16 = const()[name = tensor<string, []>("op_59438_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5041_cast_fp16 = mul(x = var_59437_cast_fp16, y = var_59438_to_fp16)[name = tensor<string, []>("aw_5041_cast_fp16")];
            tensor<string, []> var_59441_equation_0 = const()[name = tensor<string, []>("op_59441_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_59441_cast_fp16 = einsum(equation = var_59441_equation_0, values = (var_59363_cast_fp16, var_59320_cast_fp16))[name = tensor<string, []>("op_59441_cast_fp16")];
            tensor<fp16, []> var_59442_to_fp16 = const()[name = tensor<string, []>("op_59442_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5043_cast_fp16 = mul(x = var_59441_cast_fp16, y = var_59442_to_fp16)[name = tensor<string, []>("aw_5043_cast_fp16")];
            tensor<string, []> var_59445_equation_0 = const()[name = tensor<string, []>("op_59445_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_59445_cast_fp16 = einsum(equation = var_59445_equation_0, values = (var_59367_cast_fp16, var_59324_cast_fp16))[name = tensor<string, []>("op_59445_cast_fp16")];
            tensor<fp16, []> var_59446_to_fp16 = const()[name = tensor<string, []>("op_59446_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5045_cast_fp16 = mul(x = var_59445_cast_fp16, y = var_59446_to_fp16)[name = tensor<string, []>("aw_5045_cast_fp16")];
            tensor<string, []> var_59449_equation_0 = const()[name = tensor<string, []>("op_59449_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_59449_cast_fp16 = einsum(equation = var_59449_equation_0, values = (var_59371_cast_fp16, var_59328_cast_fp16))[name = tensor<string, []>("op_59449_cast_fp16")];
            tensor<fp16, []> var_59450_to_fp16 = const()[name = tensor<string, []>("op_59450_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5047_cast_fp16 = mul(x = var_59449_cast_fp16, y = var_59450_to_fp16)[name = tensor<string, []>("aw_5047_cast_fp16")];
            tensor<string, []> var_59453_equation_0 = const()[name = tensor<string, []>("op_59453_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_59453_cast_fp16 = einsum(equation = var_59453_equation_0, values = (var_59375_cast_fp16, var_59332_cast_fp16))[name = tensor<string, []>("op_59453_cast_fp16")];
            tensor<fp16, []> var_59454_to_fp16 = const()[name = tensor<string, []>("op_59454_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5049_cast_fp16 = mul(x = var_59453_cast_fp16, y = var_59454_to_fp16)[name = tensor<string, []>("aw_5049_cast_fp16")];
            tensor<string, []> var_59457_equation_0 = const()[name = tensor<string, []>("op_59457_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_59457_cast_fp16 = einsum(equation = var_59457_equation_0, values = (var_59379_cast_fp16, var_59336_cast_fp16))[name = tensor<string, []>("op_59457_cast_fp16")];
            tensor<fp16, []> var_59458_to_fp16 = const()[name = tensor<string, []>("op_59458_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5051_cast_fp16 = mul(x = var_59457_cast_fp16, y = var_59458_to_fp16)[name = tensor<string, []>("aw_5051_cast_fp16")];
            tensor<string, []> var_59461_equation_0 = const()[name = tensor<string, []>("op_59461_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_59461_cast_fp16 = einsum(equation = var_59461_equation_0, values = (var_59383_cast_fp16, var_59340_cast_fp16))[name = tensor<string, []>("op_59461_cast_fp16")];
            tensor<fp16, []> var_59462_to_fp16 = const()[name = tensor<string, []>("op_59462_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5053_cast_fp16 = mul(x = var_59461_cast_fp16, y = var_59462_to_fp16)[name = tensor<string, []>("aw_5053_cast_fp16")];
            tensor<string, []> var_59465_equation_0 = const()[name = tensor<string, []>("op_59465_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_59465_cast_fp16 = einsum(equation = var_59465_equation_0, values = (var_59387_cast_fp16, var_59344_cast_fp16))[name = tensor<string, []>("op_59465_cast_fp16")];
            tensor<fp16, []> var_59466_to_fp16 = const()[name = tensor<string, []>("op_59466_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5055_cast_fp16 = mul(x = var_59465_cast_fp16, y = var_59466_to_fp16)[name = tensor<string, []>("aw_5055_cast_fp16")];
            tensor<string, []> var_59469_equation_0 = const()[name = tensor<string, []>("op_59469_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_59469_cast_fp16 = einsum(equation = var_59469_equation_0, values = (var_59391_cast_fp16, var_59348_cast_fp16))[name = tensor<string, []>("op_59469_cast_fp16")];
            tensor<fp16, []> var_59470_to_fp16 = const()[name = tensor<string, []>("op_59470_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5057_cast_fp16 = mul(x = var_59469_cast_fp16, y = var_59470_to_fp16)[name = tensor<string, []>("aw_5057_cast_fp16")];
            tensor<string, []> var_59473_equation_0 = const()[name = tensor<string, []>("op_59473_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_59473_cast_fp16 = einsum(equation = var_59473_equation_0, values = (var_59395_cast_fp16, var_59352_cast_fp16))[name = tensor<string, []>("op_59473_cast_fp16")];
            tensor<fp16, []> var_59474_to_fp16 = const()[name = tensor<string, []>("op_59474_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5059_cast_fp16 = mul(x = var_59473_cast_fp16, y = var_59474_to_fp16)[name = tensor<string, []>("aw_5059_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_59476_cast_fp16 = softmax(axis = var_58034, x = aw_5041_cast_fp16)[name = tensor<string, []>("op_59476_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_59477_cast_fp16 = softmax(axis = var_58034, x = aw_5043_cast_fp16)[name = tensor<string, []>("op_59477_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_59478_cast_fp16 = softmax(axis = var_58034, x = aw_5045_cast_fp16)[name = tensor<string, []>("op_59478_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_59479_cast_fp16 = softmax(axis = var_58034, x = aw_5047_cast_fp16)[name = tensor<string, []>("op_59479_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_59480_cast_fp16 = softmax(axis = var_58034, x = aw_5049_cast_fp16)[name = tensor<string, []>("op_59480_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_59481_cast_fp16 = softmax(axis = var_58034, x = aw_5051_cast_fp16)[name = tensor<string, []>("op_59481_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_59482_cast_fp16 = softmax(axis = var_58034, x = aw_5053_cast_fp16)[name = tensor<string, []>("op_59482_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_59483_cast_fp16 = softmax(axis = var_58034, x = aw_5055_cast_fp16)[name = tensor<string, []>("op_59483_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_59484_cast_fp16 = softmax(axis = var_58034, x = aw_5057_cast_fp16)[name = tensor<string, []>("op_59484_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_59485_cast_fp16 = softmax(axis = var_58034, x = aw_5059_cast_fp16)[name = tensor<string, []>("op_59485_cast_fp16")];
            tensor<string, []> var_59487_equation_0 = const()[name = tensor<string, []>("op_59487_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59487_cast_fp16 = einsum(equation = var_59487_equation_0, values = (var_59397_cast_fp16, var_59476_cast_fp16))[name = tensor<string, []>("op_59487_cast_fp16")];
            tensor<string, []> var_59489_equation_0 = const()[name = tensor<string, []>("op_59489_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59489_cast_fp16 = einsum(equation = var_59489_equation_0, values = (var_59401_cast_fp16, var_59477_cast_fp16))[name = tensor<string, []>("op_59489_cast_fp16")];
            tensor<string, []> var_59491_equation_0 = const()[name = tensor<string, []>("op_59491_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59491_cast_fp16 = einsum(equation = var_59491_equation_0, values = (var_59405_cast_fp16, var_59478_cast_fp16))[name = tensor<string, []>("op_59491_cast_fp16")];
            tensor<string, []> var_59493_equation_0 = const()[name = tensor<string, []>("op_59493_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59493_cast_fp16 = einsum(equation = var_59493_equation_0, values = (var_59409_cast_fp16, var_59479_cast_fp16))[name = tensor<string, []>("op_59493_cast_fp16")];
            tensor<string, []> var_59495_equation_0 = const()[name = tensor<string, []>("op_59495_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59495_cast_fp16 = einsum(equation = var_59495_equation_0, values = (var_59413_cast_fp16, var_59480_cast_fp16))[name = tensor<string, []>("op_59495_cast_fp16")];
            tensor<string, []> var_59497_equation_0 = const()[name = tensor<string, []>("op_59497_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59497_cast_fp16 = einsum(equation = var_59497_equation_0, values = (var_59417_cast_fp16, var_59481_cast_fp16))[name = tensor<string, []>("op_59497_cast_fp16")];
            tensor<string, []> var_59499_equation_0 = const()[name = tensor<string, []>("op_59499_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59499_cast_fp16 = einsum(equation = var_59499_equation_0, values = (var_59421_cast_fp16, var_59482_cast_fp16))[name = tensor<string, []>("op_59499_cast_fp16")];
            tensor<string, []> var_59501_equation_0 = const()[name = tensor<string, []>("op_59501_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59501_cast_fp16 = einsum(equation = var_59501_equation_0, values = (var_59425_cast_fp16, var_59483_cast_fp16))[name = tensor<string, []>("op_59501_cast_fp16")];
            tensor<string, []> var_59503_equation_0 = const()[name = tensor<string, []>("op_59503_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59503_cast_fp16 = einsum(equation = var_59503_equation_0, values = (var_59429_cast_fp16, var_59484_cast_fp16))[name = tensor<string, []>("op_59503_cast_fp16")];
            tensor<string, []> var_59505_equation_0 = const()[name = tensor<string, []>("op_59505_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59505_cast_fp16 = einsum(equation = var_59505_equation_0, values = (var_59433_cast_fp16, var_59485_cast_fp16))[name = tensor<string, []>("op_59505_cast_fp16")];
            tensor<bool, []> input_771_interleave_0 = const()[name = tensor<string, []>("input_771_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 640, 1, 4096]> input_771_cast_fp16 = concat(axis = var_58034, interleave = input_771_interleave_0, values = (var_59487_cast_fp16, var_59489_cast_fp16, var_59491_cast_fp16, var_59493_cast_fp16, var_59495_cast_fp16, var_59497_cast_fp16, var_59499_cast_fp16, var_59501_cast_fp16, var_59503_cast_fp16, var_59505_cast_fp16))[name = tensor<string, []>("input_771_cast_fp16")];
            tensor<string, []> var_59515_pad_type_0 = const()[name = tensor<string, []>("op_59515_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_59515_strides_0 = const()[name = tensor<string, []>("op_59515_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_59515_pad_0 = const()[name = tensor<string, []>("op_59515_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_59515_dilations_0 = const()[name = tensor<string, []>("op_59515_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_59515_groups_0 = const()[name = tensor<string, []>("op_59515_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 640, 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, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(910557824))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(910865088))), name = tensor<string, []>("up_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]> 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, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(910865280)))];
            tensor<fp16, [2, 640, 1, 4096]> var_59515_cast_fp16 = conv(bias = up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_59515_dilations_0, groups = var_59515_groups_0, pad = var_59515_pad_0, pad_type = var_59515_pad_type_0, strides = var_59515_strides_0, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized, x = input_771_cast_fp16)[name = tensor<string, []>("op_59515_cast_fp16")];
            tensor<fp16, [2, 640, 1, 4096]> inputs_399_cast_fp16 = add(x = var_59515_cast_fp16, y = inputs_397_cast_fp16)[name = tensor<string, []>("inputs_399_cast_fp16")];
            tensor<int32, [1]> hidden_states_537_axes_0 = const()[name = tensor<string, []>("hidden_states_537_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [640]> hidden_states_537_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_537_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(910866624)))];
            tensor<fp16, [640]> hidden_states_537_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_537_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(910867968)))];
            tensor<fp16, []> var_59525_to_fp16 = const()[name = tensor<string, []>("op_59525_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 640, 1, 4096]> hidden_states_537_cast_fp16 = layer_norm(axes = hidden_states_537_axes_0, beta = hidden_states_537_beta_0_to_fp16, epsilon = var_59525_to_fp16, gamma = hidden_states_537_gamma_0_to_fp16, x = inputs_399_cast_fp16)[name = tensor<string, []>("hidden_states_537_cast_fp16")];
            tensor<string, []> q_267_pad_type_0 = const()[name = tensor<string, []>("q_267_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_267_strides_0 = const()[name = tensor<string, []>("q_267_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_267_pad_0 = const()[name = tensor<string, []>("q_267_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_267_dilations_0 = const()[name = tensor<string, []>("q_267_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_267_groups_0 = const()[name = tensor<string, []>("q_267_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(910869312))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(911176576))), name = tensor<string, []>("up_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, 4096]> q_267_cast_fp16 = conv(dilations = q_267_dilations_0, groups = q_267_groups_0, pad = q_267_pad_0, pad_type = q_267_pad_type_0, strides = q_267_strides_0, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_537_cast_fp16)[name = tensor<string, []>("q_267_cast_fp16")];
            tensor<string, []> k_533_pad_type_0 = const()[name = tensor<string, []>("k_533_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_533_strides_0 = const()[name = tensor<string, []>("k_533_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_533_pad_0 = const()[name = tensor<string, []>("k_533_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_533_dilations_0 = const()[name = tensor<string, []>("k_533_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_533_groups_0 = const()[name = tensor<string, []>("k_533_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 2048, 1, 1]> up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [983040]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(911176768))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(912159872))), name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 2048, 1, 1])];
            tensor<fp16, [2, 640, 1, 77]> k_533_cast_fp16 = conv(dilations = k_533_dilations_0, groups = k_533_groups_0, pad = k_533_pad_0, pad_type = k_533_pad_type_0, strides = k_533_strides_0, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("k_533_cast_fp16")];
            tensor<string, []> v_267_pad_type_0 = const()[name = tensor<string, []>("v_267_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_267_strides_0 = const()[name = tensor<string, []>("v_267_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_267_pad_0 = const()[name = tensor<string, []>("v_267_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_267_dilations_0 = const()[name = tensor<string, []>("v_267_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_267_groups_0 = const()[name = tensor<string, []>("v_267_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 2048, 1, 1]> up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [983040]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(912160064))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(913143168))), name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 2048, 1, 1])];
            tensor<fp16, [2, 640, 1, 77]> v_267_cast_fp16 = conv(dilations = v_267_dilations_0, groups = v_267_groups_0, pad = v_267_pad_0, pad_type = v_267_pad_type_0, strides = v_267_strides_0, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("v_267_cast_fp16")];
            tensor<int32, [4]> var_59558_begin_0 = const()[name = tensor<string, []>("op_59558_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_59558_end_0 = const()[name = tensor<string, []>("op_59558_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_59558_end_mask_0 = const()[name = tensor<string, []>("op_59558_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59558_cast_fp16 = slice_by_index(begin = var_59558_begin_0, end = var_59558_end_0, end_mask = var_59558_end_mask_0, x = q_267_cast_fp16)[name = tensor<string, []>("op_59558_cast_fp16")];
            tensor<int32, [4]> var_59562_begin_0 = const()[name = tensor<string, []>("op_59562_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_59562_end_0 = const()[name = tensor<string, []>("op_59562_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_59562_end_mask_0 = const()[name = tensor<string, []>("op_59562_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59562_cast_fp16 = slice_by_index(begin = var_59562_begin_0, end = var_59562_end_0, end_mask = var_59562_end_mask_0, x = q_267_cast_fp16)[name = tensor<string, []>("op_59562_cast_fp16")];
            tensor<int32, [4]> var_59566_begin_0 = const()[name = tensor<string, []>("op_59566_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_59566_end_0 = const()[name = tensor<string, []>("op_59566_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_59566_end_mask_0 = const()[name = tensor<string, []>("op_59566_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59566_cast_fp16 = slice_by_index(begin = var_59566_begin_0, end = var_59566_end_0, end_mask = var_59566_end_mask_0, x = q_267_cast_fp16)[name = tensor<string, []>("op_59566_cast_fp16")];
            tensor<int32, [4]> var_59570_begin_0 = const()[name = tensor<string, []>("op_59570_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_59570_end_0 = const()[name = tensor<string, []>("op_59570_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_59570_end_mask_0 = const()[name = tensor<string, []>("op_59570_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59570_cast_fp16 = slice_by_index(begin = var_59570_begin_0, end = var_59570_end_0, end_mask = var_59570_end_mask_0, x = q_267_cast_fp16)[name = tensor<string, []>("op_59570_cast_fp16")];
            tensor<int32, [4]> var_59574_begin_0 = const()[name = tensor<string, []>("op_59574_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_59574_end_0 = const()[name = tensor<string, []>("op_59574_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_59574_end_mask_0 = const()[name = tensor<string, []>("op_59574_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59574_cast_fp16 = slice_by_index(begin = var_59574_begin_0, end = var_59574_end_0, end_mask = var_59574_end_mask_0, x = q_267_cast_fp16)[name = tensor<string, []>("op_59574_cast_fp16")];
            tensor<int32, [4]> var_59578_begin_0 = const()[name = tensor<string, []>("op_59578_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_59578_end_0 = const()[name = tensor<string, []>("op_59578_end_0"), val = tensor<int32, [4]>([2, 384, 1, 4096])];
            tensor<bool, [4]> var_59578_end_mask_0 = const()[name = tensor<string, []>("op_59578_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59578_cast_fp16 = slice_by_index(begin = var_59578_begin_0, end = var_59578_end_0, end_mask = var_59578_end_mask_0, x = q_267_cast_fp16)[name = tensor<string, []>("op_59578_cast_fp16")];
            tensor<int32, [4]> var_59582_begin_0 = const()[name = tensor<string, []>("op_59582_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_59582_end_0 = const()[name = tensor<string, []>("op_59582_end_0"), val = tensor<int32, [4]>([2, 448, 1, 4096])];
            tensor<bool, [4]> var_59582_end_mask_0 = const()[name = tensor<string, []>("op_59582_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59582_cast_fp16 = slice_by_index(begin = var_59582_begin_0, end = var_59582_end_0, end_mask = var_59582_end_mask_0, x = q_267_cast_fp16)[name = tensor<string, []>("op_59582_cast_fp16")];
            tensor<int32, [4]> var_59586_begin_0 = const()[name = tensor<string, []>("op_59586_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_59586_end_0 = const()[name = tensor<string, []>("op_59586_end_0"), val = tensor<int32, [4]>([2, 512, 1, 4096])];
            tensor<bool, [4]> var_59586_end_mask_0 = const()[name = tensor<string, []>("op_59586_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59586_cast_fp16 = slice_by_index(begin = var_59586_begin_0, end = var_59586_end_0, end_mask = var_59586_end_mask_0, x = q_267_cast_fp16)[name = tensor<string, []>("op_59586_cast_fp16")];
            tensor<int32, [4]> var_59590_begin_0 = const()[name = tensor<string, []>("op_59590_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_59590_end_0 = const()[name = tensor<string, []>("op_59590_end_0"), val = tensor<int32, [4]>([2, 576, 1, 4096])];
            tensor<bool, [4]> var_59590_end_mask_0 = const()[name = tensor<string, []>("op_59590_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59590_cast_fp16 = slice_by_index(begin = var_59590_begin_0, end = var_59590_end_0, end_mask = var_59590_end_mask_0, x = q_267_cast_fp16)[name = tensor<string, []>("op_59590_cast_fp16")];
            tensor<int32, [4]> var_59594_begin_0 = const()[name = tensor<string, []>("op_59594_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_59594_end_0 = const()[name = tensor<string, []>("op_59594_end_0"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<bool, [4]> var_59594_end_mask_0 = const()[name = tensor<string, []>("op_59594_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59594_cast_fp16 = slice_by_index(begin = var_59594_begin_0, end = var_59594_end_0, end_mask = var_59594_end_mask_0, x = q_267_cast_fp16)[name = tensor<string, []>("op_59594_cast_fp16")];
            tensor<int32, [4]> k_535_perm_0 = const()[name = tensor<string, []>("k_535_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_59601_begin_0 = const()[name = tensor<string, []>("op_59601_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_59601_end_0 = const()[name = tensor<string, []>("op_59601_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_59601_end_mask_0 = const()[name = tensor<string, []>("op_59601_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 640]> k_535_cast_fp16 = transpose(perm = k_535_perm_0, x = k_533_cast_fp16)[name = tensor<string, []>("transpose_6")];
            tensor<fp16, [2, 77, 1, 64]> var_59601_cast_fp16 = slice_by_index(begin = var_59601_begin_0, end = var_59601_end_0, end_mask = var_59601_end_mask_0, x = k_535_cast_fp16)[name = tensor<string, []>("op_59601_cast_fp16")];
            tensor<int32, [4]> var_59605_begin_0 = const()[name = tensor<string, []>("op_59605_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_59605_end_0 = const()[name = tensor<string, []>("op_59605_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_59605_end_mask_0 = const()[name = tensor<string, []>("op_59605_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_59605_cast_fp16 = slice_by_index(begin = var_59605_begin_0, end = var_59605_end_0, end_mask = var_59605_end_mask_0, x = k_535_cast_fp16)[name = tensor<string, []>("op_59605_cast_fp16")];
            tensor<int32, [4]> var_59609_begin_0 = const()[name = tensor<string, []>("op_59609_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_59609_end_0 = const()[name = tensor<string, []>("op_59609_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_59609_end_mask_0 = const()[name = tensor<string, []>("op_59609_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_59609_cast_fp16 = slice_by_index(begin = var_59609_begin_0, end = var_59609_end_0, end_mask = var_59609_end_mask_0, x = k_535_cast_fp16)[name = tensor<string, []>("op_59609_cast_fp16")];
            tensor<int32, [4]> var_59613_begin_0 = const()[name = tensor<string, []>("op_59613_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_59613_end_0 = const()[name = tensor<string, []>("op_59613_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_59613_end_mask_0 = const()[name = tensor<string, []>("op_59613_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_59613_cast_fp16 = slice_by_index(begin = var_59613_begin_0, end = var_59613_end_0, end_mask = var_59613_end_mask_0, x = k_535_cast_fp16)[name = tensor<string, []>("op_59613_cast_fp16")];
            tensor<int32, [4]> var_59617_begin_0 = const()[name = tensor<string, []>("op_59617_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_59617_end_0 = const()[name = tensor<string, []>("op_59617_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_59617_end_mask_0 = const()[name = tensor<string, []>("op_59617_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_59617_cast_fp16 = slice_by_index(begin = var_59617_begin_0, end = var_59617_end_0, end_mask = var_59617_end_mask_0, x = k_535_cast_fp16)[name = tensor<string, []>("op_59617_cast_fp16")];
            tensor<int32, [4]> var_59621_begin_0 = const()[name = tensor<string, []>("op_59621_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_59621_end_0 = const()[name = tensor<string, []>("op_59621_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_59621_end_mask_0 = const()[name = tensor<string, []>("op_59621_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_59621_cast_fp16 = slice_by_index(begin = var_59621_begin_0, end = var_59621_end_0, end_mask = var_59621_end_mask_0, x = k_535_cast_fp16)[name = tensor<string, []>("op_59621_cast_fp16")];
            tensor<int32, [4]> var_59625_begin_0 = const()[name = tensor<string, []>("op_59625_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_59625_end_0 = const()[name = tensor<string, []>("op_59625_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_59625_end_mask_0 = const()[name = tensor<string, []>("op_59625_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_59625_cast_fp16 = slice_by_index(begin = var_59625_begin_0, end = var_59625_end_0, end_mask = var_59625_end_mask_0, x = k_535_cast_fp16)[name = tensor<string, []>("op_59625_cast_fp16")];
            tensor<int32, [4]> var_59629_begin_0 = const()[name = tensor<string, []>("op_59629_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_59629_end_0 = const()[name = tensor<string, []>("op_59629_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_59629_end_mask_0 = const()[name = tensor<string, []>("op_59629_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_59629_cast_fp16 = slice_by_index(begin = var_59629_begin_0, end = var_59629_end_0, end_mask = var_59629_end_mask_0, x = k_535_cast_fp16)[name = tensor<string, []>("op_59629_cast_fp16")];
            tensor<int32, [4]> var_59633_begin_0 = const()[name = tensor<string, []>("op_59633_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_59633_end_0 = const()[name = tensor<string, []>("op_59633_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_59633_end_mask_0 = const()[name = tensor<string, []>("op_59633_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_59633_cast_fp16 = slice_by_index(begin = var_59633_begin_0, end = var_59633_end_0, end_mask = var_59633_end_mask_0, x = k_535_cast_fp16)[name = tensor<string, []>("op_59633_cast_fp16")];
            tensor<int32, [4]> var_59637_begin_0 = const()[name = tensor<string, []>("op_59637_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_59637_end_0 = const()[name = tensor<string, []>("op_59637_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_59637_end_mask_0 = const()[name = tensor<string, []>("op_59637_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_59637_cast_fp16 = slice_by_index(begin = var_59637_begin_0, end = var_59637_end_0, end_mask = var_59637_end_mask_0, x = k_535_cast_fp16)[name = tensor<string, []>("op_59637_cast_fp16")];
            tensor<int32, [4]> var_59639_begin_0 = const()[name = tensor<string, []>("op_59639_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_59639_end_0 = const()[name = tensor<string, []>("op_59639_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_59639_end_mask_0 = const()[name = tensor<string, []>("op_59639_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_59639_cast_fp16 = slice_by_index(begin = var_59639_begin_0, end = var_59639_end_0, end_mask = var_59639_end_mask_0, x = v_267_cast_fp16)[name = tensor<string, []>("op_59639_cast_fp16")];
            tensor<int32, [4]> var_59643_begin_0 = const()[name = tensor<string, []>("op_59643_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_59643_end_0 = const()[name = tensor<string, []>("op_59643_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_59643_end_mask_0 = const()[name = tensor<string, []>("op_59643_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_59643_cast_fp16 = slice_by_index(begin = var_59643_begin_0, end = var_59643_end_0, end_mask = var_59643_end_mask_0, x = v_267_cast_fp16)[name = tensor<string, []>("op_59643_cast_fp16")];
            tensor<int32, [4]> var_59647_begin_0 = const()[name = tensor<string, []>("op_59647_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_59647_end_0 = const()[name = tensor<string, []>("op_59647_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_59647_end_mask_0 = const()[name = tensor<string, []>("op_59647_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_59647_cast_fp16 = slice_by_index(begin = var_59647_begin_0, end = var_59647_end_0, end_mask = var_59647_end_mask_0, x = v_267_cast_fp16)[name = tensor<string, []>("op_59647_cast_fp16")];
            tensor<int32, [4]> var_59651_begin_0 = const()[name = tensor<string, []>("op_59651_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_59651_end_0 = const()[name = tensor<string, []>("op_59651_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_59651_end_mask_0 = const()[name = tensor<string, []>("op_59651_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_59651_cast_fp16 = slice_by_index(begin = var_59651_begin_0, end = var_59651_end_0, end_mask = var_59651_end_mask_0, x = v_267_cast_fp16)[name = tensor<string, []>("op_59651_cast_fp16")];
            tensor<int32, [4]> var_59655_begin_0 = const()[name = tensor<string, []>("op_59655_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_59655_end_0 = const()[name = tensor<string, []>("op_59655_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_59655_end_mask_0 = const()[name = tensor<string, []>("op_59655_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_59655_cast_fp16 = slice_by_index(begin = var_59655_begin_0, end = var_59655_end_0, end_mask = var_59655_end_mask_0, x = v_267_cast_fp16)[name = tensor<string, []>("op_59655_cast_fp16")];
            tensor<int32, [4]> var_59659_begin_0 = const()[name = tensor<string, []>("op_59659_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_59659_end_0 = const()[name = tensor<string, []>("op_59659_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_59659_end_mask_0 = const()[name = tensor<string, []>("op_59659_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_59659_cast_fp16 = slice_by_index(begin = var_59659_begin_0, end = var_59659_end_0, end_mask = var_59659_end_mask_0, x = v_267_cast_fp16)[name = tensor<string, []>("op_59659_cast_fp16")];
            tensor<int32, [4]> var_59663_begin_0 = const()[name = tensor<string, []>("op_59663_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_59663_end_0 = const()[name = tensor<string, []>("op_59663_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_59663_end_mask_0 = const()[name = tensor<string, []>("op_59663_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_59663_cast_fp16 = slice_by_index(begin = var_59663_begin_0, end = var_59663_end_0, end_mask = var_59663_end_mask_0, x = v_267_cast_fp16)[name = tensor<string, []>("op_59663_cast_fp16")];
            tensor<int32, [4]> var_59667_begin_0 = const()[name = tensor<string, []>("op_59667_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_59667_end_0 = const()[name = tensor<string, []>("op_59667_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_59667_end_mask_0 = const()[name = tensor<string, []>("op_59667_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_59667_cast_fp16 = slice_by_index(begin = var_59667_begin_0, end = var_59667_end_0, end_mask = var_59667_end_mask_0, x = v_267_cast_fp16)[name = tensor<string, []>("op_59667_cast_fp16")];
            tensor<int32, [4]> var_59671_begin_0 = const()[name = tensor<string, []>("op_59671_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_59671_end_0 = const()[name = tensor<string, []>("op_59671_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_59671_end_mask_0 = const()[name = tensor<string, []>("op_59671_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_59671_cast_fp16 = slice_by_index(begin = var_59671_begin_0, end = var_59671_end_0, end_mask = var_59671_end_mask_0, x = v_267_cast_fp16)[name = tensor<string, []>("op_59671_cast_fp16")];
            tensor<int32, [4]> var_59675_begin_0 = const()[name = tensor<string, []>("op_59675_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_59675_end_0 = const()[name = tensor<string, []>("op_59675_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_59675_end_mask_0 = const()[name = tensor<string, []>("op_59675_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_59675_cast_fp16 = slice_by_index(begin = var_59675_begin_0, end = var_59675_end_0, end_mask = var_59675_end_mask_0, x = v_267_cast_fp16)[name = tensor<string, []>("op_59675_cast_fp16")];
            tensor<string, []> var_59679_equation_0 = const()[name = tensor<string, []>("op_59679_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_59679_cast_fp16 = einsum(equation = var_59679_equation_0, values = (var_59601_cast_fp16, var_59558_cast_fp16))[name = tensor<string, []>("op_59679_cast_fp16")];
            tensor<fp16, []> var_59680_to_fp16 = const()[name = tensor<string, []>("op_59680_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5061_cast_fp16 = mul(x = var_59679_cast_fp16, y = var_59680_to_fp16)[name = tensor<string, []>("aw_5061_cast_fp16")];
            tensor<string, []> var_59683_equation_0 = const()[name = tensor<string, []>("op_59683_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_59683_cast_fp16 = einsum(equation = var_59683_equation_0, values = (var_59605_cast_fp16, var_59562_cast_fp16))[name = tensor<string, []>("op_59683_cast_fp16")];
            tensor<fp16, []> var_59684_to_fp16 = const()[name = tensor<string, []>("op_59684_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5063_cast_fp16 = mul(x = var_59683_cast_fp16, y = var_59684_to_fp16)[name = tensor<string, []>("aw_5063_cast_fp16")];
            tensor<string, []> var_59687_equation_0 = const()[name = tensor<string, []>("op_59687_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_59687_cast_fp16 = einsum(equation = var_59687_equation_0, values = (var_59609_cast_fp16, var_59566_cast_fp16))[name = tensor<string, []>("op_59687_cast_fp16")];
            tensor<fp16, []> var_59688_to_fp16 = const()[name = tensor<string, []>("op_59688_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5065_cast_fp16 = mul(x = var_59687_cast_fp16, y = var_59688_to_fp16)[name = tensor<string, []>("aw_5065_cast_fp16")];
            tensor<string, []> var_59691_equation_0 = const()[name = tensor<string, []>("op_59691_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_59691_cast_fp16 = einsum(equation = var_59691_equation_0, values = (var_59613_cast_fp16, var_59570_cast_fp16))[name = tensor<string, []>("op_59691_cast_fp16")];
            tensor<fp16, []> var_59692_to_fp16 = const()[name = tensor<string, []>("op_59692_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5067_cast_fp16 = mul(x = var_59691_cast_fp16, y = var_59692_to_fp16)[name = tensor<string, []>("aw_5067_cast_fp16")];
            tensor<string, []> var_59695_equation_0 = const()[name = tensor<string, []>("op_59695_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_59695_cast_fp16 = einsum(equation = var_59695_equation_0, values = (var_59617_cast_fp16, var_59574_cast_fp16))[name = tensor<string, []>("op_59695_cast_fp16")];
            tensor<fp16, []> var_59696_to_fp16 = const()[name = tensor<string, []>("op_59696_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5069_cast_fp16 = mul(x = var_59695_cast_fp16, y = var_59696_to_fp16)[name = tensor<string, []>("aw_5069_cast_fp16")];
            tensor<string, []> var_59699_equation_0 = const()[name = tensor<string, []>("op_59699_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_59699_cast_fp16 = einsum(equation = var_59699_equation_0, values = (var_59621_cast_fp16, var_59578_cast_fp16))[name = tensor<string, []>("op_59699_cast_fp16")];
            tensor<fp16, []> var_59700_to_fp16 = const()[name = tensor<string, []>("op_59700_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5071_cast_fp16 = mul(x = var_59699_cast_fp16, y = var_59700_to_fp16)[name = tensor<string, []>("aw_5071_cast_fp16")];
            tensor<string, []> var_59703_equation_0 = const()[name = tensor<string, []>("op_59703_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_59703_cast_fp16 = einsum(equation = var_59703_equation_0, values = (var_59625_cast_fp16, var_59582_cast_fp16))[name = tensor<string, []>("op_59703_cast_fp16")];
            tensor<fp16, []> var_59704_to_fp16 = const()[name = tensor<string, []>("op_59704_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5073_cast_fp16 = mul(x = var_59703_cast_fp16, y = var_59704_to_fp16)[name = tensor<string, []>("aw_5073_cast_fp16")];
            tensor<string, []> var_59707_equation_0 = const()[name = tensor<string, []>("op_59707_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_59707_cast_fp16 = einsum(equation = var_59707_equation_0, values = (var_59629_cast_fp16, var_59586_cast_fp16))[name = tensor<string, []>("op_59707_cast_fp16")];
            tensor<fp16, []> var_59708_to_fp16 = const()[name = tensor<string, []>("op_59708_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5075_cast_fp16 = mul(x = var_59707_cast_fp16, y = var_59708_to_fp16)[name = tensor<string, []>("aw_5075_cast_fp16")];
            tensor<string, []> var_59711_equation_0 = const()[name = tensor<string, []>("op_59711_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_59711_cast_fp16 = einsum(equation = var_59711_equation_0, values = (var_59633_cast_fp16, var_59590_cast_fp16))[name = tensor<string, []>("op_59711_cast_fp16")];
            tensor<fp16, []> var_59712_to_fp16 = const()[name = tensor<string, []>("op_59712_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5077_cast_fp16 = mul(x = var_59711_cast_fp16, y = var_59712_to_fp16)[name = tensor<string, []>("aw_5077_cast_fp16")];
            tensor<string, []> var_59715_equation_0 = const()[name = tensor<string, []>("op_59715_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_59715_cast_fp16 = einsum(equation = var_59715_equation_0, values = (var_59637_cast_fp16, var_59594_cast_fp16))[name = tensor<string, []>("op_59715_cast_fp16")];
            tensor<fp16, []> var_59716_to_fp16 = const()[name = tensor<string, []>("op_59716_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5079_cast_fp16 = mul(x = var_59715_cast_fp16, y = var_59716_to_fp16)[name = tensor<string, []>("aw_5079_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_59718_cast_fp16 = softmax(axis = var_58034, x = aw_5061_cast_fp16)[name = tensor<string, []>("op_59718_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_59719_cast_fp16 = softmax(axis = var_58034, x = aw_5063_cast_fp16)[name = tensor<string, []>("op_59719_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_59720_cast_fp16 = softmax(axis = var_58034, x = aw_5065_cast_fp16)[name = tensor<string, []>("op_59720_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_59721_cast_fp16 = softmax(axis = var_58034, x = aw_5067_cast_fp16)[name = tensor<string, []>("op_59721_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_59722_cast_fp16 = softmax(axis = var_58034, x = aw_5069_cast_fp16)[name = tensor<string, []>("op_59722_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_59723_cast_fp16 = softmax(axis = var_58034, x = aw_5071_cast_fp16)[name = tensor<string, []>("op_59723_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_59724_cast_fp16 = softmax(axis = var_58034, x = aw_5073_cast_fp16)[name = tensor<string, []>("op_59724_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_59725_cast_fp16 = softmax(axis = var_58034, x = aw_5075_cast_fp16)[name = tensor<string, []>("op_59725_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_59726_cast_fp16 = softmax(axis = var_58034, x = aw_5077_cast_fp16)[name = tensor<string, []>("op_59726_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_59727_cast_fp16 = softmax(axis = var_58034, x = aw_5079_cast_fp16)[name = tensor<string, []>("op_59727_cast_fp16")];
            tensor<string, []> var_59729_equation_0 = const()[name = tensor<string, []>("op_59729_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59729_cast_fp16 = einsum(equation = var_59729_equation_0, values = (var_59639_cast_fp16, var_59718_cast_fp16))[name = tensor<string, []>("op_59729_cast_fp16")];
            tensor<string, []> var_59731_equation_0 = const()[name = tensor<string, []>("op_59731_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59731_cast_fp16 = einsum(equation = var_59731_equation_0, values = (var_59643_cast_fp16, var_59719_cast_fp16))[name = tensor<string, []>("op_59731_cast_fp16")];
            tensor<string, []> var_59733_equation_0 = const()[name = tensor<string, []>("op_59733_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59733_cast_fp16 = einsum(equation = var_59733_equation_0, values = (var_59647_cast_fp16, var_59720_cast_fp16))[name = tensor<string, []>("op_59733_cast_fp16")];
            tensor<string, []> var_59735_equation_0 = const()[name = tensor<string, []>("op_59735_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59735_cast_fp16 = einsum(equation = var_59735_equation_0, values = (var_59651_cast_fp16, var_59721_cast_fp16))[name = tensor<string, []>("op_59735_cast_fp16")];
            tensor<string, []> var_59737_equation_0 = const()[name = tensor<string, []>("op_59737_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59737_cast_fp16 = einsum(equation = var_59737_equation_0, values = (var_59655_cast_fp16, var_59722_cast_fp16))[name = tensor<string, []>("op_59737_cast_fp16")];
            tensor<string, []> var_59739_equation_0 = const()[name = tensor<string, []>("op_59739_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59739_cast_fp16 = einsum(equation = var_59739_equation_0, values = (var_59659_cast_fp16, var_59723_cast_fp16))[name = tensor<string, []>("op_59739_cast_fp16")];
            tensor<string, []> var_59741_equation_0 = const()[name = tensor<string, []>("op_59741_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59741_cast_fp16 = einsum(equation = var_59741_equation_0, values = (var_59663_cast_fp16, var_59724_cast_fp16))[name = tensor<string, []>("op_59741_cast_fp16")];
            tensor<string, []> var_59743_equation_0 = const()[name = tensor<string, []>("op_59743_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59743_cast_fp16 = einsum(equation = var_59743_equation_0, values = (var_59667_cast_fp16, var_59725_cast_fp16))[name = tensor<string, []>("op_59743_cast_fp16")];
            tensor<string, []> var_59745_equation_0 = const()[name = tensor<string, []>("op_59745_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59745_cast_fp16 = einsum(equation = var_59745_equation_0, values = (var_59671_cast_fp16, var_59726_cast_fp16))[name = tensor<string, []>("op_59745_cast_fp16")];
            tensor<string, []> var_59747_equation_0 = const()[name = tensor<string, []>("op_59747_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59747_cast_fp16 = einsum(equation = var_59747_equation_0, values = (var_59675_cast_fp16, var_59727_cast_fp16))[name = tensor<string, []>("op_59747_cast_fp16")];
            tensor<bool, []> input_773_interleave_0 = const()[name = tensor<string, []>("input_773_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 640, 1, 4096]> input_773_cast_fp16 = concat(axis = var_58034, interleave = input_773_interleave_0, values = (var_59729_cast_fp16, var_59731_cast_fp16, var_59733_cast_fp16, var_59735_cast_fp16, var_59737_cast_fp16, var_59739_cast_fp16, var_59741_cast_fp16, var_59743_cast_fp16, var_59745_cast_fp16, var_59747_cast_fp16))[name = tensor<string, []>("input_773_cast_fp16")];
            tensor<string, []> var_59757_pad_type_0 = const()[name = tensor<string, []>("op_59757_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_59757_strides_0 = const()[name = tensor<string, []>("op_59757_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_59757_pad_0 = const()[name = tensor<string, []>("op_59757_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_59757_dilations_0 = const()[name = tensor<string, []>("op_59757_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_59757_groups_0 = const()[name = tensor<string, []>("op_59757_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 640, 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, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(913143360))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(913450624))), name = tensor<string, []>("up_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]> 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, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(913450816)))];
            tensor<fp16, [2, 640, 1, 4096]> var_59757_cast_fp16 = conv(bias = up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_59757_dilations_0, groups = var_59757_groups_0, pad = var_59757_pad_0, pad_type = var_59757_pad_type_0, strides = var_59757_strides_0, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized, x = input_773_cast_fp16)[name = tensor<string, []>("op_59757_cast_fp16")];
            tensor<fp16, [2, 640, 1, 4096]> inputs_401_cast_fp16 = add(x = var_59757_cast_fp16, y = inputs_399_cast_fp16)[name = tensor<string, []>("inputs_401_cast_fp16")];
            tensor<int32, [1]> input_775_axes_0 = const()[name = tensor<string, []>("input_775_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [640]> input_775_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_775_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(913452160)))];
            tensor<fp16, [640]> input_775_beta_0_to_fp16 = const()[name = tensor<string, []>("input_775_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(913453504)))];
            tensor<fp16, []> var_59767_to_fp16 = const()[name = tensor<string, []>("op_59767_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 640, 1, 4096]> input_775_cast_fp16 = layer_norm(axes = input_775_axes_0, beta = input_775_beta_0_to_fp16, epsilon = var_59767_to_fp16, gamma = input_775_gamma_0_to_fp16, x = inputs_401_cast_fp16)[name = tensor<string, []>("input_775_cast_fp16")];
            tensor<string, []> var_59787_pad_type_0 = const()[name = tensor<string, []>("op_59787_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_59787_strides_0 = const()[name = tensor<string, []>("op_59787_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_59787_pad_0 = const()[name = tensor<string, []>("op_59787_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_59787_dilations_0 = const()[name = tensor<string, []>("op_59787_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_59787_groups_0 = const()[name = tensor<string, []>("op_59787_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [5120, 640, 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, [2457600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(913454848))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(915912512))), name = tensor<string, []>("up_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]> up_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(915912704)))];
            tensor<fp16, [2, 5120, 1, 4096]> var_59787_cast_fp16 = conv(bias = up_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_59787_dilations_0, groups = var_59787_groups_0, pad = var_59787_pad_0, pad_type = var_59787_pad_type_0, strides = var_59787_strides_0, weight = up_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized, x = input_775_cast_fp16)[name = tensor<string, []>("op_59787_cast_fp16")];
            tensor<int32, [2]> var_59788_split_sizes_0 = const()[name = tensor<string, []>("op_59788_split_sizes_0"), val = tensor<int32, [2]>([2560, 2560])];
            tensor<int32, []> var_59788_axis_0 = const()[name = tensor<string, []>("op_59788_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 2560, 1, 4096]> var_59788_cast_fp16_0, tensor<fp16, [2, 2560, 1, 4096]> var_59788_cast_fp16_1 = split(axis = var_59788_axis_0, split_sizes = var_59788_split_sizes_0, x = var_59787_cast_fp16)[name = tensor<string, []>("op_59788_cast_fp16")];
            tensor<string, []> var_59790_mode_0 = const()[name = tensor<string, []>("op_59790_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 2560, 1, 4096]> var_59790_cast_fp16 = gelu(mode = var_59790_mode_0, x = var_59788_cast_fp16_1)[name = tensor<string, []>("op_59790_cast_fp16")];
            tensor<fp16, [2, 2560, 1, 4096]> input_777_cast_fp16 = mul(x = var_59788_cast_fp16_0, y = var_59790_cast_fp16)[name = tensor<string, []>("input_777_cast_fp16")];
            tensor<string, []> var_59798_pad_type_0 = const()[name = tensor<string, []>("op_59798_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_59798_strides_0 = const()[name = tensor<string, []>("op_59798_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_59798_pad_0 = const()[name = tensor<string, []>("op_59798_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_59798_dilations_0 = const()[name = tensor<string, []>("op_59798_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_59798_groups_0 = const()[name = tensor<string, []>("op_59798_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 2560, 1, 1]> up_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(915923008))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(917151872))), name = tensor<string, []>("up_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]> 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, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(917152064)))];
            tensor<fp16, [2, 640, 1, 4096]> var_59798_cast_fp16 = conv(bias = up_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_59798_dilations_0, groups = var_59798_groups_0, pad = var_59798_pad_0, pad_type = var_59798_pad_type_0, strides = var_59798_strides_0, weight = up_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_777_cast_fp16)[name = tensor<string, []>("op_59798_cast_fp16")];
            tensor<fp16, [2, 640, 1, 4096]> inputs_403_cast_fp16 = add(x = var_59798_cast_fp16, y = inputs_401_cast_fp16)[name = tensor<string, []>("inputs_403_cast_fp16")];
            tensor<int32, [1]> hidden_states_541_axes_0 = const()[name = tensor<string, []>("hidden_states_541_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [640]> hidden_states_541_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_541_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(917153408)))];
            tensor<fp16, [640]> hidden_states_541_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_541_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(917154752)))];
            tensor<fp16, []> var_59814_to_fp16 = const()[name = tensor<string, []>("op_59814_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 640, 1, 4096]> hidden_states_541_cast_fp16 = layer_norm(axes = hidden_states_541_axes_0, beta = hidden_states_541_beta_0_to_fp16, epsilon = var_59814_to_fp16, gamma = hidden_states_541_gamma_0_to_fp16, x = inputs_403_cast_fp16)[name = tensor<string, []>("hidden_states_541_cast_fp16")];
            tensor<string, []> q_269_pad_type_0 = const()[name = tensor<string, []>("q_269_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_269_strides_0 = const()[name = tensor<string, []>("q_269_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_269_pad_0 = const()[name = tensor<string, []>("q_269_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_269_dilations_0 = const()[name = tensor<string, []>("q_269_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_269_groups_0 = const()[name = tensor<string, []>("q_269_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_1_transformer_blocks_1_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(917156096))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(917463360))), name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_1_attn1_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            tensor<fp16, [2, 640, 1, 4096]> q_269_cast_fp16 = conv(dilations = q_269_dilations_0, groups = q_269_groups_0, pad = q_269_pad_0, pad_type = q_269_pad_type_0, strides = q_269_strides_0, weight = up_blocks_1_attentions_1_transformer_blocks_1_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_541_cast_fp16)[name = tensor<string, []>("q_269_cast_fp16")];
            tensor<string, []> k_537_pad_type_0 = const()[name = tensor<string, []>("k_537_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_537_strides_0 = const()[name = tensor<string, []>("k_537_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_537_pad_0 = const()[name = tensor<string, []>("k_537_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_537_dilations_0 = const()[name = tensor<string, []>("k_537_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_537_groups_0 = const()[name = tensor<string, []>("k_537_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_1_transformer_blocks_1_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(917463552))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(917770816))), name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_1_attn1_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            tensor<fp16, [2, 640, 1, 4096]> k_537_cast_fp16 = conv(dilations = k_537_dilations_0, groups = k_537_groups_0, pad = k_537_pad_0, pad_type = k_537_pad_type_0, strides = k_537_strides_0, weight = up_blocks_1_attentions_1_transformer_blocks_1_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_541_cast_fp16)[name = tensor<string, []>("k_537_cast_fp16")];
            tensor<string, []> v_269_pad_type_0 = const()[name = tensor<string, []>("v_269_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_269_strides_0 = const()[name = tensor<string, []>("v_269_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_269_pad_0 = const()[name = tensor<string, []>("v_269_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_269_dilations_0 = const()[name = tensor<string, []>("v_269_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_269_groups_0 = const()[name = tensor<string, []>("v_269_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_1_transformer_blocks_1_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(917771008))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(918078272))), name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_1_attn1_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            tensor<fp16, [2, 640, 1, 4096]> v_269_cast_fp16 = conv(dilations = v_269_dilations_0, groups = v_269_groups_0, pad = v_269_pad_0, pad_type = v_269_pad_type_0, strides = v_269_strides_0, weight = up_blocks_1_attentions_1_transformer_blocks_1_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_541_cast_fp16)[name = tensor<string, []>("v_269_cast_fp16")];
            tensor<int32, [4]> var_59847_begin_0 = const()[name = tensor<string, []>("op_59847_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_59847_end_0 = const()[name = tensor<string, []>("op_59847_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_59847_end_mask_0 = const()[name = tensor<string, []>("op_59847_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59847_cast_fp16 = slice_by_index(begin = var_59847_begin_0, end = var_59847_end_0, end_mask = var_59847_end_mask_0, x = q_269_cast_fp16)[name = tensor<string, []>("op_59847_cast_fp16")];
            tensor<int32, [4]> var_59851_begin_0 = const()[name = tensor<string, []>("op_59851_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_59851_end_0 = const()[name = tensor<string, []>("op_59851_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_59851_end_mask_0 = const()[name = tensor<string, []>("op_59851_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59851_cast_fp16 = slice_by_index(begin = var_59851_begin_0, end = var_59851_end_0, end_mask = var_59851_end_mask_0, x = q_269_cast_fp16)[name = tensor<string, []>("op_59851_cast_fp16")];
            tensor<int32, [4]> var_59855_begin_0 = const()[name = tensor<string, []>("op_59855_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_59855_end_0 = const()[name = tensor<string, []>("op_59855_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_59855_end_mask_0 = const()[name = tensor<string, []>("op_59855_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59855_cast_fp16 = slice_by_index(begin = var_59855_begin_0, end = var_59855_end_0, end_mask = var_59855_end_mask_0, x = q_269_cast_fp16)[name = tensor<string, []>("op_59855_cast_fp16")];
            tensor<int32, [4]> var_59859_begin_0 = const()[name = tensor<string, []>("op_59859_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_59859_end_0 = const()[name = tensor<string, []>("op_59859_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_59859_end_mask_0 = const()[name = tensor<string, []>("op_59859_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59859_cast_fp16 = slice_by_index(begin = var_59859_begin_0, end = var_59859_end_0, end_mask = var_59859_end_mask_0, x = q_269_cast_fp16)[name = tensor<string, []>("op_59859_cast_fp16")];
            tensor<int32, [4]> var_59863_begin_0 = const()[name = tensor<string, []>("op_59863_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_59863_end_0 = const()[name = tensor<string, []>("op_59863_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_59863_end_mask_0 = const()[name = tensor<string, []>("op_59863_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59863_cast_fp16 = slice_by_index(begin = var_59863_begin_0, end = var_59863_end_0, end_mask = var_59863_end_mask_0, x = q_269_cast_fp16)[name = tensor<string, []>("op_59863_cast_fp16")];
            tensor<int32, [4]> var_59867_begin_0 = const()[name = tensor<string, []>("op_59867_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_59867_end_0 = const()[name = tensor<string, []>("op_59867_end_0"), val = tensor<int32, [4]>([2, 384, 1, 4096])];
            tensor<bool, [4]> var_59867_end_mask_0 = const()[name = tensor<string, []>("op_59867_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59867_cast_fp16 = slice_by_index(begin = var_59867_begin_0, end = var_59867_end_0, end_mask = var_59867_end_mask_0, x = q_269_cast_fp16)[name = tensor<string, []>("op_59867_cast_fp16")];
            tensor<int32, [4]> var_59871_begin_0 = const()[name = tensor<string, []>("op_59871_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_59871_end_0 = const()[name = tensor<string, []>("op_59871_end_0"), val = tensor<int32, [4]>([2, 448, 1, 4096])];
            tensor<bool, [4]> var_59871_end_mask_0 = const()[name = tensor<string, []>("op_59871_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59871_cast_fp16 = slice_by_index(begin = var_59871_begin_0, end = var_59871_end_0, end_mask = var_59871_end_mask_0, x = q_269_cast_fp16)[name = tensor<string, []>("op_59871_cast_fp16")];
            tensor<int32, [4]> var_59875_begin_0 = const()[name = tensor<string, []>("op_59875_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_59875_end_0 = const()[name = tensor<string, []>("op_59875_end_0"), val = tensor<int32, [4]>([2, 512, 1, 4096])];
            tensor<bool, [4]> var_59875_end_mask_0 = const()[name = tensor<string, []>("op_59875_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59875_cast_fp16 = slice_by_index(begin = var_59875_begin_0, end = var_59875_end_0, end_mask = var_59875_end_mask_0, x = q_269_cast_fp16)[name = tensor<string, []>("op_59875_cast_fp16")];
            tensor<int32, [4]> var_59879_begin_0 = const()[name = tensor<string, []>("op_59879_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_59879_end_0 = const()[name = tensor<string, []>("op_59879_end_0"), val = tensor<int32, [4]>([2, 576, 1, 4096])];
            tensor<bool, [4]> var_59879_end_mask_0 = const()[name = tensor<string, []>("op_59879_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59879_cast_fp16 = slice_by_index(begin = var_59879_begin_0, end = var_59879_end_0, end_mask = var_59879_end_mask_0, x = q_269_cast_fp16)[name = tensor<string, []>("op_59879_cast_fp16")];
            tensor<int32, [4]> var_59883_begin_0 = const()[name = tensor<string, []>("op_59883_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_59883_end_0 = const()[name = tensor<string, []>("op_59883_end_0"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<bool, [4]> var_59883_end_mask_0 = const()[name = tensor<string, []>("op_59883_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59883_cast_fp16 = slice_by_index(begin = var_59883_begin_0, end = var_59883_end_0, end_mask = var_59883_end_mask_0, x = q_269_cast_fp16)[name = tensor<string, []>("op_59883_cast_fp16")];
            tensor<int32, [4]> k_539_perm_0 = const()[name = tensor<string, []>("k_539_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_59890_begin_0 = const()[name = tensor<string, []>("op_59890_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_59890_end_0 = const()[name = tensor<string, []>("op_59890_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 64])];
            tensor<bool, [4]> var_59890_end_mask_0 = const()[name = tensor<string, []>("op_59890_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 640]> k_539_cast_fp16 = transpose(perm = k_539_perm_0, x = k_537_cast_fp16)[name = tensor<string, []>("transpose_5")];
            tensor<fp16, [2, 4096, 1, 64]> var_59890_cast_fp16 = slice_by_index(begin = var_59890_begin_0, end = var_59890_end_0, end_mask = var_59890_end_mask_0, x = k_539_cast_fp16)[name = tensor<string, []>("op_59890_cast_fp16")];
            tensor<int32, [4]> var_59894_begin_0 = const()[name = tensor<string, []>("op_59894_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_59894_end_0 = const()[name = tensor<string, []>("op_59894_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 128])];
            tensor<bool, [4]> var_59894_end_mask_0 = const()[name = tensor<string, []>("op_59894_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_59894_cast_fp16 = slice_by_index(begin = var_59894_begin_0, end = var_59894_end_0, end_mask = var_59894_end_mask_0, x = k_539_cast_fp16)[name = tensor<string, []>("op_59894_cast_fp16")];
            tensor<int32, [4]> var_59898_begin_0 = const()[name = tensor<string, []>("op_59898_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_59898_end_0 = const()[name = tensor<string, []>("op_59898_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 192])];
            tensor<bool, [4]> var_59898_end_mask_0 = const()[name = tensor<string, []>("op_59898_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_59898_cast_fp16 = slice_by_index(begin = var_59898_begin_0, end = var_59898_end_0, end_mask = var_59898_end_mask_0, x = k_539_cast_fp16)[name = tensor<string, []>("op_59898_cast_fp16")];
            tensor<int32, [4]> var_59902_begin_0 = const()[name = tensor<string, []>("op_59902_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_59902_end_0 = const()[name = tensor<string, []>("op_59902_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 256])];
            tensor<bool, [4]> var_59902_end_mask_0 = const()[name = tensor<string, []>("op_59902_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_59902_cast_fp16 = slice_by_index(begin = var_59902_begin_0, end = var_59902_end_0, end_mask = var_59902_end_mask_0, x = k_539_cast_fp16)[name = tensor<string, []>("op_59902_cast_fp16")];
            tensor<int32, [4]> var_59906_begin_0 = const()[name = tensor<string, []>("op_59906_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_59906_end_0 = const()[name = tensor<string, []>("op_59906_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 320])];
            tensor<bool, [4]> var_59906_end_mask_0 = const()[name = tensor<string, []>("op_59906_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_59906_cast_fp16 = slice_by_index(begin = var_59906_begin_0, end = var_59906_end_0, end_mask = var_59906_end_mask_0, x = k_539_cast_fp16)[name = tensor<string, []>("op_59906_cast_fp16")];
            tensor<int32, [4]> var_59910_begin_0 = const()[name = tensor<string, []>("op_59910_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_59910_end_0 = const()[name = tensor<string, []>("op_59910_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 384])];
            tensor<bool, [4]> var_59910_end_mask_0 = const()[name = tensor<string, []>("op_59910_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_59910_cast_fp16 = slice_by_index(begin = var_59910_begin_0, end = var_59910_end_0, end_mask = var_59910_end_mask_0, x = k_539_cast_fp16)[name = tensor<string, []>("op_59910_cast_fp16")];
            tensor<int32, [4]> var_59914_begin_0 = const()[name = tensor<string, []>("op_59914_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_59914_end_0 = const()[name = tensor<string, []>("op_59914_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 448])];
            tensor<bool, [4]> var_59914_end_mask_0 = const()[name = tensor<string, []>("op_59914_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_59914_cast_fp16 = slice_by_index(begin = var_59914_begin_0, end = var_59914_end_0, end_mask = var_59914_end_mask_0, x = k_539_cast_fp16)[name = tensor<string, []>("op_59914_cast_fp16")];
            tensor<int32, [4]> var_59918_begin_0 = const()[name = tensor<string, []>("op_59918_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_59918_end_0 = const()[name = tensor<string, []>("op_59918_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 512])];
            tensor<bool, [4]> var_59918_end_mask_0 = const()[name = tensor<string, []>("op_59918_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_59918_cast_fp16 = slice_by_index(begin = var_59918_begin_0, end = var_59918_end_0, end_mask = var_59918_end_mask_0, x = k_539_cast_fp16)[name = tensor<string, []>("op_59918_cast_fp16")];
            tensor<int32, [4]> var_59922_begin_0 = const()[name = tensor<string, []>("op_59922_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_59922_end_0 = const()[name = tensor<string, []>("op_59922_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 576])];
            tensor<bool, [4]> var_59922_end_mask_0 = const()[name = tensor<string, []>("op_59922_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_59922_cast_fp16 = slice_by_index(begin = var_59922_begin_0, end = var_59922_end_0, end_mask = var_59922_end_mask_0, x = k_539_cast_fp16)[name = tensor<string, []>("op_59922_cast_fp16")];
            tensor<int32, [4]> var_59926_begin_0 = const()[name = tensor<string, []>("op_59926_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_59926_end_0 = const()[name = tensor<string, []>("op_59926_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 640])];
            tensor<bool, [4]> var_59926_end_mask_0 = const()[name = tensor<string, []>("op_59926_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_59926_cast_fp16 = slice_by_index(begin = var_59926_begin_0, end = var_59926_end_0, end_mask = var_59926_end_mask_0, x = k_539_cast_fp16)[name = tensor<string, []>("op_59926_cast_fp16")];
            tensor<int32, [4]> var_59928_begin_0 = const()[name = tensor<string, []>("op_59928_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_59928_end_0 = const()[name = tensor<string, []>("op_59928_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_59928_end_mask_0 = const()[name = tensor<string, []>("op_59928_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59928_cast_fp16 = slice_by_index(begin = var_59928_begin_0, end = var_59928_end_0, end_mask = var_59928_end_mask_0, x = v_269_cast_fp16)[name = tensor<string, []>("op_59928_cast_fp16")];
            tensor<int32, [4]> var_59932_begin_0 = const()[name = tensor<string, []>("op_59932_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_59932_end_0 = const()[name = tensor<string, []>("op_59932_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_59932_end_mask_0 = const()[name = tensor<string, []>("op_59932_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59932_cast_fp16 = slice_by_index(begin = var_59932_begin_0, end = var_59932_end_0, end_mask = var_59932_end_mask_0, x = v_269_cast_fp16)[name = tensor<string, []>("op_59932_cast_fp16")];
            tensor<int32, [4]> var_59936_begin_0 = const()[name = tensor<string, []>("op_59936_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_59936_end_0 = const()[name = tensor<string, []>("op_59936_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_59936_end_mask_0 = const()[name = tensor<string, []>("op_59936_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59936_cast_fp16 = slice_by_index(begin = var_59936_begin_0, end = var_59936_end_0, end_mask = var_59936_end_mask_0, x = v_269_cast_fp16)[name = tensor<string, []>("op_59936_cast_fp16")];
            tensor<int32, [4]> var_59940_begin_0 = const()[name = tensor<string, []>("op_59940_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_59940_end_0 = const()[name = tensor<string, []>("op_59940_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_59940_end_mask_0 = const()[name = tensor<string, []>("op_59940_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59940_cast_fp16 = slice_by_index(begin = var_59940_begin_0, end = var_59940_end_0, end_mask = var_59940_end_mask_0, x = v_269_cast_fp16)[name = tensor<string, []>("op_59940_cast_fp16")];
            tensor<int32, [4]> var_59944_begin_0 = const()[name = tensor<string, []>("op_59944_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_59944_end_0 = const()[name = tensor<string, []>("op_59944_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_59944_end_mask_0 = const()[name = tensor<string, []>("op_59944_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59944_cast_fp16 = slice_by_index(begin = var_59944_begin_0, end = var_59944_end_0, end_mask = var_59944_end_mask_0, x = v_269_cast_fp16)[name = tensor<string, []>("op_59944_cast_fp16")];
            tensor<int32, [4]> var_59948_begin_0 = const()[name = tensor<string, []>("op_59948_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_59948_end_0 = const()[name = tensor<string, []>("op_59948_end_0"), val = tensor<int32, [4]>([2, 384, 1, 4096])];
            tensor<bool, [4]> var_59948_end_mask_0 = const()[name = tensor<string, []>("op_59948_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59948_cast_fp16 = slice_by_index(begin = var_59948_begin_0, end = var_59948_end_0, end_mask = var_59948_end_mask_0, x = v_269_cast_fp16)[name = tensor<string, []>("op_59948_cast_fp16")];
            tensor<int32, [4]> var_59952_begin_0 = const()[name = tensor<string, []>("op_59952_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_59952_end_0 = const()[name = tensor<string, []>("op_59952_end_0"), val = tensor<int32, [4]>([2, 448, 1, 4096])];
            tensor<bool, [4]> var_59952_end_mask_0 = const()[name = tensor<string, []>("op_59952_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59952_cast_fp16 = slice_by_index(begin = var_59952_begin_0, end = var_59952_end_0, end_mask = var_59952_end_mask_0, x = v_269_cast_fp16)[name = tensor<string, []>("op_59952_cast_fp16")];
            tensor<int32, [4]> var_59956_begin_0 = const()[name = tensor<string, []>("op_59956_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_59956_end_0 = const()[name = tensor<string, []>("op_59956_end_0"), val = tensor<int32, [4]>([2, 512, 1, 4096])];
            tensor<bool, [4]> var_59956_end_mask_0 = const()[name = tensor<string, []>("op_59956_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59956_cast_fp16 = slice_by_index(begin = var_59956_begin_0, end = var_59956_end_0, end_mask = var_59956_end_mask_0, x = v_269_cast_fp16)[name = tensor<string, []>("op_59956_cast_fp16")];
            tensor<int32, [4]> var_59960_begin_0 = const()[name = tensor<string, []>("op_59960_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_59960_end_0 = const()[name = tensor<string, []>("op_59960_end_0"), val = tensor<int32, [4]>([2, 576, 1, 4096])];
            tensor<bool, [4]> var_59960_end_mask_0 = const()[name = tensor<string, []>("op_59960_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59960_cast_fp16 = slice_by_index(begin = var_59960_begin_0, end = var_59960_end_0, end_mask = var_59960_end_mask_0, x = v_269_cast_fp16)[name = tensor<string, []>("op_59960_cast_fp16")];
            tensor<int32, [4]> var_59964_begin_0 = const()[name = tensor<string, []>("op_59964_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_59964_end_0 = const()[name = tensor<string, []>("op_59964_end_0"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<bool, [4]> var_59964_end_mask_0 = const()[name = tensor<string, []>("op_59964_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59964_cast_fp16 = slice_by_index(begin = var_59964_begin_0, end = var_59964_end_0, end_mask = var_59964_end_mask_0, x = v_269_cast_fp16)[name = tensor<string, []>("op_59964_cast_fp16")];
            tensor<string, []> var_59968_equation_0 = const()[name = tensor<string, []>("op_59968_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_59968_cast_fp16 = einsum(equation = var_59968_equation_0, values = (var_59890_cast_fp16, var_59847_cast_fp16))[name = tensor<string, []>("op_59968_cast_fp16")];
            tensor<fp16, []> var_59969_to_fp16 = const()[name = tensor<string, []>("op_59969_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5081_cast_fp16 = mul(x = var_59968_cast_fp16, y = var_59969_to_fp16)[name = tensor<string, []>("aw_5081_cast_fp16")];
            tensor<string, []> var_59972_equation_0 = const()[name = tensor<string, []>("op_59972_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_59972_cast_fp16 = einsum(equation = var_59972_equation_0, values = (var_59894_cast_fp16, var_59851_cast_fp16))[name = tensor<string, []>("op_59972_cast_fp16")];
            tensor<fp16, []> var_59973_to_fp16 = const()[name = tensor<string, []>("op_59973_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5083_cast_fp16 = mul(x = var_59972_cast_fp16, y = var_59973_to_fp16)[name = tensor<string, []>("aw_5083_cast_fp16")];
            tensor<string, []> var_59976_equation_0 = const()[name = tensor<string, []>("op_59976_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_59976_cast_fp16 = einsum(equation = var_59976_equation_0, values = (var_59898_cast_fp16, var_59855_cast_fp16))[name = tensor<string, []>("op_59976_cast_fp16")];
            tensor<fp16, []> var_59977_to_fp16 = const()[name = tensor<string, []>("op_59977_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5085_cast_fp16 = mul(x = var_59976_cast_fp16, y = var_59977_to_fp16)[name = tensor<string, []>("aw_5085_cast_fp16")];
            tensor<string, []> var_59980_equation_0 = const()[name = tensor<string, []>("op_59980_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_59980_cast_fp16 = einsum(equation = var_59980_equation_0, values = (var_59902_cast_fp16, var_59859_cast_fp16))[name = tensor<string, []>("op_59980_cast_fp16")];
            tensor<fp16, []> var_59981_to_fp16 = const()[name = tensor<string, []>("op_59981_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5087_cast_fp16 = mul(x = var_59980_cast_fp16, y = var_59981_to_fp16)[name = tensor<string, []>("aw_5087_cast_fp16")];
            tensor<string, []> var_59984_equation_0 = const()[name = tensor<string, []>("op_59984_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_59984_cast_fp16 = einsum(equation = var_59984_equation_0, values = (var_59906_cast_fp16, var_59863_cast_fp16))[name = tensor<string, []>("op_59984_cast_fp16")];
            tensor<fp16, []> var_59985_to_fp16 = const()[name = tensor<string, []>("op_59985_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5089_cast_fp16 = mul(x = var_59984_cast_fp16, y = var_59985_to_fp16)[name = tensor<string, []>("aw_5089_cast_fp16")];
            tensor<string, []> var_59988_equation_0 = const()[name = tensor<string, []>("op_59988_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_59988_cast_fp16 = einsum(equation = var_59988_equation_0, values = (var_59910_cast_fp16, var_59867_cast_fp16))[name = tensor<string, []>("op_59988_cast_fp16")];
            tensor<fp16, []> var_59989_to_fp16 = const()[name = tensor<string, []>("op_59989_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5091_cast_fp16 = mul(x = var_59988_cast_fp16, y = var_59989_to_fp16)[name = tensor<string, []>("aw_5091_cast_fp16")];
            tensor<string, []> var_59992_equation_0 = const()[name = tensor<string, []>("op_59992_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_59992_cast_fp16 = einsum(equation = var_59992_equation_0, values = (var_59914_cast_fp16, var_59871_cast_fp16))[name = tensor<string, []>("op_59992_cast_fp16")];
            tensor<fp16, []> var_59993_to_fp16 = const()[name = tensor<string, []>("op_59993_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5093_cast_fp16 = mul(x = var_59992_cast_fp16, y = var_59993_to_fp16)[name = tensor<string, []>("aw_5093_cast_fp16")];
            tensor<string, []> var_59996_equation_0 = const()[name = tensor<string, []>("op_59996_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_59996_cast_fp16 = einsum(equation = var_59996_equation_0, values = (var_59918_cast_fp16, var_59875_cast_fp16))[name = tensor<string, []>("op_59996_cast_fp16")];
            tensor<fp16, []> var_59997_to_fp16 = const()[name = tensor<string, []>("op_59997_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5095_cast_fp16 = mul(x = var_59996_cast_fp16, y = var_59997_to_fp16)[name = tensor<string, []>("aw_5095_cast_fp16")];
            tensor<string, []> var_60000_equation_0 = const()[name = tensor<string, []>("op_60000_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60000_cast_fp16 = einsum(equation = var_60000_equation_0, values = (var_59922_cast_fp16, var_59879_cast_fp16))[name = tensor<string, []>("op_60000_cast_fp16")];
            tensor<fp16, []> var_60001_to_fp16 = const()[name = tensor<string, []>("op_60001_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5097_cast_fp16 = mul(x = var_60000_cast_fp16, y = var_60001_to_fp16)[name = tensor<string, []>("aw_5097_cast_fp16")];
            tensor<string, []> var_60004_equation_0 = const()[name = tensor<string, []>("op_60004_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60004_cast_fp16 = einsum(equation = var_60004_equation_0, values = (var_59926_cast_fp16, var_59883_cast_fp16))[name = tensor<string, []>("op_60004_cast_fp16")];
            tensor<fp16, []> var_60005_to_fp16 = const()[name = tensor<string, []>("op_60005_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5099_cast_fp16 = mul(x = var_60004_cast_fp16, y = var_60005_to_fp16)[name = tensor<string, []>("aw_5099_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60007_cast_fp16 = softmax(axis = var_58034, x = aw_5081_cast_fp16)[name = tensor<string, []>("op_60007_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60008_cast_fp16 = softmax(axis = var_58034, x = aw_5083_cast_fp16)[name = tensor<string, []>("op_60008_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60009_cast_fp16 = softmax(axis = var_58034, x = aw_5085_cast_fp16)[name = tensor<string, []>("op_60009_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60010_cast_fp16 = softmax(axis = var_58034, x = aw_5087_cast_fp16)[name = tensor<string, []>("op_60010_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60011_cast_fp16 = softmax(axis = var_58034, x = aw_5089_cast_fp16)[name = tensor<string, []>("op_60011_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60012_cast_fp16 = softmax(axis = var_58034, x = aw_5091_cast_fp16)[name = tensor<string, []>("op_60012_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60013_cast_fp16 = softmax(axis = var_58034, x = aw_5093_cast_fp16)[name = tensor<string, []>("op_60013_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60014_cast_fp16 = softmax(axis = var_58034, x = aw_5095_cast_fp16)[name = tensor<string, []>("op_60014_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60015_cast_fp16 = softmax(axis = var_58034, x = aw_5097_cast_fp16)[name = tensor<string, []>("op_60015_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60016_cast_fp16 = softmax(axis = var_58034, x = aw_5099_cast_fp16)[name = tensor<string, []>("op_60016_cast_fp16")];
            tensor<string, []> var_60018_equation_0 = const()[name = tensor<string, []>("op_60018_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60018_cast_fp16 = einsum(equation = var_60018_equation_0, values = (var_59928_cast_fp16, var_60007_cast_fp16))[name = tensor<string, []>("op_60018_cast_fp16")];
            tensor<string, []> var_60020_equation_0 = const()[name = tensor<string, []>("op_60020_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60020_cast_fp16 = einsum(equation = var_60020_equation_0, values = (var_59932_cast_fp16, var_60008_cast_fp16))[name = tensor<string, []>("op_60020_cast_fp16")];
            tensor<string, []> var_60022_equation_0 = const()[name = tensor<string, []>("op_60022_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60022_cast_fp16 = einsum(equation = var_60022_equation_0, values = (var_59936_cast_fp16, var_60009_cast_fp16))[name = tensor<string, []>("op_60022_cast_fp16")];
            tensor<string, []> var_60024_equation_0 = const()[name = tensor<string, []>("op_60024_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60024_cast_fp16 = einsum(equation = var_60024_equation_0, values = (var_59940_cast_fp16, var_60010_cast_fp16))[name = tensor<string, []>("op_60024_cast_fp16")];
            tensor<string, []> var_60026_equation_0 = const()[name = tensor<string, []>("op_60026_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60026_cast_fp16 = einsum(equation = var_60026_equation_0, values = (var_59944_cast_fp16, var_60011_cast_fp16))[name = tensor<string, []>("op_60026_cast_fp16")];
            tensor<string, []> var_60028_equation_0 = const()[name = tensor<string, []>("op_60028_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60028_cast_fp16 = einsum(equation = var_60028_equation_0, values = (var_59948_cast_fp16, var_60012_cast_fp16))[name = tensor<string, []>("op_60028_cast_fp16")];
            tensor<string, []> var_60030_equation_0 = const()[name = tensor<string, []>("op_60030_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60030_cast_fp16 = einsum(equation = var_60030_equation_0, values = (var_59952_cast_fp16, var_60013_cast_fp16))[name = tensor<string, []>("op_60030_cast_fp16")];
            tensor<string, []> var_60032_equation_0 = const()[name = tensor<string, []>("op_60032_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60032_cast_fp16 = einsum(equation = var_60032_equation_0, values = (var_59956_cast_fp16, var_60014_cast_fp16))[name = tensor<string, []>("op_60032_cast_fp16")];
            tensor<string, []> var_60034_equation_0 = const()[name = tensor<string, []>("op_60034_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60034_cast_fp16 = einsum(equation = var_60034_equation_0, values = (var_59960_cast_fp16, var_60015_cast_fp16))[name = tensor<string, []>("op_60034_cast_fp16")];
            tensor<string, []> var_60036_equation_0 = const()[name = tensor<string, []>("op_60036_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60036_cast_fp16 = einsum(equation = var_60036_equation_0, values = (var_59964_cast_fp16, var_60016_cast_fp16))[name = tensor<string, []>("op_60036_cast_fp16")];
            tensor<bool, []> input_779_interleave_0 = const()[name = tensor<string, []>("input_779_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 640, 1, 4096]> input_779_cast_fp16 = concat(axis = var_58034, interleave = input_779_interleave_0, values = (var_60018_cast_fp16, var_60020_cast_fp16, var_60022_cast_fp16, var_60024_cast_fp16, var_60026_cast_fp16, var_60028_cast_fp16, var_60030_cast_fp16, var_60032_cast_fp16, var_60034_cast_fp16, var_60036_cast_fp16))[name = tensor<string, []>("input_779_cast_fp16")];
            tensor<string, []> var_60046_pad_type_0 = const()[name = tensor<string, []>("op_60046_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_60046_strides_0 = const()[name = tensor<string, []>("op_60046_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_60046_pad_0 = const()[name = tensor<string, []>("op_60046_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_60046_dilations_0 = const()[name = tensor<string, []>("op_60046_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_60046_groups_0 = const()[name = tensor<string, []>("op_60046_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_1_transformer_blocks_1_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(918078464))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(918385728))), name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_1_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            tensor<fp16, [640]> up_blocks_1_attentions_1_transformer_blocks_1_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_1_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(918385920)))];
            tensor<fp16, [2, 640, 1, 4096]> var_60046_cast_fp16 = conv(bias = up_blocks_1_attentions_1_transformer_blocks_1_attn1_to_out_0_bias_to_fp16, dilations = var_60046_dilations_0, groups = var_60046_groups_0, pad = var_60046_pad_0, pad_type = var_60046_pad_type_0, strides = var_60046_strides_0, weight = up_blocks_1_attentions_1_transformer_blocks_1_attn1_to_out_0_weight_to_fp16_palettized, x = input_779_cast_fp16)[name = tensor<string, []>("op_60046_cast_fp16")];
            tensor<fp16, [2, 640, 1, 4096]> inputs_405_cast_fp16 = add(x = var_60046_cast_fp16, y = inputs_403_cast_fp16)[name = tensor<string, []>("inputs_405_cast_fp16")];
            tensor<int32, [1]> hidden_states_543_axes_0 = const()[name = tensor<string, []>("hidden_states_543_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [640]> hidden_states_543_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_543_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(918387264)))];
            tensor<fp16, [640]> hidden_states_543_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_543_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(918388608)))];
            tensor<fp16, []> var_60056_to_fp16 = const()[name = tensor<string, []>("op_60056_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 640, 1, 4096]> hidden_states_543_cast_fp16 = layer_norm(axes = hidden_states_543_axes_0, beta = hidden_states_543_beta_0_to_fp16, epsilon = var_60056_to_fp16, gamma = hidden_states_543_gamma_0_to_fp16, x = inputs_405_cast_fp16)[name = tensor<string, []>("hidden_states_543_cast_fp16")];
            tensor<string, []> q_271_pad_type_0 = const()[name = tensor<string, []>("q_271_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_271_strides_0 = const()[name = tensor<string, []>("q_271_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_271_pad_0 = const()[name = tensor<string, []>("q_271_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_271_dilations_0 = const()[name = tensor<string, []>("q_271_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_271_groups_0 = const()[name = tensor<string, []>("q_271_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_1_transformer_blocks_1_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(918389952))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(918697216))), name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_1_attn2_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            tensor<fp16, [2, 640, 1, 4096]> q_271_cast_fp16 = conv(dilations = q_271_dilations_0, groups = q_271_groups_0, pad = q_271_pad_0, pad_type = q_271_pad_type_0, strides = q_271_strides_0, weight = up_blocks_1_attentions_1_transformer_blocks_1_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_543_cast_fp16)[name = tensor<string, []>("q_271_cast_fp16")];
            tensor<string, []> k_541_pad_type_0 = const()[name = tensor<string, []>("k_541_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_541_strides_0 = const()[name = tensor<string, []>("k_541_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_541_pad_0 = const()[name = tensor<string, []>("k_541_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_541_dilations_0 = const()[name = tensor<string, []>("k_541_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_541_groups_0 = const()[name = tensor<string, []>("k_541_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 2048, 1, 1]> up_blocks_1_attentions_1_transformer_blocks_1_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [983040]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(918697408))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(919680512))), name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_1_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 2048, 1, 1])];
            tensor<fp16, [2, 640, 1, 77]> k_541_cast_fp16 = conv(dilations = k_541_dilations_0, groups = k_541_groups_0, pad = k_541_pad_0, pad_type = k_541_pad_type_0, strides = k_541_strides_0, weight = up_blocks_1_attentions_1_transformer_blocks_1_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("k_541_cast_fp16")];
            tensor<string, []> v_271_pad_type_0 = const()[name = tensor<string, []>("v_271_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_271_strides_0 = const()[name = tensor<string, []>("v_271_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_271_pad_0 = const()[name = tensor<string, []>("v_271_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_271_dilations_0 = const()[name = tensor<string, []>("v_271_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_271_groups_0 = const()[name = tensor<string, []>("v_271_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 2048, 1, 1]> up_blocks_1_attentions_1_transformer_blocks_1_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [983040]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(919680704))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(920663808))), name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_1_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 2048, 1, 1])];
            tensor<fp16, [2, 640, 1, 77]> v_271_cast_fp16 = conv(dilations = v_271_dilations_0, groups = v_271_groups_0, pad = v_271_pad_0, pad_type = v_271_pad_type_0, strides = v_271_strides_0, weight = up_blocks_1_attentions_1_transformer_blocks_1_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("v_271_cast_fp16")];
            tensor<int32, [4]> var_60089_begin_0 = const()[name = tensor<string, []>("op_60089_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_60089_end_0 = const()[name = tensor<string, []>("op_60089_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_60089_end_mask_0 = const()[name = tensor<string, []>("op_60089_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60089_cast_fp16 = slice_by_index(begin = var_60089_begin_0, end = var_60089_end_0, end_mask = var_60089_end_mask_0, x = q_271_cast_fp16)[name = tensor<string, []>("op_60089_cast_fp16")];
            tensor<int32, [4]> var_60093_begin_0 = const()[name = tensor<string, []>("op_60093_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_60093_end_0 = const()[name = tensor<string, []>("op_60093_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_60093_end_mask_0 = const()[name = tensor<string, []>("op_60093_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60093_cast_fp16 = slice_by_index(begin = var_60093_begin_0, end = var_60093_end_0, end_mask = var_60093_end_mask_0, x = q_271_cast_fp16)[name = tensor<string, []>("op_60093_cast_fp16")];
            tensor<int32, [4]> var_60097_begin_0 = const()[name = tensor<string, []>("op_60097_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_60097_end_0 = const()[name = tensor<string, []>("op_60097_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_60097_end_mask_0 = const()[name = tensor<string, []>("op_60097_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60097_cast_fp16 = slice_by_index(begin = var_60097_begin_0, end = var_60097_end_0, end_mask = var_60097_end_mask_0, x = q_271_cast_fp16)[name = tensor<string, []>("op_60097_cast_fp16")];
            tensor<int32, [4]> var_60101_begin_0 = const()[name = tensor<string, []>("op_60101_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_60101_end_0 = const()[name = tensor<string, []>("op_60101_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_60101_end_mask_0 = const()[name = tensor<string, []>("op_60101_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60101_cast_fp16 = slice_by_index(begin = var_60101_begin_0, end = var_60101_end_0, end_mask = var_60101_end_mask_0, x = q_271_cast_fp16)[name = tensor<string, []>("op_60101_cast_fp16")];
            tensor<int32, [4]> var_60105_begin_0 = const()[name = tensor<string, []>("op_60105_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_60105_end_0 = const()[name = tensor<string, []>("op_60105_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_60105_end_mask_0 = const()[name = tensor<string, []>("op_60105_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60105_cast_fp16 = slice_by_index(begin = var_60105_begin_0, end = var_60105_end_0, end_mask = var_60105_end_mask_0, x = q_271_cast_fp16)[name = tensor<string, []>("op_60105_cast_fp16")];
            tensor<int32, [4]> var_60109_begin_0 = const()[name = tensor<string, []>("op_60109_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_60109_end_0 = const()[name = tensor<string, []>("op_60109_end_0"), val = tensor<int32, [4]>([2, 384, 1, 4096])];
            tensor<bool, [4]> var_60109_end_mask_0 = const()[name = tensor<string, []>("op_60109_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60109_cast_fp16 = slice_by_index(begin = var_60109_begin_0, end = var_60109_end_0, end_mask = var_60109_end_mask_0, x = q_271_cast_fp16)[name = tensor<string, []>("op_60109_cast_fp16")];
            tensor<int32, [4]> var_60113_begin_0 = const()[name = tensor<string, []>("op_60113_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_60113_end_0 = const()[name = tensor<string, []>("op_60113_end_0"), val = tensor<int32, [4]>([2, 448, 1, 4096])];
            tensor<bool, [4]> var_60113_end_mask_0 = const()[name = tensor<string, []>("op_60113_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60113_cast_fp16 = slice_by_index(begin = var_60113_begin_0, end = var_60113_end_0, end_mask = var_60113_end_mask_0, x = q_271_cast_fp16)[name = tensor<string, []>("op_60113_cast_fp16")];
            tensor<int32, [4]> var_60117_begin_0 = const()[name = tensor<string, []>("op_60117_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_60117_end_0 = const()[name = tensor<string, []>("op_60117_end_0"), val = tensor<int32, [4]>([2, 512, 1, 4096])];
            tensor<bool, [4]> var_60117_end_mask_0 = const()[name = tensor<string, []>("op_60117_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60117_cast_fp16 = slice_by_index(begin = var_60117_begin_0, end = var_60117_end_0, end_mask = var_60117_end_mask_0, x = q_271_cast_fp16)[name = tensor<string, []>("op_60117_cast_fp16")];
            tensor<int32, [4]> var_60121_begin_0 = const()[name = tensor<string, []>("op_60121_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_60121_end_0 = const()[name = tensor<string, []>("op_60121_end_0"), val = tensor<int32, [4]>([2, 576, 1, 4096])];
            tensor<bool, [4]> var_60121_end_mask_0 = const()[name = tensor<string, []>("op_60121_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60121_cast_fp16 = slice_by_index(begin = var_60121_begin_0, end = var_60121_end_0, end_mask = var_60121_end_mask_0, x = q_271_cast_fp16)[name = tensor<string, []>("op_60121_cast_fp16")];
            tensor<int32, [4]> var_60125_begin_0 = const()[name = tensor<string, []>("op_60125_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_60125_end_0 = const()[name = tensor<string, []>("op_60125_end_0"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<bool, [4]> var_60125_end_mask_0 = const()[name = tensor<string, []>("op_60125_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60125_cast_fp16 = slice_by_index(begin = var_60125_begin_0, end = var_60125_end_0, end_mask = var_60125_end_mask_0, x = q_271_cast_fp16)[name = tensor<string, []>("op_60125_cast_fp16")];
            tensor<int32, [4]> k_543_perm_0 = const()[name = tensor<string, []>("k_543_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_60132_begin_0 = const()[name = tensor<string, []>("op_60132_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_60132_end_0 = const()[name = tensor<string, []>("op_60132_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_60132_end_mask_0 = const()[name = tensor<string, []>("op_60132_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 640]> k_543_cast_fp16 = transpose(perm = k_543_perm_0, x = k_541_cast_fp16)[name = tensor<string, []>("transpose_4")];
            tensor<fp16, [2, 77, 1, 64]> var_60132_cast_fp16 = slice_by_index(begin = var_60132_begin_0, end = var_60132_end_0, end_mask = var_60132_end_mask_0, x = k_543_cast_fp16)[name = tensor<string, []>("op_60132_cast_fp16")];
            tensor<int32, [4]> var_60136_begin_0 = const()[name = tensor<string, []>("op_60136_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_60136_end_0 = const()[name = tensor<string, []>("op_60136_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_60136_end_mask_0 = const()[name = tensor<string, []>("op_60136_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_60136_cast_fp16 = slice_by_index(begin = var_60136_begin_0, end = var_60136_end_0, end_mask = var_60136_end_mask_0, x = k_543_cast_fp16)[name = tensor<string, []>("op_60136_cast_fp16")];
            tensor<int32, [4]> var_60140_begin_0 = const()[name = tensor<string, []>("op_60140_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_60140_end_0 = const()[name = tensor<string, []>("op_60140_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_60140_end_mask_0 = const()[name = tensor<string, []>("op_60140_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_60140_cast_fp16 = slice_by_index(begin = var_60140_begin_0, end = var_60140_end_0, end_mask = var_60140_end_mask_0, x = k_543_cast_fp16)[name = tensor<string, []>("op_60140_cast_fp16")];
            tensor<int32, [4]> var_60144_begin_0 = const()[name = tensor<string, []>("op_60144_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_60144_end_0 = const()[name = tensor<string, []>("op_60144_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_60144_end_mask_0 = const()[name = tensor<string, []>("op_60144_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_60144_cast_fp16 = slice_by_index(begin = var_60144_begin_0, end = var_60144_end_0, end_mask = var_60144_end_mask_0, x = k_543_cast_fp16)[name = tensor<string, []>("op_60144_cast_fp16")];
            tensor<int32, [4]> var_60148_begin_0 = const()[name = tensor<string, []>("op_60148_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_60148_end_0 = const()[name = tensor<string, []>("op_60148_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_60148_end_mask_0 = const()[name = tensor<string, []>("op_60148_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_60148_cast_fp16 = slice_by_index(begin = var_60148_begin_0, end = var_60148_end_0, end_mask = var_60148_end_mask_0, x = k_543_cast_fp16)[name = tensor<string, []>("op_60148_cast_fp16")];
            tensor<int32, [4]> var_60152_begin_0 = const()[name = tensor<string, []>("op_60152_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_60152_end_0 = const()[name = tensor<string, []>("op_60152_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_60152_end_mask_0 = const()[name = tensor<string, []>("op_60152_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_60152_cast_fp16 = slice_by_index(begin = var_60152_begin_0, end = var_60152_end_0, end_mask = var_60152_end_mask_0, x = k_543_cast_fp16)[name = tensor<string, []>("op_60152_cast_fp16")];
            tensor<int32, [4]> var_60156_begin_0 = const()[name = tensor<string, []>("op_60156_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_60156_end_0 = const()[name = tensor<string, []>("op_60156_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_60156_end_mask_0 = const()[name = tensor<string, []>("op_60156_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_60156_cast_fp16 = slice_by_index(begin = var_60156_begin_0, end = var_60156_end_0, end_mask = var_60156_end_mask_0, x = k_543_cast_fp16)[name = tensor<string, []>("op_60156_cast_fp16")];
            tensor<int32, [4]> var_60160_begin_0 = const()[name = tensor<string, []>("op_60160_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_60160_end_0 = const()[name = tensor<string, []>("op_60160_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_60160_end_mask_0 = const()[name = tensor<string, []>("op_60160_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_60160_cast_fp16 = slice_by_index(begin = var_60160_begin_0, end = var_60160_end_0, end_mask = var_60160_end_mask_0, x = k_543_cast_fp16)[name = tensor<string, []>("op_60160_cast_fp16")];
            tensor<int32, [4]> var_60164_begin_0 = const()[name = tensor<string, []>("op_60164_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_60164_end_0 = const()[name = tensor<string, []>("op_60164_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_60164_end_mask_0 = const()[name = tensor<string, []>("op_60164_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_60164_cast_fp16 = slice_by_index(begin = var_60164_begin_0, end = var_60164_end_0, end_mask = var_60164_end_mask_0, x = k_543_cast_fp16)[name = tensor<string, []>("op_60164_cast_fp16")];
            tensor<int32, [4]> var_60168_begin_0 = const()[name = tensor<string, []>("op_60168_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_60168_end_0 = const()[name = tensor<string, []>("op_60168_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_60168_end_mask_0 = const()[name = tensor<string, []>("op_60168_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_60168_cast_fp16 = slice_by_index(begin = var_60168_begin_0, end = var_60168_end_0, end_mask = var_60168_end_mask_0, x = k_543_cast_fp16)[name = tensor<string, []>("op_60168_cast_fp16")];
            tensor<int32, [4]> var_60170_begin_0 = const()[name = tensor<string, []>("op_60170_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_60170_end_0 = const()[name = tensor<string, []>("op_60170_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_60170_end_mask_0 = const()[name = tensor<string, []>("op_60170_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_60170_cast_fp16 = slice_by_index(begin = var_60170_begin_0, end = var_60170_end_0, end_mask = var_60170_end_mask_0, x = v_271_cast_fp16)[name = tensor<string, []>("op_60170_cast_fp16")];
            tensor<int32, [4]> var_60174_begin_0 = const()[name = tensor<string, []>("op_60174_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_60174_end_0 = const()[name = tensor<string, []>("op_60174_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_60174_end_mask_0 = const()[name = tensor<string, []>("op_60174_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_60174_cast_fp16 = slice_by_index(begin = var_60174_begin_0, end = var_60174_end_0, end_mask = var_60174_end_mask_0, x = v_271_cast_fp16)[name = tensor<string, []>("op_60174_cast_fp16")];
            tensor<int32, [4]> var_60178_begin_0 = const()[name = tensor<string, []>("op_60178_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_60178_end_0 = const()[name = tensor<string, []>("op_60178_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_60178_end_mask_0 = const()[name = tensor<string, []>("op_60178_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_60178_cast_fp16 = slice_by_index(begin = var_60178_begin_0, end = var_60178_end_0, end_mask = var_60178_end_mask_0, x = v_271_cast_fp16)[name = tensor<string, []>("op_60178_cast_fp16")];
            tensor<int32, [4]> var_60182_begin_0 = const()[name = tensor<string, []>("op_60182_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_60182_end_0 = const()[name = tensor<string, []>("op_60182_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_60182_end_mask_0 = const()[name = tensor<string, []>("op_60182_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_60182_cast_fp16 = slice_by_index(begin = var_60182_begin_0, end = var_60182_end_0, end_mask = var_60182_end_mask_0, x = v_271_cast_fp16)[name = tensor<string, []>("op_60182_cast_fp16")];
            tensor<int32, [4]> var_60186_begin_0 = const()[name = tensor<string, []>("op_60186_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_60186_end_0 = const()[name = tensor<string, []>("op_60186_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_60186_end_mask_0 = const()[name = tensor<string, []>("op_60186_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_60186_cast_fp16 = slice_by_index(begin = var_60186_begin_0, end = var_60186_end_0, end_mask = var_60186_end_mask_0, x = v_271_cast_fp16)[name = tensor<string, []>("op_60186_cast_fp16")];
            tensor<int32, [4]> var_60190_begin_0 = const()[name = tensor<string, []>("op_60190_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_60190_end_0 = const()[name = tensor<string, []>("op_60190_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_60190_end_mask_0 = const()[name = tensor<string, []>("op_60190_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_60190_cast_fp16 = slice_by_index(begin = var_60190_begin_0, end = var_60190_end_0, end_mask = var_60190_end_mask_0, x = v_271_cast_fp16)[name = tensor<string, []>("op_60190_cast_fp16")];
            tensor<int32, [4]> var_60194_begin_0 = const()[name = tensor<string, []>("op_60194_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_60194_end_0 = const()[name = tensor<string, []>("op_60194_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_60194_end_mask_0 = const()[name = tensor<string, []>("op_60194_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_60194_cast_fp16 = slice_by_index(begin = var_60194_begin_0, end = var_60194_end_0, end_mask = var_60194_end_mask_0, x = v_271_cast_fp16)[name = tensor<string, []>("op_60194_cast_fp16")];
            tensor<int32, [4]> var_60198_begin_0 = const()[name = tensor<string, []>("op_60198_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_60198_end_0 = const()[name = tensor<string, []>("op_60198_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_60198_end_mask_0 = const()[name = tensor<string, []>("op_60198_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_60198_cast_fp16 = slice_by_index(begin = var_60198_begin_0, end = var_60198_end_0, end_mask = var_60198_end_mask_0, x = v_271_cast_fp16)[name = tensor<string, []>("op_60198_cast_fp16")];
            tensor<int32, [4]> var_60202_begin_0 = const()[name = tensor<string, []>("op_60202_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_60202_end_0 = const()[name = tensor<string, []>("op_60202_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_60202_end_mask_0 = const()[name = tensor<string, []>("op_60202_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_60202_cast_fp16 = slice_by_index(begin = var_60202_begin_0, end = var_60202_end_0, end_mask = var_60202_end_mask_0, x = v_271_cast_fp16)[name = tensor<string, []>("op_60202_cast_fp16")];
            tensor<int32, [4]> var_60206_begin_0 = const()[name = tensor<string, []>("op_60206_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_60206_end_0 = const()[name = tensor<string, []>("op_60206_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_60206_end_mask_0 = const()[name = tensor<string, []>("op_60206_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_60206_cast_fp16 = slice_by_index(begin = var_60206_begin_0, end = var_60206_end_0, end_mask = var_60206_end_mask_0, x = v_271_cast_fp16)[name = tensor<string, []>("op_60206_cast_fp16")];
            tensor<string, []> var_60210_equation_0 = const()[name = tensor<string, []>("op_60210_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_60210_cast_fp16 = einsum(equation = var_60210_equation_0, values = (var_60132_cast_fp16, var_60089_cast_fp16))[name = tensor<string, []>("op_60210_cast_fp16")];
            tensor<fp16, []> var_60211_to_fp16 = const()[name = tensor<string, []>("op_60211_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5101_cast_fp16 = mul(x = var_60210_cast_fp16, y = var_60211_to_fp16)[name = tensor<string, []>("aw_5101_cast_fp16")];
            tensor<string, []> var_60214_equation_0 = const()[name = tensor<string, []>("op_60214_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_60214_cast_fp16 = einsum(equation = var_60214_equation_0, values = (var_60136_cast_fp16, var_60093_cast_fp16))[name = tensor<string, []>("op_60214_cast_fp16")];
            tensor<fp16, []> var_60215_to_fp16 = const()[name = tensor<string, []>("op_60215_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5103_cast_fp16 = mul(x = var_60214_cast_fp16, y = var_60215_to_fp16)[name = tensor<string, []>("aw_5103_cast_fp16")];
            tensor<string, []> var_60218_equation_0 = const()[name = tensor<string, []>("op_60218_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_60218_cast_fp16 = einsum(equation = var_60218_equation_0, values = (var_60140_cast_fp16, var_60097_cast_fp16))[name = tensor<string, []>("op_60218_cast_fp16")];
            tensor<fp16, []> var_60219_to_fp16 = const()[name = tensor<string, []>("op_60219_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5105_cast_fp16 = mul(x = var_60218_cast_fp16, y = var_60219_to_fp16)[name = tensor<string, []>("aw_5105_cast_fp16")];
            tensor<string, []> var_60222_equation_0 = const()[name = tensor<string, []>("op_60222_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_60222_cast_fp16 = einsum(equation = var_60222_equation_0, values = (var_60144_cast_fp16, var_60101_cast_fp16))[name = tensor<string, []>("op_60222_cast_fp16")];
            tensor<fp16, []> var_60223_to_fp16 = const()[name = tensor<string, []>("op_60223_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5107_cast_fp16 = mul(x = var_60222_cast_fp16, y = var_60223_to_fp16)[name = tensor<string, []>("aw_5107_cast_fp16")];
            tensor<string, []> var_60226_equation_0 = const()[name = tensor<string, []>("op_60226_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_60226_cast_fp16 = einsum(equation = var_60226_equation_0, values = (var_60148_cast_fp16, var_60105_cast_fp16))[name = tensor<string, []>("op_60226_cast_fp16")];
            tensor<fp16, []> var_60227_to_fp16 = const()[name = tensor<string, []>("op_60227_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5109_cast_fp16 = mul(x = var_60226_cast_fp16, y = var_60227_to_fp16)[name = tensor<string, []>("aw_5109_cast_fp16")];
            tensor<string, []> var_60230_equation_0 = const()[name = tensor<string, []>("op_60230_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_60230_cast_fp16 = einsum(equation = var_60230_equation_0, values = (var_60152_cast_fp16, var_60109_cast_fp16))[name = tensor<string, []>("op_60230_cast_fp16")];
            tensor<fp16, []> var_60231_to_fp16 = const()[name = tensor<string, []>("op_60231_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5111_cast_fp16 = mul(x = var_60230_cast_fp16, y = var_60231_to_fp16)[name = tensor<string, []>("aw_5111_cast_fp16")];
            tensor<string, []> var_60234_equation_0 = const()[name = tensor<string, []>("op_60234_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_60234_cast_fp16 = einsum(equation = var_60234_equation_0, values = (var_60156_cast_fp16, var_60113_cast_fp16))[name = tensor<string, []>("op_60234_cast_fp16")];
            tensor<fp16, []> var_60235_to_fp16 = const()[name = tensor<string, []>("op_60235_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5113_cast_fp16 = mul(x = var_60234_cast_fp16, y = var_60235_to_fp16)[name = tensor<string, []>("aw_5113_cast_fp16")];
            tensor<string, []> var_60238_equation_0 = const()[name = tensor<string, []>("op_60238_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_60238_cast_fp16 = einsum(equation = var_60238_equation_0, values = (var_60160_cast_fp16, var_60117_cast_fp16))[name = tensor<string, []>("op_60238_cast_fp16")];
            tensor<fp16, []> var_60239_to_fp16 = const()[name = tensor<string, []>("op_60239_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5115_cast_fp16 = mul(x = var_60238_cast_fp16, y = var_60239_to_fp16)[name = tensor<string, []>("aw_5115_cast_fp16")];
            tensor<string, []> var_60242_equation_0 = const()[name = tensor<string, []>("op_60242_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_60242_cast_fp16 = einsum(equation = var_60242_equation_0, values = (var_60164_cast_fp16, var_60121_cast_fp16))[name = tensor<string, []>("op_60242_cast_fp16")];
            tensor<fp16, []> var_60243_to_fp16 = const()[name = tensor<string, []>("op_60243_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5117_cast_fp16 = mul(x = var_60242_cast_fp16, y = var_60243_to_fp16)[name = tensor<string, []>("aw_5117_cast_fp16")];
            tensor<string, []> var_60246_equation_0 = const()[name = tensor<string, []>("op_60246_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_60246_cast_fp16 = einsum(equation = var_60246_equation_0, values = (var_60168_cast_fp16, var_60125_cast_fp16))[name = tensor<string, []>("op_60246_cast_fp16")];
            tensor<fp16, []> var_60247_to_fp16 = const()[name = tensor<string, []>("op_60247_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5119_cast_fp16 = mul(x = var_60246_cast_fp16, y = var_60247_to_fp16)[name = tensor<string, []>("aw_5119_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_60249_cast_fp16 = softmax(axis = var_58034, x = aw_5101_cast_fp16)[name = tensor<string, []>("op_60249_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_60250_cast_fp16 = softmax(axis = var_58034, x = aw_5103_cast_fp16)[name = tensor<string, []>("op_60250_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_60251_cast_fp16 = softmax(axis = var_58034, x = aw_5105_cast_fp16)[name = tensor<string, []>("op_60251_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_60252_cast_fp16 = softmax(axis = var_58034, x = aw_5107_cast_fp16)[name = tensor<string, []>("op_60252_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_60253_cast_fp16 = softmax(axis = var_58034, x = aw_5109_cast_fp16)[name = tensor<string, []>("op_60253_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_60254_cast_fp16 = softmax(axis = var_58034, x = aw_5111_cast_fp16)[name = tensor<string, []>("op_60254_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_60255_cast_fp16 = softmax(axis = var_58034, x = aw_5113_cast_fp16)[name = tensor<string, []>("op_60255_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_60256_cast_fp16 = softmax(axis = var_58034, x = aw_5115_cast_fp16)[name = tensor<string, []>("op_60256_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_60257_cast_fp16 = softmax(axis = var_58034, x = aw_5117_cast_fp16)[name = tensor<string, []>("op_60257_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_60258_cast_fp16 = softmax(axis = var_58034, x = aw_5119_cast_fp16)[name = tensor<string, []>("op_60258_cast_fp16")];
            tensor<string, []> var_60260_equation_0 = const()[name = tensor<string, []>("op_60260_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60260_cast_fp16 = einsum(equation = var_60260_equation_0, values = (var_60170_cast_fp16, var_60249_cast_fp16))[name = tensor<string, []>("op_60260_cast_fp16")];
            tensor<string, []> var_60262_equation_0 = const()[name = tensor<string, []>("op_60262_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60262_cast_fp16 = einsum(equation = var_60262_equation_0, values = (var_60174_cast_fp16, var_60250_cast_fp16))[name = tensor<string, []>("op_60262_cast_fp16")];
            tensor<string, []> var_60264_equation_0 = const()[name = tensor<string, []>("op_60264_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60264_cast_fp16 = einsum(equation = var_60264_equation_0, values = (var_60178_cast_fp16, var_60251_cast_fp16))[name = tensor<string, []>("op_60264_cast_fp16")];
            tensor<string, []> var_60266_equation_0 = const()[name = tensor<string, []>("op_60266_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60266_cast_fp16 = einsum(equation = var_60266_equation_0, values = (var_60182_cast_fp16, var_60252_cast_fp16))[name = tensor<string, []>("op_60266_cast_fp16")];
            tensor<string, []> var_60268_equation_0 = const()[name = tensor<string, []>("op_60268_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60268_cast_fp16 = einsum(equation = var_60268_equation_0, values = (var_60186_cast_fp16, var_60253_cast_fp16))[name = tensor<string, []>("op_60268_cast_fp16")];
            tensor<string, []> var_60270_equation_0 = const()[name = tensor<string, []>("op_60270_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60270_cast_fp16 = einsum(equation = var_60270_equation_0, values = (var_60190_cast_fp16, var_60254_cast_fp16))[name = tensor<string, []>("op_60270_cast_fp16")];
            tensor<string, []> var_60272_equation_0 = const()[name = tensor<string, []>("op_60272_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60272_cast_fp16 = einsum(equation = var_60272_equation_0, values = (var_60194_cast_fp16, var_60255_cast_fp16))[name = tensor<string, []>("op_60272_cast_fp16")];
            tensor<string, []> var_60274_equation_0 = const()[name = tensor<string, []>("op_60274_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60274_cast_fp16 = einsum(equation = var_60274_equation_0, values = (var_60198_cast_fp16, var_60256_cast_fp16))[name = tensor<string, []>("op_60274_cast_fp16")];
            tensor<string, []> var_60276_equation_0 = const()[name = tensor<string, []>("op_60276_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60276_cast_fp16 = einsum(equation = var_60276_equation_0, values = (var_60202_cast_fp16, var_60257_cast_fp16))[name = tensor<string, []>("op_60276_cast_fp16")];
            tensor<string, []> var_60278_equation_0 = const()[name = tensor<string, []>("op_60278_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60278_cast_fp16 = einsum(equation = var_60278_equation_0, values = (var_60206_cast_fp16, var_60258_cast_fp16))[name = tensor<string, []>("op_60278_cast_fp16")];
            tensor<bool, []> input_781_interleave_0 = const()[name = tensor<string, []>("input_781_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 640, 1, 4096]> input_781_cast_fp16 = concat(axis = var_58034, interleave = input_781_interleave_0, values = (var_60260_cast_fp16, var_60262_cast_fp16, var_60264_cast_fp16, var_60266_cast_fp16, var_60268_cast_fp16, var_60270_cast_fp16, var_60272_cast_fp16, var_60274_cast_fp16, var_60276_cast_fp16, var_60278_cast_fp16))[name = tensor<string, []>("input_781_cast_fp16")];
            tensor<string, []> var_60288_pad_type_0 = const()[name = tensor<string, []>("op_60288_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_60288_strides_0 = const()[name = tensor<string, []>("op_60288_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_60288_pad_0 = const()[name = tensor<string, []>("op_60288_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_60288_dilations_0 = const()[name = tensor<string, []>("op_60288_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_60288_groups_0 = const()[name = tensor<string, []>("op_60288_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_1_transformer_blocks_1_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(920664000))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(920971264))), name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_1_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            tensor<fp16, [640]> up_blocks_1_attentions_1_transformer_blocks_1_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_1_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(920971456)))];
            tensor<fp16, [2, 640, 1, 4096]> var_60288_cast_fp16 = conv(bias = up_blocks_1_attentions_1_transformer_blocks_1_attn2_to_out_0_bias_to_fp16, dilations = var_60288_dilations_0, groups = var_60288_groups_0, pad = var_60288_pad_0, pad_type = var_60288_pad_type_0, strides = var_60288_strides_0, weight = up_blocks_1_attentions_1_transformer_blocks_1_attn2_to_out_0_weight_to_fp16_palettized, x = input_781_cast_fp16)[name = tensor<string, []>("op_60288_cast_fp16")];
            tensor<fp16, [2, 640, 1, 4096]> inputs_407_cast_fp16 = add(x = var_60288_cast_fp16, y = inputs_405_cast_fp16)[name = tensor<string, []>("inputs_407_cast_fp16")];
            tensor<int32, [1]> input_783_axes_0 = const()[name = tensor<string, []>("input_783_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [640]> input_783_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_783_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(920972800)))];
            tensor<fp16, [640]> input_783_beta_0_to_fp16 = const()[name = tensor<string, []>("input_783_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(920974144)))];
            tensor<fp16, []> var_60298_to_fp16 = const()[name = tensor<string, []>("op_60298_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 640, 1, 4096]> input_783_cast_fp16 = layer_norm(axes = input_783_axes_0, beta = input_783_beta_0_to_fp16, epsilon = var_60298_to_fp16, gamma = input_783_gamma_0_to_fp16, x = inputs_407_cast_fp16)[name = tensor<string, []>("input_783_cast_fp16")];
            tensor<string, []> var_60318_pad_type_0 = const()[name = tensor<string, []>("op_60318_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_60318_strides_0 = const()[name = tensor<string, []>("op_60318_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_60318_pad_0 = const()[name = tensor<string, []>("op_60318_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_60318_dilations_0 = const()[name = tensor<string, []>("op_60318_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_60318_groups_0 = const()[name = tensor<string, []>("op_60318_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [5120, 640, 1, 1]> up_blocks_1_attentions_1_transformer_blocks_1_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2457600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(920975488))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(923433152))), name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_1_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([5120, 640, 1, 1])];
            tensor<fp16, [5120]> up_blocks_1_attentions_1_transformer_blocks_1_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_1_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(923433344)))];
            tensor<fp16, [2, 5120, 1, 4096]> var_60318_cast_fp16 = conv(bias = up_blocks_1_attentions_1_transformer_blocks_1_ff_net_0_proj_bias_to_fp16, dilations = var_60318_dilations_0, groups = var_60318_groups_0, pad = var_60318_pad_0, pad_type = var_60318_pad_type_0, strides = var_60318_strides_0, weight = up_blocks_1_attentions_1_transformer_blocks_1_ff_net_0_proj_weight_to_fp16_palettized, x = input_783_cast_fp16)[name = tensor<string, []>("op_60318_cast_fp16")];
            tensor<int32, [2]> var_60319_split_sizes_0 = const()[name = tensor<string, []>("op_60319_split_sizes_0"), val = tensor<int32, [2]>([2560, 2560])];
            tensor<int32, []> var_60319_axis_0 = const()[name = tensor<string, []>("op_60319_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 2560, 1, 4096]> var_60319_cast_fp16_0, tensor<fp16, [2, 2560, 1, 4096]> var_60319_cast_fp16_1 = split(axis = var_60319_axis_0, split_sizes = var_60319_split_sizes_0, x = var_60318_cast_fp16)[name = tensor<string, []>("op_60319_cast_fp16")];
            tensor<string, []> var_60321_mode_0 = const()[name = tensor<string, []>("op_60321_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 2560, 1, 4096]> var_60321_cast_fp16 = gelu(mode = var_60321_mode_0, x = var_60319_cast_fp16_1)[name = tensor<string, []>("op_60321_cast_fp16")];
            tensor<fp16, [2, 2560, 1, 4096]> input_785_cast_fp16 = mul(x = var_60319_cast_fp16_0, y = var_60321_cast_fp16)[name = tensor<string, []>("input_785_cast_fp16")];
            tensor<string, []> var_60329_pad_type_0 = const()[name = tensor<string, []>("op_60329_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_60329_strides_0 = const()[name = tensor<string, []>("op_60329_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_60329_pad_0 = const()[name = tensor<string, []>("op_60329_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_60329_dilations_0 = const()[name = tensor<string, []>("op_60329_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_60329_groups_0 = const()[name = tensor<string, []>("op_60329_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 2560, 1, 1]> up_blocks_1_attentions_1_transformer_blocks_1_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(923443648))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(924672512))), name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_1_ff_net_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 2560, 1, 1])];
            tensor<fp16, [640]> up_blocks_1_attentions_1_transformer_blocks_1_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_1_ff_net_2_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(924672704)))];
            tensor<fp16, [2, 640, 1, 4096]> var_60329_cast_fp16 = conv(bias = up_blocks_1_attentions_1_transformer_blocks_1_ff_net_2_bias_to_fp16, dilations = var_60329_dilations_0, groups = var_60329_groups_0, pad = var_60329_pad_0, pad_type = var_60329_pad_type_0, strides = var_60329_strides_0, weight = up_blocks_1_attentions_1_transformer_blocks_1_ff_net_2_weight_to_fp16_palettized, x = input_785_cast_fp16)[name = tensor<string, []>("op_60329_cast_fp16")];
            tensor<fp16, [2, 640, 1, 4096]> hidden_states_547_cast_fp16 = add(x = var_60329_cast_fp16, y = inputs_407_cast_fp16)[name = tensor<string, []>("hidden_states_547_cast_fp16")];
            tensor<int32, [4]> var_60331 = const()[name = tensor<string, []>("op_60331"), val = tensor<int32, [4]>([2, 640, 64, 64])];
            tensor<fp16, [2, 640, 64, 64]> input_787_cast_fp16 = reshape(shape = var_60331, x = hidden_states_547_cast_fp16)[name = tensor<string, []>("input_787_cast_fp16")];
            tensor<string, []> hidden_states_549_pad_type_0 = const()[name = tensor<string, []>("hidden_states_549_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> hidden_states_549_strides_0 = const()[name = tensor<string, []>("hidden_states_549_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_549_pad_0 = const()[name = tensor<string, []>("hidden_states_549_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_549_dilations_0 = const()[name = tensor<string, []>("hidden_states_549_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_549_groups_0 = const()[name = tensor<string, []>("hidden_states_549_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_1_proj_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(924674048))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(924981312))), name = tensor<string, []>("up_blocks_1_attentions_1_proj_out_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            tensor<fp16, [640]> 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, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(924981504)))];
            tensor<fp16, [2, 640, 64, 64]> hidden_states_549_cast_fp16 = conv(bias = up_blocks_1_attentions_1_proj_out_bias_to_fp16, dilations = hidden_states_549_dilations_0, groups = hidden_states_549_groups_0, pad = hidden_states_549_pad_0, pad_type = hidden_states_549_pad_type_0, strides = hidden_states_549_strides_0, weight = up_blocks_1_attentions_1_proj_out_weight_to_fp16_palettized, x = input_787_cast_fp16)[name = tensor<string, []>("hidden_states_549_cast_fp16")];
            tensor<fp16, [2, 640, 64, 64]> hidden_states_551_cast_fp16 = add(x = hidden_states_549_cast_fp16, y = hidden_states_531_cast_fp16)[name = tensor<string, []>("hidden_states_551_cast_fp16")];
            tensor<bool, []> input_789_interleave_0 = const()[name = tensor<string, []>("input_789_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 320, 64, 64]> cast_5 = cast(dtype = cast_5_dtype_0, x = input_45_cast_fp16)[name = tensor<string, []>("cast_7")];
            tensor<fp16, [2, 960, 64, 64]> input_789_cast_fp16 = concat(axis = var_58034, interleave = input_789_interleave_0, values = (hidden_states_551_cast_fp16, cast_5))[name = tensor<string, []>("input_789_cast_fp16")];
            tensor<int32, [5]> reshape_144_shape_0 = const()[name = tensor<string, []>("reshape_144_shape_0"), val = tensor<int32, [5]>([2, 32, 30, 64, 64])];
            tensor<fp16, [2, 32, 30, 64, 64]> reshape_144_cast_fp16 = reshape(shape = reshape_144_shape_0, x = input_789_cast_fp16)[name = tensor<string, []>("reshape_144_cast_fp16")];
            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_fp16 = reduce_mean(axes = reduce_mean_108_axes_0, keep_dims = reduce_mean_108_keep_dims_0, x = reshape_144_cast_fp16)[name = tensor<string, []>("reduce_mean_108_cast_fp16")];
            tensor<fp16, [2, 32, 30, 64, 64]> sub_72_cast_fp16 = sub(x = reshape_144_cast_fp16, y = reduce_mean_108_cast_fp16)[name = tensor<string, []>("sub_72_cast_fp16")];
            tensor<fp16, [2, 32, 30, 64, 64]> square_36_cast_fp16 = square(x = sub_72_cast_fp16)[name = tensor<string, []>("square_36_cast_fp16")];
            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_fp16 = reduce_mean(axes = reduce_mean_110_axes_0, keep_dims = reduce_mean_110_keep_dims_0, x = square_36_cast_fp16)[name = tensor<string, []>("reduce_mean_110_cast_fp16")];
            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_fp16 = add(x = reduce_mean_110_cast_fp16, y = add_72_y_0_to_fp16)[name = tensor<string, []>("add_72_cast_fp16")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_36_cast_fp16 = sqrt(x = add_72_cast_fp16)[name = tensor<string, []>("sqrt_36_cast_fp16")];
            tensor<fp16, [2, 32, 30, 64, 64]> real_div_36_cast_fp16 = real_div(x = sub_72_cast_fp16, y = sqrt_36_cast_fp16)[name = tensor<string, []>("real_div_36_cast_fp16")];
            tensor<int32, [4]> reshape_145_shape_0 = const()[name = tensor<string, []>("reshape_145_shape_0"), val = tensor<int32, [4]>([2, 960, 64, 64])];
            tensor<fp16, [2, 960, 64, 64]> reshape_145_cast_fp16 = reshape(shape = reshape_145_shape_0, x = real_div_36_cast_fp16)[name = tensor<string, []>("reshape_145_cast_fp16")];
            tensor<fp16, [960]> add_73_mean_0_to_fp16 = const()[name = tensor<string, []>("add_73_mean_0_to_fp16"), val = tensor<fp16, [960]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(924982848)))];
            tensor<fp16, [960]> add_73_variance_0_to_fp16 = const()[name = tensor<string, []>("add_73_variance_0_to_fp16"), val = tensor<fp16, [960]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(924984832)))];
            tensor<fp16, [960]> add_73_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_73_gamma_0_to_fp16"), val = tensor<fp16, [960]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(924986816)))];
            tensor<fp16, [960]> add_73_beta_0_to_fp16 = const()[name = tensor<string, []>("add_73_beta_0_to_fp16"), val = tensor<fp16, [960]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(924988800)))];
            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, 960, 64, 64]> add_73_cast_fp16 = batch_norm(beta = add_73_beta_0_to_fp16, epsilon = add_73_epsilon_0_to_fp16, gamma = add_73_gamma_0_to_fp16, mean = add_73_mean_0_to_fp16, variance = add_73_variance_0_to_fp16, x = reshape_145_cast_fp16)[name = tensor<string, []>("add_73_cast_fp16")];
            tensor<fp16, [2, 960, 64, 64]> input_793_cast_fp16 = silu(x = add_73_cast_fp16)[name = tensor<string, []>("input_793_cast_fp16")];
            tensor<string, []> hidden_states_553_pad_type_0 = const()[name = tensor<string, []>("hidden_states_553_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_553_pad_0 = const()[name = tensor<string, []>("hidden_states_553_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> hidden_states_553_strides_0 = const()[name = tensor<string, []>("hidden_states_553_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> hidden_states_553_dilations_0 = const()[name = tensor<string, []>("hidden_states_553_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_553_groups_0 = const()[name = tensor<string, []>("hidden_states_553_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 960, 3, 3]> up_blocks_1_resnets_2_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [4147200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(924990784))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(929138048))), name = tensor<string, []>("up_blocks_1_resnets_2_conv1_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 960, 3, 3])];
            tensor<fp16, [640]> 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, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(929138240)))];
            tensor<fp16, [2, 640, 64, 64]> hidden_states_553_cast_fp16 = conv(bias = up_blocks_1_resnets_2_conv1_bias_to_fp16, dilations = hidden_states_553_dilations_0, groups = hidden_states_553_groups_0, pad = hidden_states_553_pad_0, pad_type = hidden_states_553_pad_type_0, strides = hidden_states_553_strides_0, weight = up_blocks_1_resnets_2_conv1_weight_to_fp16_palettized, x = input_793_cast_fp16)[name = tensor<string, []>("hidden_states_553_cast_fp16")];
            tensor<string, []> temb_27_pad_type_0 = const()[name = tensor<string, []>("temb_27_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> temb_27_strides_0 = const()[name = tensor<string, []>("temb_27_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> temb_27_pad_0 = const()[name = tensor<string, []>("temb_27_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> temb_27_dilations_0 = const()[name = tensor<string, []>("temb_27_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> temb_27_groups_0 = const()[name = tensor<string, []>("temb_27_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 1280, 1, 1]> up_blocks_1_resnets_2_time_emb_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [614400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(929139584))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(929754048))), name = tensor<string, []>("up_blocks_1_resnets_2_time_emb_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 1280, 1, 1])];
            tensor<fp16, [640]> 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, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(929754240)))];
            tensor<fp16, [2, 640, 1, 1]> temb_27_cast_fp16 = conv(bias = up_blocks_1_resnets_2_time_emb_proj_bias_to_fp16, dilations = temb_27_dilations_0, groups = temb_27_groups_0, pad = temb_27_pad_0, pad_type = temb_27_pad_type_0, strides = temb_27_strides_0, weight = up_blocks_1_resnets_2_time_emb_proj_weight_to_fp16_palettized, x = cast_2)[name = tensor<string, []>("temb_27_cast_fp16")];
            tensor<fp16, [2, 640, 64, 64]> input_797_cast_fp16 = add(x = hidden_states_553_cast_fp16, y = temb_27_cast_fp16)[name = tensor<string, []>("input_797_cast_fp16")];
            tensor<int32, [5]> reshape_148_shape_0 = const()[name = tensor<string, []>("reshape_148_shape_0"), val = tensor<int32, [5]>([2, 32, 20, 64, 64])];
            tensor<fp16, [2, 32, 20, 64, 64]> reshape_148_cast_fp16 = reshape(shape = reshape_148_shape_0, x = input_797_cast_fp16)[name = tensor<string, []>("reshape_148_cast_fp16")];
            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_fp16 = reduce_mean(axes = reduce_mean_111_axes_0, keep_dims = reduce_mean_111_keep_dims_0, x = reshape_148_cast_fp16)[name = tensor<string, []>("reduce_mean_111_cast_fp16")];
            tensor<fp16, [2, 32, 20, 64, 64]> sub_74_cast_fp16 = sub(x = reshape_148_cast_fp16, y = reduce_mean_111_cast_fp16)[name = tensor<string, []>("sub_74_cast_fp16")];
            tensor<fp16, [2, 32, 20, 64, 64]> square_37_cast_fp16 = square(x = sub_74_cast_fp16)[name = tensor<string, []>("square_37_cast_fp16")];
            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_fp16 = reduce_mean(axes = reduce_mean_113_axes_0, keep_dims = reduce_mean_113_keep_dims_0, x = square_37_cast_fp16)[name = tensor<string, []>("reduce_mean_113_cast_fp16")];
            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_fp16 = add(x = reduce_mean_113_cast_fp16, y = add_74_y_0_to_fp16)[name = tensor<string, []>("add_74_cast_fp16")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_37_cast_fp16 = sqrt(x = add_74_cast_fp16)[name = tensor<string, []>("sqrt_37_cast_fp16")];
            tensor<fp16, [2, 32, 20, 64, 64]> real_div_37_cast_fp16 = real_div(x = sub_74_cast_fp16, y = sqrt_37_cast_fp16)[name = tensor<string, []>("real_div_37_cast_fp16")];
            tensor<int32, [4]> reshape_149_shape_0 = const()[name = tensor<string, []>("reshape_149_shape_0"), val = tensor<int32, [4]>([2, 640, 64, 64])];
            tensor<fp16, [2, 640, 64, 64]> reshape_149_cast_fp16 = reshape(shape = reshape_149_shape_0, x = real_div_37_cast_fp16)[name = tensor<string, []>("reshape_149_cast_fp16")];
            tensor<fp16, [640]> add_75_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_75_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(929755584)))];
            tensor<fp16, [640]> add_75_beta_0_to_fp16 = const()[name = tensor<string, []>("add_75_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(929756928)))];
            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, 640, 64, 64]> add_75_cast_fp16 = 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_11_mean_0_to_fp16, variance = add_11_variance_0_to_fp16, x = reshape_149_cast_fp16)[name = tensor<string, []>("add_75_cast_fp16")];
            tensor<fp16, [2, 640, 64, 64]> input_801_cast_fp16 = silu(x = add_75_cast_fp16)[name = tensor<string, []>("input_801_cast_fp16")];
            tensor<string, []> hidden_states_555_pad_type_0 = const()[name = tensor<string, []>("hidden_states_555_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_555_pad_0 = const()[name = tensor<string, []>("hidden_states_555_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> hidden_states_555_strides_0 = const()[name = tensor<string, []>("hidden_states_555_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> hidden_states_555_dilations_0 = const()[name = tensor<string, []>("hidden_states_555_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_555_groups_0 = const()[name = tensor<string, []>("hidden_states_555_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 640, 3, 3]> up_blocks_1_resnets_2_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2764800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(929758272))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(932523136))), name = tensor<string, []>("up_blocks_1_resnets_2_conv2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 3, 3])];
            tensor<fp16, [640]> 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, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(932523328)))];
            tensor<fp16, [2, 640, 64, 64]> hidden_states_555_cast_fp16 = conv(bias = up_blocks_1_resnets_2_conv2_bias_to_fp16, dilations = hidden_states_555_dilations_0, groups = hidden_states_555_groups_0, pad = hidden_states_555_pad_0, pad_type = hidden_states_555_pad_type_0, strides = hidden_states_555_strides_0, weight = up_blocks_1_resnets_2_conv2_weight_to_fp16_palettized, x = input_801_cast_fp16)[name = tensor<string, []>("hidden_states_555_cast_fp16")];
            tensor<string, []> x_15_pad_type_0 = const()[name = tensor<string, []>("x_15_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> x_15_strides_0 = const()[name = tensor<string, []>("x_15_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> x_15_pad_0 = const()[name = tensor<string, []>("x_15_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> x_15_dilations_0 = const()[name = tensor<string, []>("x_15_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> x_15_groups_0 = const()[name = tensor<string, []>("x_15_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 960, 1, 1]> up_blocks_1_resnets_2_conv_shortcut_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [460800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(932524672))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(932985536))), name = tensor<string, []>("up_blocks_1_resnets_2_conv_shortcut_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 960, 1, 1])];
            tensor<fp16, [640]> 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, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(932985728)))];
            tensor<fp16, [2, 640, 64, 64]> x_15_cast_fp16 = conv(bias = up_blocks_1_resnets_2_conv_shortcut_bias_to_fp16, dilations = x_15_dilations_0, groups = x_15_groups_0, pad = x_15_pad_0, pad_type = x_15_pad_type_0, strides = x_15_strides_0, weight = up_blocks_1_resnets_2_conv_shortcut_weight_to_fp16_palettized, x = input_789_cast_fp16)[name = tensor<string, []>("x_15_cast_fp16")];
            tensor<fp16, [2, 640, 64, 64]> hidden_states_557_cast_fp16 = add(x = x_15_cast_fp16, y = hidden_states_555_cast_fp16)[name = tensor<string, []>("hidden_states_557_cast_fp16")];
            tensor<int32, [5]> reshape_152_shape_0 = const()[name = tensor<string, []>("reshape_152_shape_0"), val = tensor<int32, [5]>([2, 32, 20, 64, 64])];
            tensor<fp16, [2, 32, 20, 64, 64]> reshape_152_cast_fp16 = reshape(shape = reshape_152_shape_0, x = hidden_states_557_cast_fp16)[name = tensor<string, []>("reshape_152_cast_fp16")];
            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_fp16 = reduce_mean(axes = reduce_mean_114_axes_0, keep_dims = reduce_mean_114_keep_dims_0, x = reshape_152_cast_fp16)[name = tensor<string, []>("reduce_mean_114_cast_fp16")];
            tensor<fp16, [2, 32, 20, 64, 64]> sub_76_cast_fp16 = sub(x = reshape_152_cast_fp16, y = reduce_mean_114_cast_fp16)[name = tensor<string, []>("sub_76_cast_fp16")];
            tensor<fp16, [2, 32, 20, 64, 64]> square_38_cast_fp16 = square(x = sub_76_cast_fp16)[name = tensor<string, []>("square_38_cast_fp16")];
            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_fp16 = reduce_mean(axes = reduce_mean_116_axes_0, keep_dims = reduce_mean_116_keep_dims_0, x = square_38_cast_fp16)[name = tensor<string, []>("reduce_mean_116_cast_fp16")];
            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_fp16 = add(x = reduce_mean_116_cast_fp16, y = add_76_y_0_to_fp16)[name = tensor<string, []>("add_76_cast_fp16")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_38_cast_fp16 = sqrt(x = add_76_cast_fp16)[name = tensor<string, []>("sqrt_38_cast_fp16")];
            tensor<fp16, [2, 32, 20, 64, 64]> real_div_38_cast_fp16 = real_div(x = sub_76_cast_fp16, y = sqrt_38_cast_fp16)[name = tensor<string, []>("real_div_38_cast_fp16")];
            tensor<int32, [4]> reshape_153_shape_0 = const()[name = tensor<string, []>("reshape_153_shape_0"), val = tensor<int32, [4]>([2, 640, 64, 64])];
            tensor<fp16, [2, 640, 64, 64]> reshape_153_cast_fp16 = reshape(shape = reshape_153_shape_0, x = real_div_38_cast_fp16)[name = tensor<string, []>("reshape_153_cast_fp16")];
            tensor<fp16, [640]> add_77_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_77_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(932987072)))];
            tensor<fp16, [640]> add_77_beta_0_to_fp16 = const()[name = tensor<string, []>("add_77_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(932988416)))];
            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, 640, 64, 64]> add_77_cast_fp16 = 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_11_mean_0_to_fp16, variance = add_11_variance_0_to_fp16, x = reshape_153_cast_fp16)[name = tensor<string, []>("add_77_cast_fp16")];
            tensor<string, []> hidden_states_559_pad_type_0 = const()[name = tensor<string, []>("hidden_states_559_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> hidden_states_559_strides_0 = const()[name = tensor<string, []>("hidden_states_559_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_559_pad_0 = const()[name = tensor<string, []>("hidden_states_559_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_559_dilations_0 = const()[name = tensor<string, []>("hidden_states_559_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_559_groups_0 = const()[name = tensor<string, []>("hidden_states_559_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_2_proj_in_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(932989760))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(933297024))), name = tensor<string, []>("up_blocks_1_attentions_2_proj_in_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            tensor<fp16, [640]> 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, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(933297216)))];
            tensor<fp16, [2, 640, 64, 64]> hidden_states_559_cast_fp16 = conv(bias = up_blocks_1_attentions_2_proj_in_bias_to_fp16, dilations = hidden_states_559_dilations_0, groups = hidden_states_559_groups_0, pad = hidden_states_559_pad_0, pad_type = hidden_states_559_pad_type_0, strides = hidden_states_559_strides_0, weight = up_blocks_1_attentions_2_proj_in_weight_to_fp16_palettized, x = add_77_cast_fp16)[name = tensor<string, []>("hidden_states_559_cast_fp16")];
            tensor<int32, [4]> var_60413 = const()[name = tensor<string, []>("op_60413"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<fp16, [2, 640, 1, 4096]> inputs_409_cast_fp16 = reshape(shape = var_60413, x = hidden_states_559_cast_fp16)[name = tensor<string, []>("inputs_409_cast_fp16")];
            tensor<int32, [1]> hidden_states_561_axes_0 = const()[name = tensor<string, []>("hidden_states_561_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [640]> hidden_states_561_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_561_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(933298560)))];
            tensor<fp16, [640]> hidden_states_561_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_561_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(933299904)))];
            tensor<fp16, []> var_60429_to_fp16 = const()[name = tensor<string, []>("op_60429_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 640, 1, 4096]> hidden_states_561_cast_fp16 = layer_norm(axes = hidden_states_561_axes_0, beta = hidden_states_561_beta_0_to_fp16, epsilon = var_60429_to_fp16, gamma = hidden_states_561_gamma_0_to_fp16, x = inputs_409_cast_fp16)[name = tensor<string, []>("hidden_states_561_cast_fp16")];
            tensor<string, []> q_273_pad_type_0 = const()[name = tensor<string, []>("q_273_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_273_strides_0 = const()[name = tensor<string, []>("q_273_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_273_pad_0 = const()[name = tensor<string, []>("q_273_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_273_dilations_0 = const()[name = tensor<string, []>("q_273_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_273_groups_0 = const()[name = tensor<string, []>("q_273_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(933301248))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(933608512))), name = tensor<string, []>("up_blocks_1_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, 4096]> q_273_cast_fp16 = conv(dilations = q_273_dilations_0, groups = q_273_groups_0, pad = q_273_pad_0, pad_type = q_273_pad_type_0, strides = q_273_strides_0, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_561_cast_fp16)[name = tensor<string, []>("q_273_cast_fp16")];
            tensor<string, []> k_545_pad_type_0 = const()[name = tensor<string, []>("k_545_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_545_strides_0 = const()[name = tensor<string, []>("k_545_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_545_pad_0 = const()[name = tensor<string, []>("k_545_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_545_dilations_0 = const()[name = tensor<string, []>("k_545_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_545_groups_0 = const()[name = tensor<string, []>("k_545_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(933608704))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(933915968))), name = tensor<string, []>("up_blocks_1_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, 4096]> k_545_cast_fp16 = conv(dilations = k_545_dilations_0, groups = k_545_groups_0, pad = k_545_pad_0, pad_type = k_545_pad_type_0, strides = k_545_strides_0, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_561_cast_fp16)[name = tensor<string, []>("k_545_cast_fp16")];
            tensor<string, []> v_273_pad_type_0 = const()[name = tensor<string, []>("v_273_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_273_strides_0 = const()[name = tensor<string, []>("v_273_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_273_pad_0 = const()[name = tensor<string, []>("v_273_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_273_dilations_0 = const()[name = tensor<string, []>("v_273_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_273_groups_0 = const()[name = tensor<string, []>("v_273_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(933916160))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(934223424))), name = tensor<string, []>("up_blocks_1_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, 4096]> v_273_cast_fp16 = conv(dilations = v_273_dilations_0, groups = v_273_groups_0, pad = v_273_pad_0, pad_type = v_273_pad_type_0, strides = v_273_strides_0, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_561_cast_fp16)[name = tensor<string, []>("v_273_cast_fp16")];
            tensor<int32, [4]> var_60462_begin_0 = const()[name = tensor<string, []>("op_60462_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_60462_end_0 = const()[name = tensor<string, []>("op_60462_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_60462_end_mask_0 = const()[name = tensor<string, []>("op_60462_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60462_cast_fp16 = slice_by_index(begin = var_60462_begin_0, end = var_60462_end_0, end_mask = var_60462_end_mask_0, x = q_273_cast_fp16)[name = tensor<string, []>("op_60462_cast_fp16")];
            tensor<int32, [4]> var_60466_begin_0 = const()[name = tensor<string, []>("op_60466_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_60466_end_0 = const()[name = tensor<string, []>("op_60466_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_60466_end_mask_0 = const()[name = tensor<string, []>("op_60466_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60466_cast_fp16 = slice_by_index(begin = var_60466_begin_0, end = var_60466_end_0, end_mask = var_60466_end_mask_0, x = q_273_cast_fp16)[name = tensor<string, []>("op_60466_cast_fp16")];
            tensor<int32, [4]> var_60470_begin_0 = const()[name = tensor<string, []>("op_60470_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_60470_end_0 = const()[name = tensor<string, []>("op_60470_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_60470_end_mask_0 = const()[name = tensor<string, []>("op_60470_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60470_cast_fp16 = slice_by_index(begin = var_60470_begin_0, end = var_60470_end_0, end_mask = var_60470_end_mask_0, x = q_273_cast_fp16)[name = tensor<string, []>("op_60470_cast_fp16")];
            tensor<int32, [4]> var_60474_begin_0 = const()[name = tensor<string, []>("op_60474_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_60474_end_0 = const()[name = tensor<string, []>("op_60474_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_60474_end_mask_0 = const()[name = tensor<string, []>("op_60474_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60474_cast_fp16 = slice_by_index(begin = var_60474_begin_0, end = var_60474_end_0, end_mask = var_60474_end_mask_0, x = q_273_cast_fp16)[name = tensor<string, []>("op_60474_cast_fp16")];
            tensor<int32, [4]> var_60478_begin_0 = const()[name = tensor<string, []>("op_60478_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_60478_end_0 = const()[name = tensor<string, []>("op_60478_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_60478_end_mask_0 = const()[name = tensor<string, []>("op_60478_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60478_cast_fp16 = slice_by_index(begin = var_60478_begin_0, end = var_60478_end_0, end_mask = var_60478_end_mask_0, x = q_273_cast_fp16)[name = tensor<string, []>("op_60478_cast_fp16")];
            tensor<int32, [4]> var_60482_begin_0 = const()[name = tensor<string, []>("op_60482_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_60482_end_0 = const()[name = tensor<string, []>("op_60482_end_0"), val = tensor<int32, [4]>([2, 384, 1, 4096])];
            tensor<bool, [4]> var_60482_end_mask_0 = const()[name = tensor<string, []>("op_60482_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60482_cast_fp16 = slice_by_index(begin = var_60482_begin_0, end = var_60482_end_0, end_mask = var_60482_end_mask_0, x = q_273_cast_fp16)[name = tensor<string, []>("op_60482_cast_fp16")];
            tensor<int32, [4]> var_60486_begin_0 = const()[name = tensor<string, []>("op_60486_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_60486_end_0 = const()[name = tensor<string, []>("op_60486_end_0"), val = tensor<int32, [4]>([2, 448, 1, 4096])];
            tensor<bool, [4]> var_60486_end_mask_0 = const()[name = tensor<string, []>("op_60486_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60486_cast_fp16 = slice_by_index(begin = var_60486_begin_0, end = var_60486_end_0, end_mask = var_60486_end_mask_0, x = q_273_cast_fp16)[name = tensor<string, []>("op_60486_cast_fp16")];
            tensor<int32, [4]> var_60490_begin_0 = const()[name = tensor<string, []>("op_60490_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_60490_end_0 = const()[name = tensor<string, []>("op_60490_end_0"), val = tensor<int32, [4]>([2, 512, 1, 4096])];
            tensor<bool, [4]> var_60490_end_mask_0 = const()[name = tensor<string, []>("op_60490_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60490_cast_fp16 = slice_by_index(begin = var_60490_begin_0, end = var_60490_end_0, end_mask = var_60490_end_mask_0, x = q_273_cast_fp16)[name = tensor<string, []>("op_60490_cast_fp16")];
            tensor<int32, [4]> var_60494_begin_0 = const()[name = tensor<string, []>("op_60494_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_60494_end_0 = const()[name = tensor<string, []>("op_60494_end_0"), val = tensor<int32, [4]>([2, 576, 1, 4096])];
            tensor<bool, [4]> var_60494_end_mask_0 = const()[name = tensor<string, []>("op_60494_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60494_cast_fp16 = slice_by_index(begin = var_60494_begin_0, end = var_60494_end_0, end_mask = var_60494_end_mask_0, x = q_273_cast_fp16)[name = tensor<string, []>("op_60494_cast_fp16")];
            tensor<int32, [4]> var_60498_begin_0 = const()[name = tensor<string, []>("op_60498_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_60498_end_0 = const()[name = tensor<string, []>("op_60498_end_0"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<bool, [4]> var_60498_end_mask_0 = const()[name = tensor<string, []>("op_60498_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60498_cast_fp16 = slice_by_index(begin = var_60498_begin_0, end = var_60498_end_0, end_mask = var_60498_end_mask_0, x = q_273_cast_fp16)[name = tensor<string, []>("op_60498_cast_fp16")];
            tensor<int32, [4]> k_547_perm_0 = const()[name = tensor<string, []>("k_547_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_60505_begin_0 = const()[name = tensor<string, []>("op_60505_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_60505_end_0 = const()[name = tensor<string, []>("op_60505_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 64])];
            tensor<bool, [4]> var_60505_end_mask_0 = const()[name = tensor<string, []>("op_60505_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 640]> k_547_cast_fp16 = transpose(perm = k_547_perm_0, x = k_545_cast_fp16)[name = tensor<string, []>("transpose_3")];
            tensor<fp16, [2, 4096, 1, 64]> var_60505_cast_fp16 = slice_by_index(begin = var_60505_begin_0, end = var_60505_end_0, end_mask = var_60505_end_mask_0, x = k_547_cast_fp16)[name = tensor<string, []>("op_60505_cast_fp16")];
            tensor<int32, [4]> var_60509_begin_0 = const()[name = tensor<string, []>("op_60509_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_60509_end_0 = const()[name = tensor<string, []>("op_60509_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 128])];
            tensor<bool, [4]> var_60509_end_mask_0 = const()[name = tensor<string, []>("op_60509_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_60509_cast_fp16 = slice_by_index(begin = var_60509_begin_0, end = var_60509_end_0, end_mask = var_60509_end_mask_0, x = k_547_cast_fp16)[name = tensor<string, []>("op_60509_cast_fp16")];
            tensor<int32, [4]> var_60513_begin_0 = const()[name = tensor<string, []>("op_60513_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_60513_end_0 = const()[name = tensor<string, []>("op_60513_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 192])];
            tensor<bool, [4]> var_60513_end_mask_0 = const()[name = tensor<string, []>("op_60513_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_60513_cast_fp16 = slice_by_index(begin = var_60513_begin_0, end = var_60513_end_0, end_mask = var_60513_end_mask_0, x = k_547_cast_fp16)[name = tensor<string, []>("op_60513_cast_fp16")];
            tensor<int32, [4]> var_60517_begin_0 = const()[name = tensor<string, []>("op_60517_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_60517_end_0 = const()[name = tensor<string, []>("op_60517_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 256])];
            tensor<bool, [4]> var_60517_end_mask_0 = const()[name = tensor<string, []>("op_60517_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_60517_cast_fp16 = slice_by_index(begin = var_60517_begin_0, end = var_60517_end_0, end_mask = var_60517_end_mask_0, x = k_547_cast_fp16)[name = tensor<string, []>("op_60517_cast_fp16")];
            tensor<int32, [4]> var_60521_begin_0 = const()[name = tensor<string, []>("op_60521_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_60521_end_0 = const()[name = tensor<string, []>("op_60521_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 320])];
            tensor<bool, [4]> var_60521_end_mask_0 = const()[name = tensor<string, []>("op_60521_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_60521_cast_fp16 = slice_by_index(begin = var_60521_begin_0, end = var_60521_end_0, end_mask = var_60521_end_mask_0, x = k_547_cast_fp16)[name = tensor<string, []>("op_60521_cast_fp16")];
            tensor<int32, [4]> var_60525_begin_0 = const()[name = tensor<string, []>("op_60525_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_60525_end_0 = const()[name = tensor<string, []>("op_60525_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 384])];
            tensor<bool, [4]> var_60525_end_mask_0 = const()[name = tensor<string, []>("op_60525_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_60525_cast_fp16 = slice_by_index(begin = var_60525_begin_0, end = var_60525_end_0, end_mask = var_60525_end_mask_0, x = k_547_cast_fp16)[name = tensor<string, []>("op_60525_cast_fp16")];
            tensor<int32, [4]> var_60529_begin_0 = const()[name = tensor<string, []>("op_60529_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_60529_end_0 = const()[name = tensor<string, []>("op_60529_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 448])];
            tensor<bool, [4]> var_60529_end_mask_0 = const()[name = tensor<string, []>("op_60529_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_60529_cast_fp16 = slice_by_index(begin = var_60529_begin_0, end = var_60529_end_0, end_mask = var_60529_end_mask_0, x = k_547_cast_fp16)[name = tensor<string, []>("op_60529_cast_fp16")];
            tensor<int32, [4]> var_60533_begin_0 = const()[name = tensor<string, []>("op_60533_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_60533_end_0 = const()[name = tensor<string, []>("op_60533_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 512])];
            tensor<bool, [4]> var_60533_end_mask_0 = const()[name = tensor<string, []>("op_60533_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_60533_cast_fp16 = slice_by_index(begin = var_60533_begin_0, end = var_60533_end_0, end_mask = var_60533_end_mask_0, x = k_547_cast_fp16)[name = tensor<string, []>("op_60533_cast_fp16")];
            tensor<int32, [4]> var_60537_begin_0 = const()[name = tensor<string, []>("op_60537_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_60537_end_0 = const()[name = tensor<string, []>("op_60537_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 576])];
            tensor<bool, [4]> var_60537_end_mask_0 = const()[name = tensor<string, []>("op_60537_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_60537_cast_fp16 = slice_by_index(begin = var_60537_begin_0, end = var_60537_end_0, end_mask = var_60537_end_mask_0, x = k_547_cast_fp16)[name = tensor<string, []>("op_60537_cast_fp16")];
            tensor<int32, [4]> var_60541_begin_0 = const()[name = tensor<string, []>("op_60541_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_60541_end_0 = const()[name = tensor<string, []>("op_60541_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 640])];
            tensor<bool, [4]> var_60541_end_mask_0 = const()[name = tensor<string, []>("op_60541_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_60541_cast_fp16 = slice_by_index(begin = var_60541_begin_0, end = var_60541_end_0, end_mask = var_60541_end_mask_0, x = k_547_cast_fp16)[name = tensor<string, []>("op_60541_cast_fp16")];
            tensor<int32, [4]> var_60543_begin_0 = const()[name = tensor<string, []>("op_60543_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_60543_end_0 = const()[name = tensor<string, []>("op_60543_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_60543_end_mask_0 = const()[name = tensor<string, []>("op_60543_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60543_cast_fp16 = slice_by_index(begin = var_60543_begin_0, end = var_60543_end_0, end_mask = var_60543_end_mask_0, x = v_273_cast_fp16)[name = tensor<string, []>("op_60543_cast_fp16")];
            tensor<int32, [4]> var_60547_begin_0 = const()[name = tensor<string, []>("op_60547_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_60547_end_0 = const()[name = tensor<string, []>("op_60547_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_60547_end_mask_0 = const()[name = tensor<string, []>("op_60547_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60547_cast_fp16 = slice_by_index(begin = var_60547_begin_0, end = var_60547_end_0, end_mask = var_60547_end_mask_0, x = v_273_cast_fp16)[name = tensor<string, []>("op_60547_cast_fp16")];
            tensor<int32, [4]> var_60551_begin_0 = const()[name = tensor<string, []>("op_60551_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_60551_end_0 = const()[name = tensor<string, []>("op_60551_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_60551_end_mask_0 = const()[name = tensor<string, []>("op_60551_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60551_cast_fp16 = slice_by_index(begin = var_60551_begin_0, end = var_60551_end_0, end_mask = var_60551_end_mask_0, x = v_273_cast_fp16)[name = tensor<string, []>("op_60551_cast_fp16")];
            tensor<int32, [4]> var_60555_begin_0 = const()[name = tensor<string, []>("op_60555_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_60555_end_0 = const()[name = tensor<string, []>("op_60555_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_60555_end_mask_0 = const()[name = tensor<string, []>("op_60555_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60555_cast_fp16 = slice_by_index(begin = var_60555_begin_0, end = var_60555_end_0, end_mask = var_60555_end_mask_0, x = v_273_cast_fp16)[name = tensor<string, []>("op_60555_cast_fp16")];
            tensor<int32, [4]> var_60559_begin_0 = const()[name = tensor<string, []>("op_60559_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_60559_end_0 = const()[name = tensor<string, []>("op_60559_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_60559_end_mask_0 = const()[name = tensor<string, []>("op_60559_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60559_cast_fp16 = slice_by_index(begin = var_60559_begin_0, end = var_60559_end_0, end_mask = var_60559_end_mask_0, x = v_273_cast_fp16)[name = tensor<string, []>("op_60559_cast_fp16")];
            tensor<int32, [4]> var_60563_begin_0 = const()[name = tensor<string, []>("op_60563_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_60563_end_0 = const()[name = tensor<string, []>("op_60563_end_0"), val = tensor<int32, [4]>([2, 384, 1, 4096])];
            tensor<bool, [4]> var_60563_end_mask_0 = const()[name = tensor<string, []>("op_60563_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60563_cast_fp16 = slice_by_index(begin = var_60563_begin_0, end = var_60563_end_0, end_mask = var_60563_end_mask_0, x = v_273_cast_fp16)[name = tensor<string, []>("op_60563_cast_fp16")];
            tensor<int32, [4]> var_60567_begin_0 = const()[name = tensor<string, []>("op_60567_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_60567_end_0 = const()[name = tensor<string, []>("op_60567_end_0"), val = tensor<int32, [4]>([2, 448, 1, 4096])];
            tensor<bool, [4]> var_60567_end_mask_0 = const()[name = tensor<string, []>("op_60567_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60567_cast_fp16 = slice_by_index(begin = var_60567_begin_0, end = var_60567_end_0, end_mask = var_60567_end_mask_0, x = v_273_cast_fp16)[name = tensor<string, []>("op_60567_cast_fp16")];
            tensor<int32, [4]> var_60571_begin_0 = const()[name = tensor<string, []>("op_60571_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_60571_end_0 = const()[name = tensor<string, []>("op_60571_end_0"), val = tensor<int32, [4]>([2, 512, 1, 4096])];
            tensor<bool, [4]> var_60571_end_mask_0 = const()[name = tensor<string, []>("op_60571_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60571_cast_fp16 = slice_by_index(begin = var_60571_begin_0, end = var_60571_end_0, end_mask = var_60571_end_mask_0, x = v_273_cast_fp16)[name = tensor<string, []>("op_60571_cast_fp16")];
            tensor<int32, [4]> var_60575_begin_0 = const()[name = tensor<string, []>("op_60575_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_60575_end_0 = const()[name = tensor<string, []>("op_60575_end_0"), val = tensor<int32, [4]>([2, 576, 1, 4096])];
            tensor<bool, [4]> var_60575_end_mask_0 = const()[name = tensor<string, []>("op_60575_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60575_cast_fp16 = slice_by_index(begin = var_60575_begin_0, end = var_60575_end_0, end_mask = var_60575_end_mask_0, x = v_273_cast_fp16)[name = tensor<string, []>("op_60575_cast_fp16")];
            tensor<int32, [4]> var_60579_begin_0 = const()[name = tensor<string, []>("op_60579_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_60579_end_0 = const()[name = tensor<string, []>("op_60579_end_0"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<bool, [4]> var_60579_end_mask_0 = const()[name = tensor<string, []>("op_60579_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60579_cast_fp16 = slice_by_index(begin = var_60579_begin_0, end = var_60579_end_0, end_mask = var_60579_end_mask_0, x = v_273_cast_fp16)[name = tensor<string, []>("op_60579_cast_fp16")];
            tensor<string, []> var_60583_equation_0 = const()[name = tensor<string, []>("op_60583_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60583_cast_fp16 = einsum(equation = var_60583_equation_0, values = (var_60505_cast_fp16, var_60462_cast_fp16))[name = tensor<string, []>("op_60583_cast_fp16")];
            tensor<fp16, []> var_60584_to_fp16 = const()[name = tensor<string, []>("op_60584_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5121_cast_fp16 = mul(x = var_60583_cast_fp16, y = var_60584_to_fp16)[name = tensor<string, []>("aw_5121_cast_fp16")];
            tensor<string, []> var_60587_equation_0 = const()[name = tensor<string, []>("op_60587_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60587_cast_fp16 = einsum(equation = var_60587_equation_0, values = (var_60509_cast_fp16, var_60466_cast_fp16))[name = tensor<string, []>("op_60587_cast_fp16")];
            tensor<fp16, []> var_60588_to_fp16 = const()[name = tensor<string, []>("op_60588_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5123_cast_fp16 = mul(x = var_60587_cast_fp16, y = var_60588_to_fp16)[name = tensor<string, []>("aw_5123_cast_fp16")];
            tensor<string, []> var_60591_equation_0 = const()[name = tensor<string, []>("op_60591_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60591_cast_fp16 = einsum(equation = var_60591_equation_0, values = (var_60513_cast_fp16, var_60470_cast_fp16))[name = tensor<string, []>("op_60591_cast_fp16")];
            tensor<fp16, []> var_60592_to_fp16 = const()[name = tensor<string, []>("op_60592_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5125_cast_fp16 = mul(x = var_60591_cast_fp16, y = var_60592_to_fp16)[name = tensor<string, []>("aw_5125_cast_fp16")];
            tensor<string, []> var_60595_equation_0 = const()[name = tensor<string, []>("op_60595_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60595_cast_fp16 = einsum(equation = var_60595_equation_0, values = (var_60517_cast_fp16, var_60474_cast_fp16))[name = tensor<string, []>("op_60595_cast_fp16")];
            tensor<fp16, []> var_60596_to_fp16 = const()[name = tensor<string, []>("op_60596_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5127_cast_fp16 = mul(x = var_60595_cast_fp16, y = var_60596_to_fp16)[name = tensor<string, []>("aw_5127_cast_fp16")];
            tensor<string, []> var_60599_equation_0 = const()[name = tensor<string, []>("op_60599_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60599_cast_fp16 = einsum(equation = var_60599_equation_0, values = (var_60521_cast_fp16, var_60478_cast_fp16))[name = tensor<string, []>("op_60599_cast_fp16")];
            tensor<fp16, []> var_60600_to_fp16 = const()[name = tensor<string, []>("op_60600_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5129_cast_fp16 = mul(x = var_60599_cast_fp16, y = var_60600_to_fp16)[name = tensor<string, []>("aw_5129_cast_fp16")];
            tensor<string, []> var_60603_equation_0 = const()[name = tensor<string, []>("op_60603_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60603_cast_fp16 = einsum(equation = var_60603_equation_0, values = (var_60525_cast_fp16, var_60482_cast_fp16))[name = tensor<string, []>("op_60603_cast_fp16")];
            tensor<fp16, []> var_60604_to_fp16 = const()[name = tensor<string, []>("op_60604_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5131_cast_fp16 = mul(x = var_60603_cast_fp16, y = var_60604_to_fp16)[name = tensor<string, []>("aw_5131_cast_fp16")];
            tensor<string, []> var_60607_equation_0 = const()[name = tensor<string, []>("op_60607_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60607_cast_fp16 = einsum(equation = var_60607_equation_0, values = (var_60529_cast_fp16, var_60486_cast_fp16))[name = tensor<string, []>("op_60607_cast_fp16")];
            tensor<fp16, []> var_60608_to_fp16 = const()[name = tensor<string, []>("op_60608_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5133_cast_fp16 = mul(x = var_60607_cast_fp16, y = var_60608_to_fp16)[name = tensor<string, []>("aw_5133_cast_fp16")];
            tensor<string, []> var_60611_equation_0 = const()[name = tensor<string, []>("op_60611_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60611_cast_fp16 = einsum(equation = var_60611_equation_0, values = (var_60533_cast_fp16, var_60490_cast_fp16))[name = tensor<string, []>("op_60611_cast_fp16")];
            tensor<fp16, []> var_60612_to_fp16 = const()[name = tensor<string, []>("op_60612_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5135_cast_fp16 = mul(x = var_60611_cast_fp16, y = var_60612_to_fp16)[name = tensor<string, []>("aw_5135_cast_fp16")];
            tensor<string, []> var_60615_equation_0 = const()[name = tensor<string, []>("op_60615_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60615_cast_fp16 = einsum(equation = var_60615_equation_0, values = (var_60537_cast_fp16, var_60494_cast_fp16))[name = tensor<string, []>("op_60615_cast_fp16")];
            tensor<fp16, []> var_60616_to_fp16 = const()[name = tensor<string, []>("op_60616_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5137_cast_fp16 = mul(x = var_60615_cast_fp16, y = var_60616_to_fp16)[name = tensor<string, []>("aw_5137_cast_fp16")];
            tensor<string, []> var_60619_equation_0 = const()[name = tensor<string, []>("op_60619_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60619_cast_fp16 = einsum(equation = var_60619_equation_0, values = (var_60541_cast_fp16, var_60498_cast_fp16))[name = tensor<string, []>("op_60619_cast_fp16")];
            tensor<fp16, []> var_60620_to_fp16 = const()[name = tensor<string, []>("op_60620_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5139_cast_fp16 = mul(x = var_60619_cast_fp16, y = var_60620_to_fp16)[name = tensor<string, []>("aw_5139_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60622_cast_fp16 = softmax(axis = var_58034, x = aw_5121_cast_fp16)[name = tensor<string, []>("op_60622_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60623_cast_fp16 = softmax(axis = var_58034, x = aw_5123_cast_fp16)[name = tensor<string, []>("op_60623_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60624_cast_fp16 = softmax(axis = var_58034, x = aw_5125_cast_fp16)[name = tensor<string, []>("op_60624_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60625_cast_fp16 = softmax(axis = var_58034, x = aw_5127_cast_fp16)[name = tensor<string, []>("op_60625_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60626_cast_fp16 = softmax(axis = var_58034, x = aw_5129_cast_fp16)[name = tensor<string, []>("op_60626_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60627_cast_fp16 = softmax(axis = var_58034, x = aw_5131_cast_fp16)[name = tensor<string, []>("op_60627_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60628_cast_fp16 = softmax(axis = var_58034, x = aw_5133_cast_fp16)[name = tensor<string, []>("op_60628_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60629_cast_fp16 = softmax(axis = var_58034, x = aw_5135_cast_fp16)[name = tensor<string, []>("op_60629_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60630_cast_fp16 = softmax(axis = var_58034, x = aw_5137_cast_fp16)[name = tensor<string, []>("op_60630_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60631_cast_fp16 = softmax(axis = var_58034, x = aw_5139_cast_fp16)[name = tensor<string, []>("op_60631_cast_fp16")];
            tensor<string, []> var_60633_equation_0 = const()[name = tensor<string, []>("op_60633_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60633_cast_fp16 = einsum(equation = var_60633_equation_0, values = (var_60543_cast_fp16, var_60622_cast_fp16))[name = tensor<string, []>("op_60633_cast_fp16")];
            tensor<string, []> var_60635_equation_0 = const()[name = tensor<string, []>("op_60635_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60635_cast_fp16 = einsum(equation = var_60635_equation_0, values = (var_60547_cast_fp16, var_60623_cast_fp16))[name = tensor<string, []>("op_60635_cast_fp16")];
            tensor<string, []> var_60637_equation_0 = const()[name = tensor<string, []>("op_60637_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60637_cast_fp16 = einsum(equation = var_60637_equation_0, values = (var_60551_cast_fp16, var_60624_cast_fp16))[name = tensor<string, []>("op_60637_cast_fp16")];
            tensor<string, []> var_60639_equation_0 = const()[name = tensor<string, []>("op_60639_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60639_cast_fp16 = einsum(equation = var_60639_equation_0, values = (var_60555_cast_fp16, var_60625_cast_fp16))[name = tensor<string, []>("op_60639_cast_fp16")];
            tensor<string, []> var_60641_equation_0 = const()[name = tensor<string, []>("op_60641_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60641_cast_fp16 = einsum(equation = var_60641_equation_0, values = (var_60559_cast_fp16, var_60626_cast_fp16))[name = tensor<string, []>("op_60641_cast_fp16")];
            tensor<string, []> var_60643_equation_0 = const()[name = tensor<string, []>("op_60643_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60643_cast_fp16 = einsum(equation = var_60643_equation_0, values = (var_60563_cast_fp16, var_60627_cast_fp16))[name = tensor<string, []>("op_60643_cast_fp16")];
            tensor<string, []> var_60645_equation_0 = const()[name = tensor<string, []>("op_60645_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60645_cast_fp16 = einsum(equation = var_60645_equation_0, values = (var_60567_cast_fp16, var_60628_cast_fp16))[name = tensor<string, []>("op_60645_cast_fp16")];
            tensor<string, []> var_60647_equation_0 = const()[name = tensor<string, []>("op_60647_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60647_cast_fp16 = einsum(equation = var_60647_equation_0, values = (var_60571_cast_fp16, var_60629_cast_fp16))[name = tensor<string, []>("op_60647_cast_fp16")];
            tensor<string, []> var_60649_equation_0 = const()[name = tensor<string, []>("op_60649_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60649_cast_fp16 = einsum(equation = var_60649_equation_0, values = (var_60575_cast_fp16, var_60630_cast_fp16))[name = tensor<string, []>("op_60649_cast_fp16")];
            tensor<string, []> var_60651_equation_0 = const()[name = tensor<string, []>("op_60651_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60651_cast_fp16 = einsum(equation = var_60651_equation_0, values = (var_60579_cast_fp16, var_60631_cast_fp16))[name = tensor<string, []>("op_60651_cast_fp16")];
            tensor<bool, []> input_805_interleave_0 = const()[name = tensor<string, []>("input_805_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 640, 1, 4096]> input_805_cast_fp16 = concat(axis = var_58034, interleave = input_805_interleave_0, values = (var_60633_cast_fp16, var_60635_cast_fp16, var_60637_cast_fp16, var_60639_cast_fp16, var_60641_cast_fp16, var_60643_cast_fp16, var_60645_cast_fp16, var_60647_cast_fp16, var_60649_cast_fp16, var_60651_cast_fp16))[name = tensor<string, []>("input_805_cast_fp16")];
            tensor<string, []> var_60661_pad_type_0 = const()[name = tensor<string, []>("op_60661_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_60661_strides_0 = const()[name = tensor<string, []>("op_60661_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_60661_pad_0 = const()[name = tensor<string, []>("op_60661_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_60661_dilations_0 = const()[name = tensor<string, []>("op_60661_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_60661_groups_0 = const()[name = tensor<string, []>("op_60661_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 640, 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, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(934223616))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(934530880))), name = tensor<string, []>("up_blocks_1_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_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, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(934531072)))];
            tensor<fp16, [2, 640, 1, 4096]> var_60661_cast_fp16 = conv(bias = up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_60661_dilations_0, groups = var_60661_groups_0, pad = var_60661_pad_0, pad_type = var_60661_pad_type_0, strides = var_60661_strides_0, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized, x = input_805_cast_fp16)[name = tensor<string, []>("op_60661_cast_fp16")];
            tensor<fp16, [2, 640, 1, 4096]> inputs_411_cast_fp16 = add(x = var_60661_cast_fp16, y = inputs_409_cast_fp16)[name = tensor<string, []>("inputs_411_cast_fp16")];
            tensor<int32, [1]> hidden_states_563_axes_0 = const()[name = tensor<string, []>("hidden_states_563_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [640]> hidden_states_563_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_563_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(934532416)))];
            tensor<fp16, [640]> hidden_states_563_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_563_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(934533760)))];
            tensor<fp16, []> var_60671_to_fp16 = const()[name = tensor<string, []>("op_60671_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 640, 1, 4096]> hidden_states_563_cast_fp16 = layer_norm(axes = hidden_states_563_axes_0, beta = hidden_states_563_beta_0_to_fp16, epsilon = var_60671_to_fp16, gamma = hidden_states_563_gamma_0_to_fp16, x = inputs_411_cast_fp16)[name = tensor<string, []>("hidden_states_563_cast_fp16")];
            tensor<string, []> q_275_pad_type_0 = const()[name = tensor<string, []>("q_275_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_275_strides_0 = const()[name = tensor<string, []>("q_275_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_275_pad_0 = const()[name = tensor<string, []>("q_275_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_275_dilations_0 = const()[name = tensor<string, []>("q_275_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_275_groups_0 = const()[name = tensor<string, []>("q_275_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(934535104))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(934842368))), name = tensor<string, []>("up_blocks_1_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, 4096]> q_275_cast_fp16 = conv(dilations = q_275_dilations_0, groups = q_275_groups_0, pad = q_275_pad_0, pad_type = q_275_pad_type_0, strides = q_275_strides_0, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_563_cast_fp16)[name = tensor<string, []>("q_275_cast_fp16")];
            tensor<string, []> k_549_pad_type_0 = const()[name = tensor<string, []>("k_549_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_549_strides_0 = const()[name = tensor<string, []>("k_549_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_549_pad_0 = const()[name = tensor<string, []>("k_549_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_549_dilations_0 = const()[name = tensor<string, []>("k_549_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_549_groups_0 = const()[name = tensor<string, []>("k_549_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 2048, 1, 1]> up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [983040]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(934842560))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(935825664))), name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 2048, 1, 1])];
            tensor<fp16, [2, 640, 1, 77]> k_549_cast_fp16 = conv(dilations = k_549_dilations_0, groups = k_549_groups_0, pad = k_549_pad_0, pad_type = k_549_pad_type_0, strides = k_549_strides_0, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("k_549_cast_fp16")];
            tensor<string, []> v_275_pad_type_0 = const()[name = tensor<string, []>("v_275_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_275_strides_0 = const()[name = tensor<string, []>("v_275_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_275_pad_0 = const()[name = tensor<string, []>("v_275_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_275_dilations_0 = const()[name = tensor<string, []>("v_275_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_275_groups_0 = const()[name = tensor<string, []>("v_275_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 2048, 1, 1]> up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [983040]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(935825856))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(936808960))), name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 2048, 1, 1])];
            tensor<fp16, [2, 640, 1, 77]> v_275_cast_fp16 = conv(dilations = v_275_dilations_0, groups = v_275_groups_0, pad = v_275_pad_0, pad_type = v_275_pad_type_0, strides = v_275_strides_0, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("v_275_cast_fp16")];
            tensor<int32, [4]> var_60704_begin_0 = const()[name = tensor<string, []>("op_60704_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_60704_end_0 = const()[name = tensor<string, []>("op_60704_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_60704_end_mask_0 = const()[name = tensor<string, []>("op_60704_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60704_cast_fp16 = slice_by_index(begin = var_60704_begin_0, end = var_60704_end_0, end_mask = var_60704_end_mask_0, x = q_275_cast_fp16)[name = tensor<string, []>("op_60704_cast_fp16")];
            tensor<int32, [4]> var_60708_begin_0 = const()[name = tensor<string, []>("op_60708_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_60708_end_0 = const()[name = tensor<string, []>("op_60708_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_60708_end_mask_0 = const()[name = tensor<string, []>("op_60708_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60708_cast_fp16 = slice_by_index(begin = var_60708_begin_0, end = var_60708_end_0, end_mask = var_60708_end_mask_0, x = q_275_cast_fp16)[name = tensor<string, []>("op_60708_cast_fp16")];
            tensor<int32, [4]> var_60712_begin_0 = const()[name = tensor<string, []>("op_60712_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_60712_end_0 = const()[name = tensor<string, []>("op_60712_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_60712_end_mask_0 = const()[name = tensor<string, []>("op_60712_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60712_cast_fp16 = slice_by_index(begin = var_60712_begin_0, end = var_60712_end_0, end_mask = var_60712_end_mask_0, x = q_275_cast_fp16)[name = tensor<string, []>("op_60712_cast_fp16")];
            tensor<int32, [4]> var_60716_begin_0 = const()[name = tensor<string, []>("op_60716_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_60716_end_0 = const()[name = tensor<string, []>("op_60716_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_60716_end_mask_0 = const()[name = tensor<string, []>("op_60716_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60716_cast_fp16 = slice_by_index(begin = var_60716_begin_0, end = var_60716_end_0, end_mask = var_60716_end_mask_0, x = q_275_cast_fp16)[name = tensor<string, []>("op_60716_cast_fp16")];
            tensor<int32, [4]> var_60720_begin_0 = const()[name = tensor<string, []>("op_60720_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_60720_end_0 = const()[name = tensor<string, []>("op_60720_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_60720_end_mask_0 = const()[name = tensor<string, []>("op_60720_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60720_cast_fp16 = slice_by_index(begin = var_60720_begin_0, end = var_60720_end_0, end_mask = var_60720_end_mask_0, x = q_275_cast_fp16)[name = tensor<string, []>("op_60720_cast_fp16")];
            tensor<int32, [4]> var_60724_begin_0 = const()[name = tensor<string, []>("op_60724_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_60724_end_0 = const()[name = tensor<string, []>("op_60724_end_0"), val = tensor<int32, [4]>([2, 384, 1, 4096])];
            tensor<bool, [4]> var_60724_end_mask_0 = const()[name = tensor<string, []>("op_60724_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60724_cast_fp16 = slice_by_index(begin = var_60724_begin_0, end = var_60724_end_0, end_mask = var_60724_end_mask_0, x = q_275_cast_fp16)[name = tensor<string, []>("op_60724_cast_fp16")];
            tensor<int32, [4]> var_60728_begin_0 = const()[name = tensor<string, []>("op_60728_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_60728_end_0 = const()[name = tensor<string, []>("op_60728_end_0"), val = tensor<int32, [4]>([2, 448, 1, 4096])];
            tensor<bool, [4]> var_60728_end_mask_0 = const()[name = tensor<string, []>("op_60728_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60728_cast_fp16 = slice_by_index(begin = var_60728_begin_0, end = var_60728_end_0, end_mask = var_60728_end_mask_0, x = q_275_cast_fp16)[name = tensor<string, []>("op_60728_cast_fp16")];
            tensor<int32, [4]> var_60732_begin_0 = const()[name = tensor<string, []>("op_60732_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_60732_end_0 = const()[name = tensor<string, []>("op_60732_end_0"), val = tensor<int32, [4]>([2, 512, 1, 4096])];
            tensor<bool, [4]> var_60732_end_mask_0 = const()[name = tensor<string, []>("op_60732_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60732_cast_fp16 = slice_by_index(begin = var_60732_begin_0, end = var_60732_end_0, end_mask = var_60732_end_mask_0, x = q_275_cast_fp16)[name = tensor<string, []>("op_60732_cast_fp16")];
            tensor<int32, [4]> var_60736_begin_0 = const()[name = tensor<string, []>("op_60736_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_60736_end_0 = const()[name = tensor<string, []>("op_60736_end_0"), val = tensor<int32, [4]>([2, 576, 1, 4096])];
            tensor<bool, [4]> var_60736_end_mask_0 = const()[name = tensor<string, []>("op_60736_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60736_cast_fp16 = slice_by_index(begin = var_60736_begin_0, end = var_60736_end_0, end_mask = var_60736_end_mask_0, x = q_275_cast_fp16)[name = tensor<string, []>("op_60736_cast_fp16")];
            tensor<int32, [4]> var_60740_begin_0 = const()[name = tensor<string, []>("op_60740_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_60740_end_0 = const()[name = tensor<string, []>("op_60740_end_0"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<bool, [4]> var_60740_end_mask_0 = const()[name = tensor<string, []>("op_60740_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60740_cast_fp16 = slice_by_index(begin = var_60740_begin_0, end = var_60740_end_0, end_mask = var_60740_end_mask_0, x = q_275_cast_fp16)[name = tensor<string, []>("op_60740_cast_fp16")];
            tensor<int32, [4]> k_551_perm_0 = const()[name = tensor<string, []>("k_551_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_60747_begin_0 = const()[name = tensor<string, []>("op_60747_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_60747_end_0 = const()[name = tensor<string, []>("op_60747_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_60747_end_mask_0 = const()[name = tensor<string, []>("op_60747_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 640]> k_551_cast_fp16 = transpose(perm = k_551_perm_0, x = k_549_cast_fp16)[name = tensor<string, []>("transpose_2")];
            tensor<fp16, [2, 77, 1, 64]> var_60747_cast_fp16 = slice_by_index(begin = var_60747_begin_0, end = var_60747_end_0, end_mask = var_60747_end_mask_0, x = k_551_cast_fp16)[name = tensor<string, []>("op_60747_cast_fp16")];
            tensor<int32, [4]> var_60751_begin_0 = const()[name = tensor<string, []>("op_60751_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_60751_end_0 = const()[name = tensor<string, []>("op_60751_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_60751_end_mask_0 = const()[name = tensor<string, []>("op_60751_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_60751_cast_fp16 = slice_by_index(begin = var_60751_begin_0, end = var_60751_end_0, end_mask = var_60751_end_mask_0, x = k_551_cast_fp16)[name = tensor<string, []>("op_60751_cast_fp16")];
            tensor<int32, [4]> var_60755_begin_0 = const()[name = tensor<string, []>("op_60755_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_60755_end_0 = const()[name = tensor<string, []>("op_60755_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_60755_end_mask_0 = const()[name = tensor<string, []>("op_60755_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_60755_cast_fp16 = slice_by_index(begin = var_60755_begin_0, end = var_60755_end_0, end_mask = var_60755_end_mask_0, x = k_551_cast_fp16)[name = tensor<string, []>("op_60755_cast_fp16")];
            tensor<int32, [4]> var_60759_begin_0 = const()[name = tensor<string, []>("op_60759_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_60759_end_0 = const()[name = tensor<string, []>("op_60759_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_60759_end_mask_0 = const()[name = tensor<string, []>("op_60759_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_60759_cast_fp16 = slice_by_index(begin = var_60759_begin_0, end = var_60759_end_0, end_mask = var_60759_end_mask_0, x = k_551_cast_fp16)[name = tensor<string, []>("op_60759_cast_fp16")];
            tensor<int32, [4]> var_60763_begin_0 = const()[name = tensor<string, []>("op_60763_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_60763_end_0 = const()[name = tensor<string, []>("op_60763_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_60763_end_mask_0 = const()[name = tensor<string, []>("op_60763_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_60763_cast_fp16 = slice_by_index(begin = var_60763_begin_0, end = var_60763_end_0, end_mask = var_60763_end_mask_0, x = k_551_cast_fp16)[name = tensor<string, []>("op_60763_cast_fp16")];
            tensor<int32, [4]> var_60767_begin_0 = const()[name = tensor<string, []>("op_60767_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_60767_end_0 = const()[name = tensor<string, []>("op_60767_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_60767_end_mask_0 = const()[name = tensor<string, []>("op_60767_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_60767_cast_fp16 = slice_by_index(begin = var_60767_begin_0, end = var_60767_end_0, end_mask = var_60767_end_mask_0, x = k_551_cast_fp16)[name = tensor<string, []>("op_60767_cast_fp16")];
            tensor<int32, [4]> var_60771_begin_0 = const()[name = tensor<string, []>("op_60771_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_60771_end_0 = const()[name = tensor<string, []>("op_60771_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_60771_end_mask_0 = const()[name = tensor<string, []>("op_60771_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_60771_cast_fp16 = slice_by_index(begin = var_60771_begin_0, end = var_60771_end_0, end_mask = var_60771_end_mask_0, x = k_551_cast_fp16)[name = tensor<string, []>("op_60771_cast_fp16")];
            tensor<int32, [4]> var_60775_begin_0 = const()[name = tensor<string, []>("op_60775_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_60775_end_0 = const()[name = tensor<string, []>("op_60775_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_60775_end_mask_0 = const()[name = tensor<string, []>("op_60775_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_60775_cast_fp16 = slice_by_index(begin = var_60775_begin_0, end = var_60775_end_0, end_mask = var_60775_end_mask_0, x = k_551_cast_fp16)[name = tensor<string, []>("op_60775_cast_fp16")];
            tensor<int32, [4]> var_60779_begin_0 = const()[name = tensor<string, []>("op_60779_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_60779_end_0 = const()[name = tensor<string, []>("op_60779_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_60779_end_mask_0 = const()[name = tensor<string, []>("op_60779_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_60779_cast_fp16 = slice_by_index(begin = var_60779_begin_0, end = var_60779_end_0, end_mask = var_60779_end_mask_0, x = k_551_cast_fp16)[name = tensor<string, []>("op_60779_cast_fp16")];
            tensor<int32, [4]> var_60783_begin_0 = const()[name = tensor<string, []>("op_60783_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_60783_end_0 = const()[name = tensor<string, []>("op_60783_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_60783_end_mask_0 = const()[name = tensor<string, []>("op_60783_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_60783_cast_fp16 = slice_by_index(begin = var_60783_begin_0, end = var_60783_end_0, end_mask = var_60783_end_mask_0, x = k_551_cast_fp16)[name = tensor<string, []>("op_60783_cast_fp16")];
            tensor<int32, [4]> var_60785_begin_0 = const()[name = tensor<string, []>("op_60785_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_60785_end_0 = const()[name = tensor<string, []>("op_60785_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_60785_end_mask_0 = const()[name = tensor<string, []>("op_60785_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_60785_cast_fp16 = slice_by_index(begin = var_60785_begin_0, end = var_60785_end_0, end_mask = var_60785_end_mask_0, x = v_275_cast_fp16)[name = tensor<string, []>("op_60785_cast_fp16")];
            tensor<int32, [4]> var_60789_begin_0 = const()[name = tensor<string, []>("op_60789_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_60789_end_0 = const()[name = tensor<string, []>("op_60789_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_60789_end_mask_0 = const()[name = tensor<string, []>("op_60789_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_60789_cast_fp16 = slice_by_index(begin = var_60789_begin_0, end = var_60789_end_0, end_mask = var_60789_end_mask_0, x = v_275_cast_fp16)[name = tensor<string, []>("op_60789_cast_fp16")];
            tensor<int32, [4]> var_60793_begin_0 = const()[name = tensor<string, []>("op_60793_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_60793_end_0 = const()[name = tensor<string, []>("op_60793_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_60793_end_mask_0 = const()[name = tensor<string, []>("op_60793_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_60793_cast_fp16 = slice_by_index(begin = var_60793_begin_0, end = var_60793_end_0, end_mask = var_60793_end_mask_0, x = v_275_cast_fp16)[name = tensor<string, []>("op_60793_cast_fp16")];
            tensor<int32, [4]> var_60797_begin_0 = const()[name = tensor<string, []>("op_60797_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_60797_end_0 = const()[name = tensor<string, []>("op_60797_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_60797_end_mask_0 = const()[name = tensor<string, []>("op_60797_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_60797_cast_fp16 = slice_by_index(begin = var_60797_begin_0, end = var_60797_end_0, end_mask = var_60797_end_mask_0, x = v_275_cast_fp16)[name = tensor<string, []>("op_60797_cast_fp16")];
            tensor<int32, [4]> var_60801_begin_0 = const()[name = tensor<string, []>("op_60801_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_60801_end_0 = const()[name = tensor<string, []>("op_60801_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_60801_end_mask_0 = const()[name = tensor<string, []>("op_60801_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_60801_cast_fp16 = slice_by_index(begin = var_60801_begin_0, end = var_60801_end_0, end_mask = var_60801_end_mask_0, x = v_275_cast_fp16)[name = tensor<string, []>("op_60801_cast_fp16")];
            tensor<int32, [4]> var_60805_begin_0 = const()[name = tensor<string, []>("op_60805_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_60805_end_0 = const()[name = tensor<string, []>("op_60805_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_60805_end_mask_0 = const()[name = tensor<string, []>("op_60805_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_60805_cast_fp16 = slice_by_index(begin = var_60805_begin_0, end = var_60805_end_0, end_mask = var_60805_end_mask_0, x = v_275_cast_fp16)[name = tensor<string, []>("op_60805_cast_fp16")];
            tensor<int32, [4]> var_60809_begin_0 = const()[name = tensor<string, []>("op_60809_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_60809_end_0 = const()[name = tensor<string, []>("op_60809_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_60809_end_mask_0 = const()[name = tensor<string, []>("op_60809_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_60809_cast_fp16 = slice_by_index(begin = var_60809_begin_0, end = var_60809_end_0, end_mask = var_60809_end_mask_0, x = v_275_cast_fp16)[name = tensor<string, []>("op_60809_cast_fp16")];
            tensor<int32, [4]> var_60813_begin_0 = const()[name = tensor<string, []>("op_60813_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_60813_end_0 = const()[name = tensor<string, []>("op_60813_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_60813_end_mask_0 = const()[name = tensor<string, []>("op_60813_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_60813_cast_fp16 = slice_by_index(begin = var_60813_begin_0, end = var_60813_end_0, end_mask = var_60813_end_mask_0, x = v_275_cast_fp16)[name = tensor<string, []>("op_60813_cast_fp16")];
            tensor<int32, [4]> var_60817_begin_0 = const()[name = tensor<string, []>("op_60817_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_60817_end_0 = const()[name = tensor<string, []>("op_60817_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_60817_end_mask_0 = const()[name = tensor<string, []>("op_60817_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_60817_cast_fp16 = slice_by_index(begin = var_60817_begin_0, end = var_60817_end_0, end_mask = var_60817_end_mask_0, x = v_275_cast_fp16)[name = tensor<string, []>("op_60817_cast_fp16")];
            tensor<int32, [4]> var_60821_begin_0 = const()[name = tensor<string, []>("op_60821_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_60821_end_0 = const()[name = tensor<string, []>("op_60821_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_60821_end_mask_0 = const()[name = tensor<string, []>("op_60821_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_60821_cast_fp16 = slice_by_index(begin = var_60821_begin_0, end = var_60821_end_0, end_mask = var_60821_end_mask_0, x = v_275_cast_fp16)[name = tensor<string, []>("op_60821_cast_fp16")];
            tensor<string, []> var_60825_equation_0 = const()[name = tensor<string, []>("op_60825_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_60825_cast_fp16 = einsum(equation = var_60825_equation_0, values = (var_60747_cast_fp16, var_60704_cast_fp16))[name = tensor<string, []>("op_60825_cast_fp16")];
            tensor<fp16, []> var_60826_to_fp16 = const()[name = tensor<string, []>("op_60826_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5141_cast_fp16 = mul(x = var_60825_cast_fp16, y = var_60826_to_fp16)[name = tensor<string, []>("aw_5141_cast_fp16")];
            tensor<string, []> var_60829_equation_0 = const()[name = tensor<string, []>("op_60829_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_60829_cast_fp16 = einsum(equation = var_60829_equation_0, values = (var_60751_cast_fp16, var_60708_cast_fp16))[name = tensor<string, []>("op_60829_cast_fp16")];
            tensor<fp16, []> var_60830_to_fp16 = const()[name = tensor<string, []>("op_60830_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5143_cast_fp16 = mul(x = var_60829_cast_fp16, y = var_60830_to_fp16)[name = tensor<string, []>("aw_5143_cast_fp16")];
            tensor<string, []> var_60833_equation_0 = const()[name = tensor<string, []>("op_60833_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_60833_cast_fp16 = einsum(equation = var_60833_equation_0, values = (var_60755_cast_fp16, var_60712_cast_fp16))[name = tensor<string, []>("op_60833_cast_fp16")];
            tensor<fp16, []> var_60834_to_fp16 = const()[name = tensor<string, []>("op_60834_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5145_cast_fp16 = mul(x = var_60833_cast_fp16, y = var_60834_to_fp16)[name = tensor<string, []>("aw_5145_cast_fp16")];
            tensor<string, []> var_60837_equation_0 = const()[name = tensor<string, []>("op_60837_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_60837_cast_fp16 = einsum(equation = var_60837_equation_0, values = (var_60759_cast_fp16, var_60716_cast_fp16))[name = tensor<string, []>("op_60837_cast_fp16")];
            tensor<fp16, []> var_60838_to_fp16 = const()[name = tensor<string, []>("op_60838_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5147_cast_fp16 = mul(x = var_60837_cast_fp16, y = var_60838_to_fp16)[name = tensor<string, []>("aw_5147_cast_fp16")];
            tensor<string, []> var_60841_equation_0 = const()[name = tensor<string, []>("op_60841_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_60841_cast_fp16 = einsum(equation = var_60841_equation_0, values = (var_60763_cast_fp16, var_60720_cast_fp16))[name = tensor<string, []>("op_60841_cast_fp16")];
            tensor<fp16, []> var_60842_to_fp16 = const()[name = tensor<string, []>("op_60842_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5149_cast_fp16 = mul(x = var_60841_cast_fp16, y = var_60842_to_fp16)[name = tensor<string, []>("aw_5149_cast_fp16")];
            tensor<string, []> var_60845_equation_0 = const()[name = tensor<string, []>("op_60845_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_60845_cast_fp16 = einsum(equation = var_60845_equation_0, values = (var_60767_cast_fp16, var_60724_cast_fp16))[name = tensor<string, []>("op_60845_cast_fp16")];
            tensor<fp16, []> var_60846_to_fp16 = const()[name = tensor<string, []>("op_60846_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5151_cast_fp16 = mul(x = var_60845_cast_fp16, y = var_60846_to_fp16)[name = tensor<string, []>("aw_5151_cast_fp16")];
            tensor<string, []> var_60849_equation_0 = const()[name = tensor<string, []>("op_60849_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_60849_cast_fp16 = einsum(equation = var_60849_equation_0, values = (var_60771_cast_fp16, var_60728_cast_fp16))[name = tensor<string, []>("op_60849_cast_fp16")];
            tensor<fp16, []> var_60850_to_fp16 = const()[name = tensor<string, []>("op_60850_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5153_cast_fp16 = mul(x = var_60849_cast_fp16, y = var_60850_to_fp16)[name = tensor<string, []>("aw_5153_cast_fp16")];
            tensor<string, []> var_60853_equation_0 = const()[name = tensor<string, []>("op_60853_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_60853_cast_fp16 = einsum(equation = var_60853_equation_0, values = (var_60775_cast_fp16, var_60732_cast_fp16))[name = tensor<string, []>("op_60853_cast_fp16")];
            tensor<fp16, []> var_60854_to_fp16 = const()[name = tensor<string, []>("op_60854_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5155_cast_fp16 = mul(x = var_60853_cast_fp16, y = var_60854_to_fp16)[name = tensor<string, []>("aw_5155_cast_fp16")];
            tensor<string, []> var_60857_equation_0 = const()[name = tensor<string, []>("op_60857_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_60857_cast_fp16 = einsum(equation = var_60857_equation_0, values = (var_60779_cast_fp16, var_60736_cast_fp16))[name = tensor<string, []>("op_60857_cast_fp16")];
            tensor<fp16, []> var_60858_to_fp16 = const()[name = tensor<string, []>("op_60858_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5157_cast_fp16 = mul(x = var_60857_cast_fp16, y = var_60858_to_fp16)[name = tensor<string, []>("aw_5157_cast_fp16")];
            tensor<string, []> var_60861_equation_0 = const()[name = tensor<string, []>("op_60861_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_60861_cast_fp16 = einsum(equation = var_60861_equation_0, values = (var_60783_cast_fp16, var_60740_cast_fp16))[name = tensor<string, []>("op_60861_cast_fp16")];
            tensor<fp16, []> var_60862_to_fp16 = const()[name = tensor<string, []>("op_60862_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5159_cast_fp16 = mul(x = var_60861_cast_fp16, y = var_60862_to_fp16)[name = tensor<string, []>("aw_5159_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_60864_cast_fp16 = softmax(axis = var_58034, x = aw_5141_cast_fp16)[name = tensor<string, []>("op_60864_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_60865_cast_fp16 = softmax(axis = var_58034, x = aw_5143_cast_fp16)[name = tensor<string, []>("op_60865_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_60866_cast_fp16 = softmax(axis = var_58034, x = aw_5145_cast_fp16)[name = tensor<string, []>("op_60866_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_60867_cast_fp16 = softmax(axis = var_58034, x = aw_5147_cast_fp16)[name = tensor<string, []>("op_60867_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_60868_cast_fp16 = softmax(axis = var_58034, x = aw_5149_cast_fp16)[name = tensor<string, []>("op_60868_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_60869_cast_fp16 = softmax(axis = var_58034, x = aw_5151_cast_fp16)[name = tensor<string, []>("op_60869_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_60870_cast_fp16 = softmax(axis = var_58034, x = aw_5153_cast_fp16)[name = tensor<string, []>("op_60870_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_60871_cast_fp16 = softmax(axis = var_58034, x = aw_5155_cast_fp16)[name = tensor<string, []>("op_60871_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_60872_cast_fp16 = softmax(axis = var_58034, x = aw_5157_cast_fp16)[name = tensor<string, []>("op_60872_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_60873_cast_fp16 = softmax(axis = var_58034, x = aw_5159_cast_fp16)[name = tensor<string, []>("op_60873_cast_fp16")];
            tensor<string, []> var_60875_equation_0 = const()[name = tensor<string, []>("op_60875_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60875_cast_fp16 = einsum(equation = var_60875_equation_0, values = (var_60785_cast_fp16, var_60864_cast_fp16))[name = tensor<string, []>("op_60875_cast_fp16")];
            tensor<string, []> var_60877_equation_0 = const()[name = tensor<string, []>("op_60877_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60877_cast_fp16 = einsum(equation = var_60877_equation_0, values = (var_60789_cast_fp16, var_60865_cast_fp16))[name = tensor<string, []>("op_60877_cast_fp16")];
            tensor<string, []> var_60879_equation_0 = const()[name = tensor<string, []>("op_60879_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60879_cast_fp16 = einsum(equation = var_60879_equation_0, values = (var_60793_cast_fp16, var_60866_cast_fp16))[name = tensor<string, []>("op_60879_cast_fp16")];
            tensor<string, []> var_60881_equation_0 = const()[name = tensor<string, []>("op_60881_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60881_cast_fp16 = einsum(equation = var_60881_equation_0, values = (var_60797_cast_fp16, var_60867_cast_fp16))[name = tensor<string, []>("op_60881_cast_fp16")];
            tensor<string, []> var_60883_equation_0 = const()[name = tensor<string, []>("op_60883_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60883_cast_fp16 = einsum(equation = var_60883_equation_0, values = (var_60801_cast_fp16, var_60868_cast_fp16))[name = tensor<string, []>("op_60883_cast_fp16")];
            tensor<string, []> var_60885_equation_0 = const()[name = tensor<string, []>("op_60885_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60885_cast_fp16 = einsum(equation = var_60885_equation_0, values = (var_60805_cast_fp16, var_60869_cast_fp16))[name = tensor<string, []>("op_60885_cast_fp16")];
            tensor<string, []> var_60887_equation_0 = const()[name = tensor<string, []>("op_60887_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60887_cast_fp16 = einsum(equation = var_60887_equation_0, values = (var_60809_cast_fp16, var_60870_cast_fp16))[name = tensor<string, []>("op_60887_cast_fp16")];
            tensor<string, []> var_60889_equation_0 = const()[name = tensor<string, []>("op_60889_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60889_cast_fp16 = einsum(equation = var_60889_equation_0, values = (var_60813_cast_fp16, var_60871_cast_fp16))[name = tensor<string, []>("op_60889_cast_fp16")];
            tensor<string, []> var_60891_equation_0 = const()[name = tensor<string, []>("op_60891_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60891_cast_fp16 = einsum(equation = var_60891_equation_0, values = (var_60817_cast_fp16, var_60872_cast_fp16))[name = tensor<string, []>("op_60891_cast_fp16")];
            tensor<string, []> var_60893_equation_0 = const()[name = tensor<string, []>("op_60893_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60893_cast_fp16 = einsum(equation = var_60893_equation_0, values = (var_60821_cast_fp16, var_60873_cast_fp16))[name = tensor<string, []>("op_60893_cast_fp16")];
            tensor<bool, []> input_807_interleave_0 = const()[name = tensor<string, []>("input_807_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 640, 1, 4096]> input_807_cast_fp16 = concat(axis = var_58034, interleave = input_807_interleave_0, values = (var_60875_cast_fp16, var_60877_cast_fp16, var_60879_cast_fp16, var_60881_cast_fp16, var_60883_cast_fp16, var_60885_cast_fp16, var_60887_cast_fp16, var_60889_cast_fp16, var_60891_cast_fp16, var_60893_cast_fp16))[name = tensor<string, []>("input_807_cast_fp16")];
            tensor<string, []> var_60903_pad_type_0 = const()[name = tensor<string, []>("op_60903_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_60903_strides_0 = const()[name = tensor<string, []>("op_60903_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_60903_pad_0 = const()[name = tensor<string, []>("op_60903_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_60903_dilations_0 = const()[name = tensor<string, []>("op_60903_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_60903_groups_0 = const()[name = tensor<string, []>("op_60903_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 640, 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, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(936809152))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(937116416))), name = tensor<string, []>("up_blocks_1_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_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, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(937116608)))];
            tensor<fp16, [2, 640, 1, 4096]> var_60903_cast_fp16 = conv(bias = up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_60903_dilations_0, groups = var_60903_groups_0, pad = var_60903_pad_0, pad_type = var_60903_pad_type_0, strides = var_60903_strides_0, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized, x = input_807_cast_fp16)[name = tensor<string, []>("op_60903_cast_fp16")];
            tensor<fp16, [2, 640, 1, 4096]> inputs_413_cast_fp16 = add(x = var_60903_cast_fp16, y = inputs_411_cast_fp16)[name = tensor<string, []>("inputs_413_cast_fp16")];
            tensor<int32, [1]> input_809_axes_0 = const()[name = tensor<string, []>("input_809_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [640]> input_809_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_809_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(937117952)))];
            tensor<fp16, [640]> input_809_beta_0_to_fp16 = const()[name = tensor<string, []>("input_809_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(937119296)))];
            tensor<fp16, []> var_60913_to_fp16 = const()[name = tensor<string, []>("op_60913_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 640, 1, 4096]> input_809_cast_fp16 = layer_norm(axes = input_809_axes_0, beta = input_809_beta_0_to_fp16, epsilon = var_60913_to_fp16, gamma = input_809_gamma_0_to_fp16, x = inputs_413_cast_fp16)[name = tensor<string, []>("input_809_cast_fp16")];
            tensor<string, []> var_60933_pad_type_0 = const()[name = tensor<string, []>("op_60933_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_60933_strides_0 = const()[name = tensor<string, []>("op_60933_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_60933_pad_0 = const()[name = tensor<string, []>("op_60933_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_60933_dilations_0 = const()[name = tensor<string, []>("op_60933_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_60933_groups_0 = const()[name = tensor<string, []>("op_60933_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [5120, 640, 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, [2457600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(937120640))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(939578304))), name = tensor<string, []>("up_blocks_1_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_1_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(939578496)))];
            tensor<fp16, [2, 5120, 1, 4096]> var_60933_cast_fp16 = conv(bias = up_blocks_1_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_60933_dilations_0, groups = var_60933_groups_0, pad = var_60933_pad_0, pad_type = var_60933_pad_type_0, strides = var_60933_strides_0, weight = up_blocks_1_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized, x = input_809_cast_fp16)[name = tensor<string, []>("op_60933_cast_fp16")];
            tensor<int32, [2]> var_60934_split_sizes_0 = const()[name = tensor<string, []>("op_60934_split_sizes_0"), val = tensor<int32, [2]>([2560, 2560])];
            tensor<int32, []> var_60934_axis_0 = const()[name = tensor<string, []>("op_60934_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 2560, 1, 4096]> var_60934_cast_fp16_0, tensor<fp16, [2, 2560, 1, 4096]> var_60934_cast_fp16_1 = split(axis = var_60934_axis_0, split_sizes = var_60934_split_sizes_0, x = var_60933_cast_fp16)[name = tensor<string, []>("op_60934_cast_fp16")];
            tensor<string, []> var_60936_mode_0 = const()[name = tensor<string, []>("op_60936_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 2560, 1, 4096]> var_60936_cast_fp16 = gelu(mode = var_60936_mode_0, x = var_60934_cast_fp16_1)[name = tensor<string, []>("op_60936_cast_fp16")];
            tensor<fp16, [2, 2560, 1, 4096]> input_811_cast_fp16 = mul(x = var_60934_cast_fp16_0, y = var_60936_cast_fp16)[name = tensor<string, []>("input_811_cast_fp16")];
            tensor<string, []> var_60944_pad_type_0 = const()[name = tensor<string, []>("op_60944_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_60944_strides_0 = const()[name = tensor<string, []>("op_60944_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_60944_pad_0 = const()[name = tensor<string, []>("op_60944_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_60944_dilations_0 = const()[name = tensor<string, []>("op_60944_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_60944_groups_0 = const()[name = tensor<string, []>("op_60944_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 2560, 1, 1]> up_blocks_1_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(939588800))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(940817664))), name = tensor<string, []>("up_blocks_1_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_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, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(940817856)))];
            tensor<fp16, [2, 640, 1, 4096]> var_60944_cast_fp16 = conv(bias = up_blocks_1_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_60944_dilations_0, groups = var_60944_groups_0, pad = var_60944_pad_0, pad_type = var_60944_pad_type_0, strides = var_60944_strides_0, weight = up_blocks_1_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_811_cast_fp16)[name = tensor<string, []>("op_60944_cast_fp16")];
            tensor<fp16, [2, 640, 1, 4096]> inputs_415_cast_fp16 = add(x = var_60944_cast_fp16, y = inputs_413_cast_fp16)[name = tensor<string, []>("inputs_415_cast_fp16")];
            tensor<int32, [1]> hidden_states_567_axes_0 = const()[name = tensor<string, []>("hidden_states_567_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [640]> hidden_states_567_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_567_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(940819200)))];
            tensor<fp16, [640]> hidden_states_567_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_567_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(940820544)))];
            tensor<fp16, []> var_60960_to_fp16 = const()[name = tensor<string, []>("op_60960_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 640, 1, 4096]> hidden_states_567_cast_fp16 = layer_norm(axes = hidden_states_567_axes_0, beta = hidden_states_567_beta_0_to_fp16, epsilon = var_60960_to_fp16, gamma = hidden_states_567_gamma_0_to_fp16, x = inputs_415_cast_fp16)[name = tensor<string, []>("hidden_states_567_cast_fp16")];
            tensor<string, []> q_277_pad_type_0 = const()[name = tensor<string, []>("q_277_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_277_strides_0 = const()[name = tensor<string, []>("q_277_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_277_pad_0 = const()[name = tensor<string, []>("q_277_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_277_dilations_0 = const()[name = tensor<string, []>("q_277_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_277_groups_0 = const()[name = tensor<string, []>("q_277_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_2_transformer_blocks_1_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(940821888))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(941129152))), name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_1_attn1_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            tensor<fp16, [2, 640, 1, 4096]> q_277_cast_fp16 = conv(dilations = q_277_dilations_0, groups = q_277_groups_0, pad = q_277_pad_0, pad_type = q_277_pad_type_0, strides = q_277_strides_0, weight = up_blocks_1_attentions_2_transformer_blocks_1_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_567_cast_fp16)[name = tensor<string, []>("q_277_cast_fp16")];
            tensor<string, []> k_553_pad_type_0 = const()[name = tensor<string, []>("k_553_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_553_strides_0 = const()[name = tensor<string, []>("k_553_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_553_pad_0 = const()[name = tensor<string, []>("k_553_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_553_dilations_0 = const()[name = tensor<string, []>("k_553_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_553_groups_0 = const()[name = tensor<string, []>("k_553_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_2_transformer_blocks_1_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(941129344))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(941436608))), name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_1_attn1_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            tensor<fp16, [2, 640, 1, 4096]> k_553_cast_fp16 = conv(dilations = k_553_dilations_0, groups = k_553_groups_0, pad = k_553_pad_0, pad_type = k_553_pad_type_0, strides = k_553_strides_0, weight = up_blocks_1_attentions_2_transformer_blocks_1_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_567_cast_fp16)[name = tensor<string, []>("k_553_cast_fp16")];
            tensor<string, []> v_277_pad_type_0 = const()[name = tensor<string, []>("v_277_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_277_strides_0 = const()[name = tensor<string, []>("v_277_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_277_pad_0 = const()[name = tensor<string, []>("v_277_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_277_dilations_0 = const()[name = tensor<string, []>("v_277_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_277_groups_0 = const()[name = tensor<string, []>("v_277_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_2_transformer_blocks_1_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(941436800))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(941744064))), name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_1_attn1_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            tensor<fp16, [2, 640, 1, 4096]> v_277_cast_fp16 = conv(dilations = v_277_dilations_0, groups = v_277_groups_0, pad = v_277_pad_0, pad_type = v_277_pad_type_0, strides = v_277_strides_0, weight = up_blocks_1_attentions_2_transformer_blocks_1_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_567_cast_fp16)[name = tensor<string, []>("v_277_cast_fp16")];
            tensor<int32, [4]> var_60993_begin_0 = const()[name = tensor<string, []>("op_60993_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_60993_end_0 = const()[name = tensor<string, []>("op_60993_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_60993_end_mask_0 = const()[name = tensor<string, []>("op_60993_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60993_cast_fp16 = slice_by_index(begin = var_60993_begin_0, end = var_60993_end_0, end_mask = var_60993_end_mask_0, x = q_277_cast_fp16)[name = tensor<string, []>("op_60993_cast_fp16")];
            tensor<int32, [4]> var_60997_begin_0 = const()[name = tensor<string, []>("op_60997_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_60997_end_0 = const()[name = tensor<string, []>("op_60997_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_60997_end_mask_0 = const()[name = tensor<string, []>("op_60997_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60997_cast_fp16 = slice_by_index(begin = var_60997_begin_0, end = var_60997_end_0, end_mask = var_60997_end_mask_0, x = q_277_cast_fp16)[name = tensor<string, []>("op_60997_cast_fp16")];
            tensor<int32, [4]> var_61001_begin_0 = const()[name = tensor<string, []>("op_61001_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_61001_end_0 = const()[name = tensor<string, []>("op_61001_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_61001_end_mask_0 = const()[name = tensor<string, []>("op_61001_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61001_cast_fp16 = slice_by_index(begin = var_61001_begin_0, end = var_61001_end_0, end_mask = var_61001_end_mask_0, x = q_277_cast_fp16)[name = tensor<string, []>("op_61001_cast_fp16")];
            tensor<int32, [4]> var_61005_begin_0 = const()[name = tensor<string, []>("op_61005_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_61005_end_0 = const()[name = tensor<string, []>("op_61005_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_61005_end_mask_0 = const()[name = tensor<string, []>("op_61005_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61005_cast_fp16 = slice_by_index(begin = var_61005_begin_0, end = var_61005_end_0, end_mask = var_61005_end_mask_0, x = q_277_cast_fp16)[name = tensor<string, []>("op_61005_cast_fp16")];
            tensor<int32, [4]> var_61009_begin_0 = const()[name = tensor<string, []>("op_61009_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_61009_end_0 = const()[name = tensor<string, []>("op_61009_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_61009_end_mask_0 = const()[name = tensor<string, []>("op_61009_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61009_cast_fp16 = slice_by_index(begin = var_61009_begin_0, end = var_61009_end_0, end_mask = var_61009_end_mask_0, x = q_277_cast_fp16)[name = tensor<string, []>("op_61009_cast_fp16")];
            tensor<int32, [4]> var_61013_begin_0 = const()[name = tensor<string, []>("op_61013_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_61013_end_0 = const()[name = tensor<string, []>("op_61013_end_0"), val = tensor<int32, [4]>([2, 384, 1, 4096])];
            tensor<bool, [4]> var_61013_end_mask_0 = const()[name = tensor<string, []>("op_61013_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61013_cast_fp16 = slice_by_index(begin = var_61013_begin_0, end = var_61013_end_0, end_mask = var_61013_end_mask_0, x = q_277_cast_fp16)[name = tensor<string, []>("op_61013_cast_fp16")];
            tensor<int32, [4]> var_61017_begin_0 = const()[name = tensor<string, []>("op_61017_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_61017_end_0 = const()[name = tensor<string, []>("op_61017_end_0"), val = tensor<int32, [4]>([2, 448, 1, 4096])];
            tensor<bool, [4]> var_61017_end_mask_0 = const()[name = tensor<string, []>("op_61017_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61017_cast_fp16 = slice_by_index(begin = var_61017_begin_0, end = var_61017_end_0, end_mask = var_61017_end_mask_0, x = q_277_cast_fp16)[name = tensor<string, []>("op_61017_cast_fp16")];
            tensor<int32, [4]> var_61021_begin_0 = const()[name = tensor<string, []>("op_61021_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_61021_end_0 = const()[name = tensor<string, []>("op_61021_end_0"), val = tensor<int32, [4]>([2, 512, 1, 4096])];
            tensor<bool, [4]> var_61021_end_mask_0 = const()[name = tensor<string, []>("op_61021_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61021_cast_fp16 = slice_by_index(begin = var_61021_begin_0, end = var_61021_end_0, end_mask = var_61021_end_mask_0, x = q_277_cast_fp16)[name = tensor<string, []>("op_61021_cast_fp16")];
            tensor<int32, [4]> var_61025_begin_0 = const()[name = tensor<string, []>("op_61025_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_61025_end_0 = const()[name = tensor<string, []>("op_61025_end_0"), val = tensor<int32, [4]>([2, 576, 1, 4096])];
            tensor<bool, [4]> var_61025_end_mask_0 = const()[name = tensor<string, []>("op_61025_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61025_cast_fp16 = slice_by_index(begin = var_61025_begin_0, end = var_61025_end_0, end_mask = var_61025_end_mask_0, x = q_277_cast_fp16)[name = tensor<string, []>("op_61025_cast_fp16")];
            tensor<int32, [4]> var_61029_begin_0 = const()[name = tensor<string, []>("op_61029_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_61029_end_0 = const()[name = tensor<string, []>("op_61029_end_0"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<bool, [4]> var_61029_end_mask_0 = const()[name = tensor<string, []>("op_61029_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61029_cast_fp16 = slice_by_index(begin = var_61029_begin_0, end = var_61029_end_0, end_mask = var_61029_end_mask_0, x = q_277_cast_fp16)[name = tensor<string, []>("op_61029_cast_fp16")];
            tensor<int32, [4]> k_555_perm_0 = const()[name = tensor<string, []>("k_555_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_61036_begin_0 = const()[name = tensor<string, []>("op_61036_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_61036_end_0 = const()[name = tensor<string, []>("op_61036_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 64])];
            tensor<bool, [4]> var_61036_end_mask_0 = const()[name = tensor<string, []>("op_61036_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 640]> k_555_cast_fp16 = transpose(perm = k_555_perm_0, x = k_553_cast_fp16)[name = tensor<string, []>("transpose_1")];
            tensor<fp16, [2, 4096, 1, 64]> var_61036_cast_fp16 = slice_by_index(begin = var_61036_begin_0, end = var_61036_end_0, end_mask = var_61036_end_mask_0, x = k_555_cast_fp16)[name = tensor<string, []>("op_61036_cast_fp16")];
            tensor<int32, [4]> var_61040_begin_0 = const()[name = tensor<string, []>("op_61040_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_61040_end_0 = const()[name = tensor<string, []>("op_61040_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 128])];
            tensor<bool, [4]> var_61040_end_mask_0 = const()[name = tensor<string, []>("op_61040_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_61040_cast_fp16 = slice_by_index(begin = var_61040_begin_0, end = var_61040_end_0, end_mask = var_61040_end_mask_0, x = k_555_cast_fp16)[name = tensor<string, []>("op_61040_cast_fp16")];
            tensor<int32, [4]> var_61044_begin_0 = const()[name = tensor<string, []>("op_61044_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_61044_end_0 = const()[name = tensor<string, []>("op_61044_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 192])];
            tensor<bool, [4]> var_61044_end_mask_0 = const()[name = tensor<string, []>("op_61044_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_61044_cast_fp16 = slice_by_index(begin = var_61044_begin_0, end = var_61044_end_0, end_mask = var_61044_end_mask_0, x = k_555_cast_fp16)[name = tensor<string, []>("op_61044_cast_fp16")];
            tensor<int32, [4]> var_61048_begin_0 = const()[name = tensor<string, []>("op_61048_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_61048_end_0 = const()[name = tensor<string, []>("op_61048_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 256])];
            tensor<bool, [4]> var_61048_end_mask_0 = const()[name = tensor<string, []>("op_61048_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_61048_cast_fp16 = slice_by_index(begin = var_61048_begin_0, end = var_61048_end_0, end_mask = var_61048_end_mask_0, x = k_555_cast_fp16)[name = tensor<string, []>("op_61048_cast_fp16")];
            tensor<int32, [4]> var_61052_begin_0 = const()[name = tensor<string, []>("op_61052_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_61052_end_0 = const()[name = tensor<string, []>("op_61052_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 320])];
            tensor<bool, [4]> var_61052_end_mask_0 = const()[name = tensor<string, []>("op_61052_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_61052_cast_fp16 = slice_by_index(begin = var_61052_begin_0, end = var_61052_end_0, end_mask = var_61052_end_mask_0, x = k_555_cast_fp16)[name = tensor<string, []>("op_61052_cast_fp16")];
            tensor<int32, [4]> var_61056_begin_0 = const()[name = tensor<string, []>("op_61056_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_61056_end_0 = const()[name = tensor<string, []>("op_61056_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 384])];
            tensor<bool, [4]> var_61056_end_mask_0 = const()[name = tensor<string, []>("op_61056_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_61056_cast_fp16 = slice_by_index(begin = var_61056_begin_0, end = var_61056_end_0, end_mask = var_61056_end_mask_0, x = k_555_cast_fp16)[name = tensor<string, []>("op_61056_cast_fp16")];
            tensor<int32, [4]> var_61060_begin_0 = const()[name = tensor<string, []>("op_61060_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_61060_end_0 = const()[name = tensor<string, []>("op_61060_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 448])];
            tensor<bool, [4]> var_61060_end_mask_0 = const()[name = tensor<string, []>("op_61060_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_61060_cast_fp16 = slice_by_index(begin = var_61060_begin_0, end = var_61060_end_0, end_mask = var_61060_end_mask_0, x = k_555_cast_fp16)[name = tensor<string, []>("op_61060_cast_fp16")];
            tensor<int32, [4]> var_61064_begin_0 = const()[name = tensor<string, []>("op_61064_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_61064_end_0 = const()[name = tensor<string, []>("op_61064_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 512])];
            tensor<bool, [4]> var_61064_end_mask_0 = const()[name = tensor<string, []>("op_61064_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_61064_cast_fp16 = slice_by_index(begin = var_61064_begin_0, end = var_61064_end_0, end_mask = var_61064_end_mask_0, x = k_555_cast_fp16)[name = tensor<string, []>("op_61064_cast_fp16")];
            tensor<int32, [4]> var_61068_begin_0 = const()[name = tensor<string, []>("op_61068_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_61068_end_0 = const()[name = tensor<string, []>("op_61068_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 576])];
            tensor<bool, [4]> var_61068_end_mask_0 = const()[name = tensor<string, []>("op_61068_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_61068_cast_fp16 = slice_by_index(begin = var_61068_begin_0, end = var_61068_end_0, end_mask = var_61068_end_mask_0, x = k_555_cast_fp16)[name = tensor<string, []>("op_61068_cast_fp16")];
            tensor<int32, [4]> var_61072_begin_0 = const()[name = tensor<string, []>("op_61072_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_61072_end_0 = const()[name = tensor<string, []>("op_61072_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 640])];
            tensor<bool, [4]> var_61072_end_mask_0 = const()[name = tensor<string, []>("op_61072_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_61072_cast_fp16 = slice_by_index(begin = var_61072_begin_0, end = var_61072_end_0, end_mask = var_61072_end_mask_0, x = k_555_cast_fp16)[name = tensor<string, []>("op_61072_cast_fp16")];
            tensor<int32, [4]> var_61074_begin_0 = const()[name = tensor<string, []>("op_61074_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_61074_end_0 = const()[name = tensor<string, []>("op_61074_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_61074_end_mask_0 = const()[name = tensor<string, []>("op_61074_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61074_cast_fp16 = slice_by_index(begin = var_61074_begin_0, end = var_61074_end_0, end_mask = var_61074_end_mask_0, x = v_277_cast_fp16)[name = tensor<string, []>("op_61074_cast_fp16")];
            tensor<int32, [4]> var_61078_begin_0 = const()[name = tensor<string, []>("op_61078_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_61078_end_0 = const()[name = tensor<string, []>("op_61078_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_61078_end_mask_0 = const()[name = tensor<string, []>("op_61078_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61078_cast_fp16 = slice_by_index(begin = var_61078_begin_0, end = var_61078_end_0, end_mask = var_61078_end_mask_0, x = v_277_cast_fp16)[name = tensor<string, []>("op_61078_cast_fp16")];
            tensor<int32, [4]> var_61082_begin_0 = const()[name = tensor<string, []>("op_61082_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_61082_end_0 = const()[name = tensor<string, []>("op_61082_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_61082_end_mask_0 = const()[name = tensor<string, []>("op_61082_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61082_cast_fp16 = slice_by_index(begin = var_61082_begin_0, end = var_61082_end_0, end_mask = var_61082_end_mask_0, x = v_277_cast_fp16)[name = tensor<string, []>("op_61082_cast_fp16")];
            tensor<int32, [4]> var_61086_begin_0 = const()[name = tensor<string, []>("op_61086_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_61086_end_0 = const()[name = tensor<string, []>("op_61086_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_61086_end_mask_0 = const()[name = tensor<string, []>("op_61086_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61086_cast_fp16 = slice_by_index(begin = var_61086_begin_0, end = var_61086_end_0, end_mask = var_61086_end_mask_0, x = v_277_cast_fp16)[name = tensor<string, []>("op_61086_cast_fp16")];
            tensor<int32, [4]> var_61090_begin_0 = const()[name = tensor<string, []>("op_61090_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_61090_end_0 = const()[name = tensor<string, []>("op_61090_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_61090_end_mask_0 = const()[name = tensor<string, []>("op_61090_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61090_cast_fp16 = slice_by_index(begin = var_61090_begin_0, end = var_61090_end_0, end_mask = var_61090_end_mask_0, x = v_277_cast_fp16)[name = tensor<string, []>("op_61090_cast_fp16")];
            tensor<int32, [4]> var_61094_begin_0 = const()[name = tensor<string, []>("op_61094_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_61094_end_0 = const()[name = tensor<string, []>("op_61094_end_0"), val = tensor<int32, [4]>([2, 384, 1, 4096])];
            tensor<bool, [4]> var_61094_end_mask_0 = const()[name = tensor<string, []>("op_61094_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61094_cast_fp16 = slice_by_index(begin = var_61094_begin_0, end = var_61094_end_0, end_mask = var_61094_end_mask_0, x = v_277_cast_fp16)[name = tensor<string, []>("op_61094_cast_fp16")];
            tensor<int32, [4]> var_61098_begin_0 = const()[name = tensor<string, []>("op_61098_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_61098_end_0 = const()[name = tensor<string, []>("op_61098_end_0"), val = tensor<int32, [4]>([2, 448, 1, 4096])];
            tensor<bool, [4]> var_61098_end_mask_0 = const()[name = tensor<string, []>("op_61098_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61098_cast_fp16 = slice_by_index(begin = var_61098_begin_0, end = var_61098_end_0, end_mask = var_61098_end_mask_0, x = v_277_cast_fp16)[name = tensor<string, []>("op_61098_cast_fp16")];
            tensor<int32, [4]> var_61102_begin_0 = const()[name = tensor<string, []>("op_61102_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_61102_end_0 = const()[name = tensor<string, []>("op_61102_end_0"), val = tensor<int32, [4]>([2, 512, 1, 4096])];
            tensor<bool, [4]> var_61102_end_mask_0 = const()[name = tensor<string, []>("op_61102_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61102_cast_fp16 = slice_by_index(begin = var_61102_begin_0, end = var_61102_end_0, end_mask = var_61102_end_mask_0, x = v_277_cast_fp16)[name = tensor<string, []>("op_61102_cast_fp16")];
            tensor<int32, [4]> var_61106_begin_0 = const()[name = tensor<string, []>("op_61106_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_61106_end_0 = const()[name = tensor<string, []>("op_61106_end_0"), val = tensor<int32, [4]>([2, 576, 1, 4096])];
            tensor<bool, [4]> var_61106_end_mask_0 = const()[name = tensor<string, []>("op_61106_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61106_cast_fp16 = slice_by_index(begin = var_61106_begin_0, end = var_61106_end_0, end_mask = var_61106_end_mask_0, x = v_277_cast_fp16)[name = tensor<string, []>("op_61106_cast_fp16")];
            tensor<int32, [4]> var_61110_begin_0 = const()[name = tensor<string, []>("op_61110_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_61110_end_0 = const()[name = tensor<string, []>("op_61110_end_0"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<bool, [4]> var_61110_end_mask_0 = const()[name = tensor<string, []>("op_61110_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61110_cast_fp16 = slice_by_index(begin = var_61110_begin_0, end = var_61110_end_0, end_mask = var_61110_end_mask_0, x = v_277_cast_fp16)[name = tensor<string, []>("op_61110_cast_fp16")];
            tensor<string, []> var_61114_equation_0 = const()[name = tensor<string, []>("op_61114_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_61114_cast_fp16 = einsum(equation = var_61114_equation_0, values = (var_61036_cast_fp16, var_60993_cast_fp16))[name = tensor<string, []>("op_61114_cast_fp16")];
            tensor<fp16, []> var_61115_to_fp16 = const()[name = tensor<string, []>("op_61115_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5161_cast_fp16 = mul(x = var_61114_cast_fp16, y = var_61115_to_fp16)[name = tensor<string, []>("aw_5161_cast_fp16")];
            tensor<string, []> var_61118_equation_0 = const()[name = tensor<string, []>("op_61118_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_61118_cast_fp16 = einsum(equation = var_61118_equation_0, values = (var_61040_cast_fp16, var_60997_cast_fp16))[name = tensor<string, []>("op_61118_cast_fp16")];
            tensor<fp16, []> var_61119_to_fp16 = const()[name = tensor<string, []>("op_61119_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5163_cast_fp16 = mul(x = var_61118_cast_fp16, y = var_61119_to_fp16)[name = tensor<string, []>("aw_5163_cast_fp16")];
            tensor<string, []> var_61122_equation_0 = const()[name = tensor<string, []>("op_61122_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_61122_cast_fp16 = einsum(equation = var_61122_equation_0, values = (var_61044_cast_fp16, var_61001_cast_fp16))[name = tensor<string, []>("op_61122_cast_fp16")];
            tensor<fp16, []> var_61123_to_fp16 = const()[name = tensor<string, []>("op_61123_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5165_cast_fp16 = mul(x = var_61122_cast_fp16, y = var_61123_to_fp16)[name = tensor<string, []>("aw_5165_cast_fp16")];
            tensor<string, []> var_61126_equation_0 = const()[name = tensor<string, []>("op_61126_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_61126_cast_fp16 = einsum(equation = var_61126_equation_0, values = (var_61048_cast_fp16, var_61005_cast_fp16))[name = tensor<string, []>("op_61126_cast_fp16")];
            tensor<fp16, []> var_61127_to_fp16 = const()[name = tensor<string, []>("op_61127_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5167_cast_fp16 = mul(x = var_61126_cast_fp16, y = var_61127_to_fp16)[name = tensor<string, []>("aw_5167_cast_fp16")];
            tensor<string, []> var_61130_equation_0 = const()[name = tensor<string, []>("op_61130_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_61130_cast_fp16 = einsum(equation = var_61130_equation_0, values = (var_61052_cast_fp16, var_61009_cast_fp16))[name = tensor<string, []>("op_61130_cast_fp16")];
            tensor<fp16, []> var_61131_to_fp16 = const()[name = tensor<string, []>("op_61131_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5169_cast_fp16 = mul(x = var_61130_cast_fp16, y = var_61131_to_fp16)[name = tensor<string, []>("aw_5169_cast_fp16")];
            tensor<string, []> var_61134_equation_0 = const()[name = tensor<string, []>("op_61134_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_61134_cast_fp16 = einsum(equation = var_61134_equation_0, values = (var_61056_cast_fp16, var_61013_cast_fp16))[name = tensor<string, []>("op_61134_cast_fp16")];
            tensor<fp16, []> var_61135_to_fp16 = const()[name = tensor<string, []>("op_61135_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5171_cast_fp16 = mul(x = var_61134_cast_fp16, y = var_61135_to_fp16)[name = tensor<string, []>("aw_5171_cast_fp16")];
            tensor<string, []> var_61138_equation_0 = const()[name = tensor<string, []>("op_61138_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_61138_cast_fp16 = einsum(equation = var_61138_equation_0, values = (var_61060_cast_fp16, var_61017_cast_fp16))[name = tensor<string, []>("op_61138_cast_fp16")];
            tensor<fp16, []> var_61139_to_fp16 = const()[name = tensor<string, []>("op_61139_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5173_cast_fp16 = mul(x = var_61138_cast_fp16, y = var_61139_to_fp16)[name = tensor<string, []>("aw_5173_cast_fp16")];
            tensor<string, []> var_61142_equation_0 = const()[name = tensor<string, []>("op_61142_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_61142_cast_fp16 = einsum(equation = var_61142_equation_0, values = (var_61064_cast_fp16, var_61021_cast_fp16))[name = tensor<string, []>("op_61142_cast_fp16")];
            tensor<fp16, []> var_61143_to_fp16 = const()[name = tensor<string, []>("op_61143_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5175_cast_fp16 = mul(x = var_61142_cast_fp16, y = var_61143_to_fp16)[name = tensor<string, []>("aw_5175_cast_fp16")];
            tensor<string, []> var_61146_equation_0 = const()[name = tensor<string, []>("op_61146_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_61146_cast_fp16 = einsum(equation = var_61146_equation_0, values = (var_61068_cast_fp16, var_61025_cast_fp16))[name = tensor<string, []>("op_61146_cast_fp16")];
            tensor<fp16, []> var_61147_to_fp16 = const()[name = tensor<string, []>("op_61147_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5177_cast_fp16 = mul(x = var_61146_cast_fp16, y = var_61147_to_fp16)[name = tensor<string, []>("aw_5177_cast_fp16")];
            tensor<string, []> var_61150_equation_0 = const()[name = tensor<string, []>("op_61150_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_61150_cast_fp16 = einsum(equation = var_61150_equation_0, values = (var_61072_cast_fp16, var_61029_cast_fp16))[name = tensor<string, []>("op_61150_cast_fp16")];
            tensor<fp16, []> var_61151_to_fp16 = const()[name = tensor<string, []>("op_61151_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5179_cast_fp16 = mul(x = var_61150_cast_fp16, y = var_61151_to_fp16)[name = tensor<string, []>("aw_5179_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_61153_cast_fp16 = softmax(axis = var_58034, x = aw_5161_cast_fp16)[name = tensor<string, []>("op_61153_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_61154_cast_fp16 = softmax(axis = var_58034, x = aw_5163_cast_fp16)[name = tensor<string, []>("op_61154_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_61155_cast_fp16 = softmax(axis = var_58034, x = aw_5165_cast_fp16)[name = tensor<string, []>("op_61155_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_61156_cast_fp16 = softmax(axis = var_58034, x = aw_5167_cast_fp16)[name = tensor<string, []>("op_61156_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_61157_cast_fp16 = softmax(axis = var_58034, x = aw_5169_cast_fp16)[name = tensor<string, []>("op_61157_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_61158_cast_fp16 = softmax(axis = var_58034, x = aw_5171_cast_fp16)[name = tensor<string, []>("op_61158_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_61159_cast_fp16 = softmax(axis = var_58034, x = aw_5173_cast_fp16)[name = tensor<string, []>("op_61159_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_61160_cast_fp16 = softmax(axis = var_58034, x = aw_5175_cast_fp16)[name = tensor<string, []>("op_61160_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_61161_cast_fp16 = softmax(axis = var_58034, x = aw_5177_cast_fp16)[name = tensor<string, []>("op_61161_cast_fp16")];
            tensor<fp16, [2, 4096, 1, 4096]> var_61162_cast_fp16 = softmax(axis = var_58034, x = aw_5179_cast_fp16)[name = tensor<string, []>("op_61162_cast_fp16")];
            tensor<string, []> var_61164_equation_0 = const()[name = tensor<string, []>("op_61164_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_61164_cast_fp16 = einsum(equation = var_61164_equation_0, values = (var_61074_cast_fp16, var_61153_cast_fp16))[name = tensor<string, []>("op_61164_cast_fp16")];
            tensor<string, []> var_61166_equation_0 = const()[name = tensor<string, []>("op_61166_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_61166_cast_fp16 = einsum(equation = var_61166_equation_0, values = (var_61078_cast_fp16, var_61154_cast_fp16))[name = tensor<string, []>("op_61166_cast_fp16")];
            tensor<string, []> var_61168_equation_0 = const()[name = tensor<string, []>("op_61168_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_61168_cast_fp16 = einsum(equation = var_61168_equation_0, values = (var_61082_cast_fp16, var_61155_cast_fp16))[name = tensor<string, []>("op_61168_cast_fp16")];
            tensor<string, []> var_61170_equation_0 = const()[name = tensor<string, []>("op_61170_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_61170_cast_fp16 = einsum(equation = var_61170_equation_0, values = (var_61086_cast_fp16, var_61156_cast_fp16))[name = tensor<string, []>("op_61170_cast_fp16")];
            tensor<string, []> var_61172_equation_0 = const()[name = tensor<string, []>("op_61172_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_61172_cast_fp16 = einsum(equation = var_61172_equation_0, values = (var_61090_cast_fp16, var_61157_cast_fp16))[name = tensor<string, []>("op_61172_cast_fp16")];
            tensor<string, []> var_61174_equation_0 = const()[name = tensor<string, []>("op_61174_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_61174_cast_fp16 = einsum(equation = var_61174_equation_0, values = (var_61094_cast_fp16, var_61158_cast_fp16))[name = tensor<string, []>("op_61174_cast_fp16")];
            tensor<string, []> var_61176_equation_0 = const()[name = tensor<string, []>("op_61176_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_61176_cast_fp16 = einsum(equation = var_61176_equation_0, values = (var_61098_cast_fp16, var_61159_cast_fp16))[name = tensor<string, []>("op_61176_cast_fp16")];
            tensor<string, []> var_61178_equation_0 = const()[name = tensor<string, []>("op_61178_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_61178_cast_fp16 = einsum(equation = var_61178_equation_0, values = (var_61102_cast_fp16, var_61160_cast_fp16))[name = tensor<string, []>("op_61178_cast_fp16")];
            tensor<string, []> var_61180_equation_0 = const()[name = tensor<string, []>("op_61180_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_61180_cast_fp16 = einsum(equation = var_61180_equation_0, values = (var_61106_cast_fp16, var_61161_cast_fp16))[name = tensor<string, []>("op_61180_cast_fp16")];
            tensor<string, []> var_61182_equation_0 = const()[name = tensor<string, []>("op_61182_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_61182_cast_fp16 = einsum(equation = var_61182_equation_0, values = (var_61110_cast_fp16, var_61162_cast_fp16))[name = tensor<string, []>("op_61182_cast_fp16")];
            tensor<bool, []> input_813_interleave_0 = const()[name = tensor<string, []>("input_813_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 640, 1, 4096]> input_813_cast_fp16 = concat(axis = var_58034, interleave = input_813_interleave_0, values = (var_61164_cast_fp16, var_61166_cast_fp16, var_61168_cast_fp16, var_61170_cast_fp16, var_61172_cast_fp16, var_61174_cast_fp16, var_61176_cast_fp16, var_61178_cast_fp16, var_61180_cast_fp16, var_61182_cast_fp16))[name = tensor<string, []>("input_813_cast_fp16")];
            tensor<string, []> var_61192_pad_type_0 = const()[name = tensor<string, []>("op_61192_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_61192_strides_0 = const()[name = tensor<string, []>("op_61192_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_61192_pad_0 = const()[name = tensor<string, []>("op_61192_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_61192_dilations_0 = const()[name = tensor<string, []>("op_61192_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_61192_groups_0 = const()[name = tensor<string, []>("op_61192_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_2_transformer_blocks_1_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(941744256))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(942051520))), name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_1_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            tensor<fp16, [640]> up_blocks_1_attentions_2_transformer_blocks_1_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_1_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(942051712)))];
            tensor<fp16, [2, 640, 1, 4096]> var_61192_cast_fp16 = conv(bias = up_blocks_1_attentions_2_transformer_blocks_1_attn1_to_out_0_bias_to_fp16, dilations = var_61192_dilations_0, groups = var_61192_groups_0, pad = var_61192_pad_0, pad_type = var_61192_pad_type_0, strides = var_61192_strides_0, weight = up_blocks_1_attentions_2_transformer_blocks_1_attn1_to_out_0_weight_to_fp16_palettized, x = input_813_cast_fp16)[name = tensor<string, []>("op_61192_cast_fp16")];
            tensor<fp16, [2, 640, 1, 4096]> inputs_417_cast_fp16 = add(x = var_61192_cast_fp16, y = inputs_415_cast_fp16)[name = tensor<string, []>("inputs_417_cast_fp16")];
            tensor<int32, [1]> hidden_states_569_axes_0 = const()[name = tensor<string, []>("hidden_states_569_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [640]> hidden_states_569_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_569_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(942053056)))];
            tensor<fp16, [640]> hidden_states_569_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_569_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(942054400)))];
            tensor<fp16, []> var_61202_to_fp16 = const()[name = tensor<string, []>("op_61202_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 640, 1, 4096]> hidden_states_569_cast_fp16 = layer_norm(axes = hidden_states_569_axes_0, beta = hidden_states_569_beta_0_to_fp16, epsilon = var_61202_to_fp16, gamma = hidden_states_569_gamma_0_to_fp16, x = inputs_417_cast_fp16)[name = tensor<string, []>("hidden_states_569_cast_fp16")];
            tensor<string, []> q_pad_type_0 = const()[name = tensor<string, []>("q_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_strides_0 = const()[name = tensor<string, []>("q_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_pad_0 = const()[name = tensor<string, []>("q_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_dilations_0 = const()[name = tensor<string, []>("q_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_groups_0 = const()[name = tensor<string, []>("q_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_2_transformer_blocks_1_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(942055744))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(942363008))), name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_1_attn2_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            tensor<fp16, [2, 640, 1, 4096]> q_cast_fp16 = conv(dilations = q_dilations_0, groups = q_groups_0, pad = q_pad_0, pad_type = q_pad_type_0, strides = q_strides_0, weight = up_blocks_1_attentions_2_transformer_blocks_1_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_569_cast_fp16)[name = tensor<string, []>("q_cast_fp16")];
            tensor<string, []> k_557_pad_type_0 = const()[name = tensor<string, []>("k_557_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_557_strides_0 = const()[name = tensor<string, []>("k_557_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_557_pad_0 = const()[name = tensor<string, []>("k_557_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_557_dilations_0 = const()[name = tensor<string, []>("k_557_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_557_groups_0 = const()[name = tensor<string, []>("k_557_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 2048, 1, 1]> up_blocks_1_attentions_2_transformer_blocks_1_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [983040]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(942363200))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(943346304))), name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_1_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 2048, 1, 1])];
            tensor<fp16, [2, 640, 1, 77]> k_557_cast_fp16 = conv(dilations = k_557_dilations_0, groups = k_557_groups_0, pad = k_557_pad_0, pad_type = k_557_pad_type_0, strides = k_557_strides_0, weight = up_blocks_1_attentions_2_transformer_blocks_1_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("k_557_cast_fp16")];
            tensor<string, []> v_pad_type_0 = const()[name = tensor<string, []>("v_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> v_strides_0 = const()[name = tensor<string, []>("v_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_pad_0 = const()[name = tensor<string, []>("v_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_dilations_0 = const()[name = tensor<string, []>("v_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> v_groups_0 = const()[name = tensor<string, []>("v_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 2048, 1, 1]> up_blocks_1_attentions_2_transformer_blocks_1_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [983040]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(943346496))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(944329600))), name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_1_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 2048, 1, 1])];
            tensor<fp16, [2, 640, 1, 77]> v_cast_fp16 = conv(dilations = v_dilations_0, groups = v_groups_0, pad = v_pad_0, pad_type = v_pad_type_0, strides = v_strides_0, weight = up_blocks_1_attentions_2_transformer_blocks_1_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("v_cast_fp16")];
            tensor<int32, [4]> var_61235_begin_0 = const()[name = tensor<string, []>("op_61235_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_61235_end_0 = const()[name = tensor<string, []>("op_61235_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_61235_end_mask_0 = const()[name = tensor<string, []>("op_61235_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61235_cast_fp16 = slice_by_index(begin = var_61235_begin_0, end = var_61235_end_0, end_mask = var_61235_end_mask_0, x = q_cast_fp16)[name = tensor<string, []>("op_61235_cast_fp16")];
            tensor<int32, [4]> var_61239_begin_0 = const()[name = tensor<string, []>("op_61239_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_61239_end_0 = const()[name = tensor<string, []>("op_61239_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_61239_end_mask_0 = const()[name = tensor<string, []>("op_61239_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61239_cast_fp16 = slice_by_index(begin = var_61239_begin_0, end = var_61239_end_0, end_mask = var_61239_end_mask_0, x = q_cast_fp16)[name = tensor<string, []>("op_61239_cast_fp16")];
            tensor<int32, [4]> var_61243_begin_0 = const()[name = tensor<string, []>("op_61243_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_61243_end_0 = const()[name = tensor<string, []>("op_61243_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_61243_end_mask_0 = const()[name = tensor<string, []>("op_61243_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61243_cast_fp16 = slice_by_index(begin = var_61243_begin_0, end = var_61243_end_0, end_mask = var_61243_end_mask_0, x = q_cast_fp16)[name = tensor<string, []>("op_61243_cast_fp16")];
            tensor<int32, [4]> var_61247_begin_0 = const()[name = tensor<string, []>("op_61247_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_61247_end_0 = const()[name = tensor<string, []>("op_61247_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_61247_end_mask_0 = const()[name = tensor<string, []>("op_61247_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61247_cast_fp16 = slice_by_index(begin = var_61247_begin_0, end = var_61247_end_0, end_mask = var_61247_end_mask_0, x = q_cast_fp16)[name = tensor<string, []>("op_61247_cast_fp16")];
            tensor<int32, [4]> var_61251_begin_0 = const()[name = tensor<string, []>("op_61251_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_61251_end_0 = const()[name = tensor<string, []>("op_61251_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_61251_end_mask_0 = const()[name = tensor<string, []>("op_61251_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61251_cast_fp16 = slice_by_index(begin = var_61251_begin_0, end = var_61251_end_0, end_mask = var_61251_end_mask_0, x = q_cast_fp16)[name = tensor<string, []>("op_61251_cast_fp16")];
            tensor<int32, [4]> var_61255_begin_0 = const()[name = tensor<string, []>("op_61255_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_61255_end_0 = const()[name = tensor<string, []>("op_61255_end_0"), val = tensor<int32, [4]>([2, 384, 1, 4096])];
            tensor<bool, [4]> var_61255_end_mask_0 = const()[name = tensor<string, []>("op_61255_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61255_cast_fp16 = slice_by_index(begin = var_61255_begin_0, end = var_61255_end_0, end_mask = var_61255_end_mask_0, x = q_cast_fp16)[name = tensor<string, []>("op_61255_cast_fp16")];
            tensor<int32, [4]> var_61259_begin_0 = const()[name = tensor<string, []>("op_61259_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_61259_end_0 = const()[name = tensor<string, []>("op_61259_end_0"), val = tensor<int32, [4]>([2, 448, 1, 4096])];
            tensor<bool, [4]> var_61259_end_mask_0 = const()[name = tensor<string, []>("op_61259_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61259_cast_fp16 = slice_by_index(begin = var_61259_begin_0, end = var_61259_end_0, end_mask = var_61259_end_mask_0, x = q_cast_fp16)[name = tensor<string, []>("op_61259_cast_fp16")];
            tensor<int32, [4]> var_61263_begin_0 = const()[name = tensor<string, []>("op_61263_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_61263_end_0 = const()[name = tensor<string, []>("op_61263_end_0"), val = tensor<int32, [4]>([2, 512, 1, 4096])];
            tensor<bool, [4]> var_61263_end_mask_0 = const()[name = tensor<string, []>("op_61263_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61263_cast_fp16 = slice_by_index(begin = var_61263_begin_0, end = var_61263_end_0, end_mask = var_61263_end_mask_0, x = q_cast_fp16)[name = tensor<string, []>("op_61263_cast_fp16")];
            tensor<int32, [4]> var_61267_begin_0 = const()[name = tensor<string, []>("op_61267_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_61267_end_0 = const()[name = tensor<string, []>("op_61267_end_0"), val = tensor<int32, [4]>([2, 576, 1, 4096])];
            tensor<bool, [4]> var_61267_end_mask_0 = const()[name = tensor<string, []>("op_61267_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61267_cast_fp16 = slice_by_index(begin = var_61267_begin_0, end = var_61267_end_0, end_mask = var_61267_end_mask_0, x = q_cast_fp16)[name = tensor<string, []>("op_61267_cast_fp16")];
            tensor<int32, [4]> var_61271_begin_0 = const()[name = tensor<string, []>("op_61271_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_61271_end_0 = const()[name = tensor<string, []>("op_61271_end_0"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<bool, [4]> var_61271_end_mask_0 = const()[name = tensor<string, []>("op_61271_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61271_cast_fp16 = slice_by_index(begin = var_61271_begin_0, end = var_61271_end_0, end_mask = var_61271_end_mask_0, x = q_cast_fp16)[name = tensor<string, []>("op_61271_cast_fp16")];
            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_61278_begin_0 = const()[name = tensor<string, []>("op_61278_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_61278_end_0 = const()[name = tensor<string, []>("op_61278_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_61278_end_mask_0 = const()[name = tensor<string, []>("op_61278_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 640]> k_cast_fp16 = transpose(perm = k_perm_0, x = k_557_cast_fp16)[name = tensor<string, []>("transpose_0")];
            tensor<fp16, [2, 77, 1, 64]> var_61278_cast_fp16 = slice_by_index(begin = var_61278_begin_0, end = var_61278_end_0, end_mask = var_61278_end_mask_0, x = k_cast_fp16)[name = tensor<string, []>("op_61278_cast_fp16")];
            tensor<int32, [4]> var_61282_begin_0 = const()[name = tensor<string, []>("op_61282_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_61282_end_0 = const()[name = tensor<string, []>("op_61282_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_61282_end_mask_0 = const()[name = tensor<string, []>("op_61282_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_61282_cast_fp16 = slice_by_index(begin = var_61282_begin_0, end = var_61282_end_0, end_mask = var_61282_end_mask_0, x = k_cast_fp16)[name = tensor<string, []>("op_61282_cast_fp16")];
            tensor<int32, [4]> var_61286_begin_0 = const()[name = tensor<string, []>("op_61286_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_61286_end_0 = const()[name = tensor<string, []>("op_61286_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_61286_end_mask_0 = const()[name = tensor<string, []>("op_61286_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_61286_cast_fp16 = slice_by_index(begin = var_61286_begin_0, end = var_61286_end_0, end_mask = var_61286_end_mask_0, x = k_cast_fp16)[name = tensor<string, []>("op_61286_cast_fp16")];
            tensor<int32, [4]> var_61290_begin_0 = const()[name = tensor<string, []>("op_61290_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_61290_end_0 = const()[name = tensor<string, []>("op_61290_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_61290_end_mask_0 = const()[name = tensor<string, []>("op_61290_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_61290_cast_fp16 = slice_by_index(begin = var_61290_begin_0, end = var_61290_end_0, end_mask = var_61290_end_mask_0, x = k_cast_fp16)[name = tensor<string, []>("op_61290_cast_fp16")];
            tensor<int32, [4]> var_61294_begin_0 = const()[name = tensor<string, []>("op_61294_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_61294_end_0 = const()[name = tensor<string, []>("op_61294_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_61294_end_mask_0 = const()[name = tensor<string, []>("op_61294_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_61294_cast_fp16 = slice_by_index(begin = var_61294_begin_0, end = var_61294_end_0, end_mask = var_61294_end_mask_0, x = k_cast_fp16)[name = tensor<string, []>("op_61294_cast_fp16")];
            tensor<int32, [4]> var_61298_begin_0 = const()[name = tensor<string, []>("op_61298_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_61298_end_0 = const()[name = tensor<string, []>("op_61298_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_61298_end_mask_0 = const()[name = tensor<string, []>("op_61298_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_61298_cast_fp16 = slice_by_index(begin = var_61298_begin_0, end = var_61298_end_0, end_mask = var_61298_end_mask_0, x = k_cast_fp16)[name = tensor<string, []>("op_61298_cast_fp16")];
            tensor<int32, [4]> var_61302_begin_0 = const()[name = tensor<string, []>("op_61302_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_61302_end_0 = const()[name = tensor<string, []>("op_61302_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_61302_end_mask_0 = const()[name = tensor<string, []>("op_61302_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_61302_cast_fp16 = slice_by_index(begin = var_61302_begin_0, end = var_61302_end_0, end_mask = var_61302_end_mask_0, x = k_cast_fp16)[name = tensor<string, []>("op_61302_cast_fp16")];
            tensor<int32, [4]> var_61306_begin_0 = const()[name = tensor<string, []>("op_61306_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_61306_end_0 = const()[name = tensor<string, []>("op_61306_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_61306_end_mask_0 = const()[name = tensor<string, []>("op_61306_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_61306_cast_fp16 = slice_by_index(begin = var_61306_begin_0, end = var_61306_end_0, end_mask = var_61306_end_mask_0, x = k_cast_fp16)[name = tensor<string, []>("op_61306_cast_fp16")];
            tensor<int32, [4]> var_61310_begin_0 = const()[name = tensor<string, []>("op_61310_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_61310_end_0 = const()[name = tensor<string, []>("op_61310_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_61310_end_mask_0 = const()[name = tensor<string, []>("op_61310_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_61310_cast_fp16 = slice_by_index(begin = var_61310_begin_0, end = var_61310_end_0, end_mask = var_61310_end_mask_0, x = k_cast_fp16)[name = tensor<string, []>("op_61310_cast_fp16")];
            tensor<int32, [4]> var_61314_begin_0 = const()[name = tensor<string, []>("op_61314_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_61314_end_0 = const()[name = tensor<string, []>("op_61314_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_61314_end_mask_0 = const()[name = tensor<string, []>("op_61314_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_61314_cast_fp16 = slice_by_index(begin = var_61314_begin_0, end = var_61314_end_0, end_mask = var_61314_end_mask_0, x = k_cast_fp16)[name = tensor<string, []>("op_61314_cast_fp16")];
            tensor<int32, [4]> var_61316_begin_0 = const()[name = tensor<string, []>("op_61316_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_61316_end_0 = const()[name = tensor<string, []>("op_61316_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_61316_end_mask_0 = const()[name = tensor<string, []>("op_61316_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_61316_cast_fp16 = slice_by_index(begin = var_61316_begin_0, end = var_61316_end_0, end_mask = var_61316_end_mask_0, x = v_cast_fp16)[name = tensor<string, []>("op_61316_cast_fp16")];
            tensor<int32, [4]> var_61320_begin_0 = const()[name = tensor<string, []>("op_61320_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_61320_end_0 = const()[name = tensor<string, []>("op_61320_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_61320_end_mask_0 = const()[name = tensor<string, []>("op_61320_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_61320_cast_fp16 = slice_by_index(begin = var_61320_begin_0, end = var_61320_end_0, end_mask = var_61320_end_mask_0, x = v_cast_fp16)[name = tensor<string, []>("op_61320_cast_fp16")];
            tensor<int32, [4]> var_61324_begin_0 = const()[name = tensor<string, []>("op_61324_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_61324_end_0 = const()[name = tensor<string, []>("op_61324_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_61324_end_mask_0 = const()[name = tensor<string, []>("op_61324_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_61324_cast_fp16 = slice_by_index(begin = var_61324_begin_0, end = var_61324_end_0, end_mask = var_61324_end_mask_0, x = v_cast_fp16)[name = tensor<string, []>("op_61324_cast_fp16")];
            tensor<int32, [4]> var_61328_begin_0 = const()[name = tensor<string, []>("op_61328_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_61328_end_0 = const()[name = tensor<string, []>("op_61328_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_61328_end_mask_0 = const()[name = tensor<string, []>("op_61328_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_61328_cast_fp16 = slice_by_index(begin = var_61328_begin_0, end = var_61328_end_0, end_mask = var_61328_end_mask_0, x = v_cast_fp16)[name = tensor<string, []>("op_61328_cast_fp16")];
            tensor<int32, [4]> var_61332_begin_0 = const()[name = tensor<string, []>("op_61332_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_61332_end_0 = const()[name = tensor<string, []>("op_61332_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_61332_end_mask_0 = const()[name = tensor<string, []>("op_61332_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_61332_cast_fp16 = slice_by_index(begin = var_61332_begin_0, end = var_61332_end_0, end_mask = var_61332_end_mask_0, x = v_cast_fp16)[name = tensor<string, []>("op_61332_cast_fp16")];
            tensor<int32, [4]> var_61336_begin_0 = const()[name = tensor<string, []>("op_61336_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_61336_end_0 = const()[name = tensor<string, []>("op_61336_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_61336_end_mask_0 = const()[name = tensor<string, []>("op_61336_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_61336_cast_fp16 = slice_by_index(begin = var_61336_begin_0, end = var_61336_end_0, end_mask = var_61336_end_mask_0, x = v_cast_fp16)[name = tensor<string, []>("op_61336_cast_fp16")];
            tensor<int32, [4]> var_61340_begin_0 = const()[name = tensor<string, []>("op_61340_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_61340_end_0 = const()[name = tensor<string, []>("op_61340_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_61340_end_mask_0 = const()[name = tensor<string, []>("op_61340_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_61340_cast_fp16 = slice_by_index(begin = var_61340_begin_0, end = var_61340_end_0, end_mask = var_61340_end_mask_0, x = v_cast_fp16)[name = tensor<string, []>("op_61340_cast_fp16")];
            tensor<int32, [4]> var_61344_begin_0 = const()[name = tensor<string, []>("op_61344_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_61344_end_0 = const()[name = tensor<string, []>("op_61344_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_61344_end_mask_0 = const()[name = tensor<string, []>("op_61344_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_61344_cast_fp16 = slice_by_index(begin = var_61344_begin_0, end = var_61344_end_0, end_mask = var_61344_end_mask_0, x = v_cast_fp16)[name = tensor<string, []>("op_61344_cast_fp16")];
            tensor<int32, [4]> var_61348_begin_0 = const()[name = tensor<string, []>("op_61348_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_61348_end_0 = const()[name = tensor<string, []>("op_61348_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_61348_end_mask_0 = const()[name = tensor<string, []>("op_61348_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_61348_cast_fp16 = slice_by_index(begin = var_61348_begin_0, end = var_61348_end_0, end_mask = var_61348_end_mask_0, x = v_cast_fp16)[name = tensor<string, []>("op_61348_cast_fp16")];
            tensor<int32, [4]> var_61352_begin_0 = const()[name = tensor<string, []>("op_61352_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_61352_end_0 = const()[name = tensor<string, []>("op_61352_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_61352_end_mask_0 = const()[name = tensor<string, []>("op_61352_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_61352_cast_fp16 = slice_by_index(begin = var_61352_begin_0, end = var_61352_end_0, end_mask = var_61352_end_mask_0, x = v_cast_fp16)[name = tensor<string, []>("op_61352_cast_fp16")];
            tensor<string, []> var_61356_equation_0 = const()[name = tensor<string, []>("op_61356_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_61356_cast_fp16 = einsum(equation = var_61356_equation_0, values = (var_61278_cast_fp16, var_61235_cast_fp16))[name = tensor<string, []>("op_61356_cast_fp16")];
            tensor<fp16, []> var_61357_to_fp16 = const()[name = tensor<string, []>("op_61357_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5181_cast_fp16 = mul(x = var_61356_cast_fp16, y = var_61357_to_fp16)[name = tensor<string, []>("aw_5181_cast_fp16")];
            tensor<string, []> var_61360_equation_0 = const()[name = tensor<string, []>("op_61360_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_61360_cast_fp16 = einsum(equation = var_61360_equation_0, values = (var_61282_cast_fp16, var_61239_cast_fp16))[name = tensor<string, []>("op_61360_cast_fp16")];
            tensor<fp16, []> var_61361_to_fp16 = const()[name = tensor<string, []>("op_61361_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5183_cast_fp16 = mul(x = var_61360_cast_fp16, y = var_61361_to_fp16)[name = tensor<string, []>("aw_5183_cast_fp16")];
            tensor<string, []> var_61364_equation_0 = const()[name = tensor<string, []>("op_61364_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_61364_cast_fp16 = einsum(equation = var_61364_equation_0, values = (var_61286_cast_fp16, var_61243_cast_fp16))[name = tensor<string, []>("op_61364_cast_fp16")];
            tensor<fp16, []> var_61365_to_fp16 = const()[name = tensor<string, []>("op_61365_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5185_cast_fp16 = mul(x = var_61364_cast_fp16, y = var_61365_to_fp16)[name = tensor<string, []>("aw_5185_cast_fp16")];
            tensor<string, []> var_61368_equation_0 = const()[name = tensor<string, []>("op_61368_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_61368_cast_fp16 = einsum(equation = var_61368_equation_0, values = (var_61290_cast_fp16, var_61247_cast_fp16))[name = tensor<string, []>("op_61368_cast_fp16")];
            tensor<fp16, []> var_61369_to_fp16 = const()[name = tensor<string, []>("op_61369_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5187_cast_fp16 = mul(x = var_61368_cast_fp16, y = var_61369_to_fp16)[name = tensor<string, []>("aw_5187_cast_fp16")];
            tensor<string, []> var_61372_equation_0 = const()[name = tensor<string, []>("op_61372_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_61372_cast_fp16 = einsum(equation = var_61372_equation_0, values = (var_61294_cast_fp16, var_61251_cast_fp16))[name = tensor<string, []>("op_61372_cast_fp16")];
            tensor<fp16, []> var_61373_to_fp16 = const()[name = tensor<string, []>("op_61373_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5189_cast_fp16 = mul(x = var_61372_cast_fp16, y = var_61373_to_fp16)[name = tensor<string, []>("aw_5189_cast_fp16")];
            tensor<string, []> var_61376_equation_0 = const()[name = tensor<string, []>("op_61376_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_61376_cast_fp16 = einsum(equation = var_61376_equation_0, values = (var_61298_cast_fp16, var_61255_cast_fp16))[name = tensor<string, []>("op_61376_cast_fp16")];
            tensor<fp16, []> var_61377_to_fp16 = const()[name = tensor<string, []>("op_61377_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5191_cast_fp16 = mul(x = var_61376_cast_fp16, y = var_61377_to_fp16)[name = tensor<string, []>("aw_5191_cast_fp16")];
            tensor<string, []> var_61380_equation_0 = const()[name = tensor<string, []>("op_61380_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_61380_cast_fp16 = einsum(equation = var_61380_equation_0, values = (var_61302_cast_fp16, var_61259_cast_fp16))[name = tensor<string, []>("op_61380_cast_fp16")];
            tensor<fp16, []> var_61381_to_fp16 = const()[name = tensor<string, []>("op_61381_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5193_cast_fp16 = mul(x = var_61380_cast_fp16, y = var_61381_to_fp16)[name = tensor<string, []>("aw_5193_cast_fp16")];
            tensor<string, []> var_61384_equation_0 = const()[name = tensor<string, []>("op_61384_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_61384_cast_fp16 = einsum(equation = var_61384_equation_0, values = (var_61306_cast_fp16, var_61263_cast_fp16))[name = tensor<string, []>("op_61384_cast_fp16")];
            tensor<fp16, []> var_61385_to_fp16 = const()[name = tensor<string, []>("op_61385_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5195_cast_fp16 = mul(x = var_61384_cast_fp16, y = var_61385_to_fp16)[name = tensor<string, []>("aw_5195_cast_fp16")];
            tensor<string, []> var_61388_equation_0 = const()[name = tensor<string, []>("op_61388_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_61388_cast_fp16 = einsum(equation = var_61388_equation_0, values = (var_61310_cast_fp16, var_61267_cast_fp16))[name = tensor<string, []>("op_61388_cast_fp16")];
            tensor<fp16, []> var_61389_to_fp16 = const()[name = tensor<string, []>("op_61389_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5197_cast_fp16 = mul(x = var_61388_cast_fp16, y = var_61389_to_fp16)[name = tensor<string, []>("aw_5197_cast_fp16")];
            tensor<string, []> var_61392_equation_0 = const()[name = tensor<string, []>("op_61392_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_61392_cast_fp16 = einsum(equation = var_61392_equation_0, values = (var_61314_cast_fp16, var_61271_cast_fp16))[name = tensor<string, []>("op_61392_cast_fp16")];
            tensor<fp16, []> var_61393_to_fp16 = const()[name = tensor<string, []>("op_61393_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_cast_fp16 = mul(x = var_61392_cast_fp16, y = var_61393_to_fp16)[name = tensor<string, []>("aw_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_61395_cast_fp16 = softmax(axis = var_58034, x = aw_5181_cast_fp16)[name = tensor<string, []>("op_61395_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_61396_cast_fp16 = softmax(axis = var_58034, x = aw_5183_cast_fp16)[name = tensor<string, []>("op_61396_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_61397_cast_fp16 = softmax(axis = var_58034, x = aw_5185_cast_fp16)[name = tensor<string, []>("op_61397_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_61398_cast_fp16 = softmax(axis = var_58034, x = aw_5187_cast_fp16)[name = tensor<string, []>("op_61398_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_61399_cast_fp16 = softmax(axis = var_58034, x = aw_5189_cast_fp16)[name = tensor<string, []>("op_61399_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_61400_cast_fp16 = softmax(axis = var_58034, x = aw_5191_cast_fp16)[name = tensor<string, []>("op_61400_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_61401_cast_fp16 = softmax(axis = var_58034, x = aw_5193_cast_fp16)[name = tensor<string, []>("op_61401_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_61402_cast_fp16 = softmax(axis = var_58034, x = aw_5195_cast_fp16)[name = tensor<string, []>("op_61402_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_61403_cast_fp16 = softmax(axis = var_58034, x = aw_5197_cast_fp16)[name = tensor<string, []>("op_61403_cast_fp16")];
            tensor<fp16, [2, 77, 1, 4096]> var_61404_cast_fp16 = softmax(axis = var_58034, x = aw_cast_fp16)[name = tensor<string, []>("op_61404_cast_fp16")];
            tensor<string, []> var_61406_equation_0 = const()[name = tensor<string, []>("op_61406_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_61406_cast_fp16 = einsum(equation = var_61406_equation_0, values = (var_61316_cast_fp16, var_61395_cast_fp16))[name = tensor<string, []>("op_61406_cast_fp16")];
            tensor<string, []> var_61408_equation_0 = const()[name = tensor<string, []>("op_61408_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_61408_cast_fp16 = einsum(equation = var_61408_equation_0, values = (var_61320_cast_fp16, var_61396_cast_fp16))[name = tensor<string, []>("op_61408_cast_fp16")];
            tensor<string, []> var_61410_equation_0 = const()[name = tensor<string, []>("op_61410_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_61410_cast_fp16 = einsum(equation = var_61410_equation_0, values = (var_61324_cast_fp16, var_61397_cast_fp16))[name = tensor<string, []>("op_61410_cast_fp16")];
            tensor<string, []> var_61412_equation_0 = const()[name = tensor<string, []>("op_61412_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_61412_cast_fp16 = einsum(equation = var_61412_equation_0, values = (var_61328_cast_fp16, var_61398_cast_fp16))[name = tensor<string, []>("op_61412_cast_fp16")];
            tensor<string, []> var_61414_equation_0 = const()[name = tensor<string, []>("op_61414_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_61414_cast_fp16 = einsum(equation = var_61414_equation_0, values = (var_61332_cast_fp16, var_61399_cast_fp16))[name = tensor<string, []>("op_61414_cast_fp16")];
            tensor<string, []> var_61416_equation_0 = const()[name = tensor<string, []>("op_61416_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_61416_cast_fp16 = einsum(equation = var_61416_equation_0, values = (var_61336_cast_fp16, var_61400_cast_fp16))[name = tensor<string, []>("op_61416_cast_fp16")];
            tensor<string, []> var_61418_equation_0 = const()[name = tensor<string, []>("op_61418_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_61418_cast_fp16 = einsum(equation = var_61418_equation_0, values = (var_61340_cast_fp16, var_61401_cast_fp16))[name = tensor<string, []>("op_61418_cast_fp16")];
            tensor<string, []> var_61420_equation_0 = const()[name = tensor<string, []>("op_61420_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_61420_cast_fp16 = einsum(equation = var_61420_equation_0, values = (var_61344_cast_fp16, var_61402_cast_fp16))[name = tensor<string, []>("op_61420_cast_fp16")];
            tensor<string, []> var_61422_equation_0 = const()[name = tensor<string, []>("op_61422_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_61422_cast_fp16 = einsum(equation = var_61422_equation_0, values = (var_61348_cast_fp16, var_61403_cast_fp16))[name = tensor<string, []>("op_61422_cast_fp16")];
            tensor<string, []> var_61424_equation_0 = const()[name = tensor<string, []>("op_61424_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_61424_cast_fp16 = einsum(equation = var_61424_equation_0, values = (var_61352_cast_fp16, var_61404_cast_fp16))[name = tensor<string, []>("op_61424_cast_fp16")];
            tensor<bool, []> input_815_interleave_0 = const()[name = tensor<string, []>("input_815_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 640, 1, 4096]> input_815_cast_fp16 = concat(axis = var_58034, interleave = input_815_interleave_0, values = (var_61406_cast_fp16, var_61408_cast_fp16, var_61410_cast_fp16, var_61412_cast_fp16, var_61414_cast_fp16, var_61416_cast_fp16, var_61418_cast_fp16, var_61420_cast_fp16, var_61422_cast_fp16, var_61424_cast_fp16))[name = tensor<string, []>("input_815_cast_fp16")];
            tensor<string, []> var_61434_pad_type_0 = const()[name = tensor<string, []>("op_61434_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_61434_strides_0 = const()[name = tensor<string, []>("op_61434_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_61434_pad_0 = const()[name = tensor<string, []>("op_61434_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_61434_dilations_0 = const()[name = tensor<string, []>("op_61434_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_61434_groups_0 = const()[name = tensor<string, []>("op_61434_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_2_transformer_blocks_1_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(944329792))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(944637056))), name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_1_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            tensor<fp16, [640]> up_blocks_1_attentions_2_transformer_blocks_1_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_1_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(944637248)))];
            tensor<fp16, [2, 640, 1, 4096]> var_61434_cast_fp16 = conv(bias = up_blocks_1_attentions_2_transformer_blocks_1_attn2_to_out_0_bias_to_fp16, dilations = var_61434_dilations_0, groups = var_61434_groups_0, pad = var_61434_pad_0, pad_type = var_61434_pad_type_0, strides = var_61434_strides_0, weight = up_blocks_1_attentions_2_transformer_blocks_1_attn2_to_out_0_weight_to_fp16_palettized, x = input_815_cast_fp16)[name = tensor<string, []>("op_61434_cast_fp16")];
            tensor<fp16, [2, 640, 1, 4096]> inputs_cast_fp16 = add(x = var_61434_cast_fp16, y = inputs_417_cast_fp16)[name = tensor<string, []>("inputs_cast_fp16")];
            tensor<int32, [1]> input_817_axes_0 = const()[name = tensor<string, []>("input_817_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [640]> input_817_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_817_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(944638592)))];
            tensor<fp16, [640]> input_817_beta_0_to_fp16 = const()[name = tensor<string, []>("input_817_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(944639936)))];
            tensor<fp16, []> var_61444_to_fp16 = const()[name = tensor<string, []>("op_61444_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 640, 1, 4096]> input_817_cast_fp16 = layer_norm(axes = input_817_axes_0, beta = input_817_beta_0_to_fp16, epsilon = var_61444_to_fp16, gamma = input_817_gamma_0_to_fp16, x = inputs_cast_fp16)[name = tensor<string, []>("input_817_cast_fp16")];
            tensor<string, []> var_61464_pad_type_0 = const()[name = tensor<string, []>("op_61464_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_61464_strides_0 = const()[name = tensor<string, []>("op_61464_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_61464_pad_0 = const()[name = tensor<string, []>("op_61464_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_61464_dilations_0 = const()[name = tensor<string, []>("op_61464_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_61464_groups_0 = const()[name = tensor<string, []>("op_61464_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [5120, 640, 1, 1]> up_blocks_1_attentions_2_transformer_blocks_1_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2457600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(944641280))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(947098944))), name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_1_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([5120, 640, 1, 1])];
            tensor<fp16, [5120]> up_blocks_1_attentions_2_transformer_blocks_1_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_1_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(947099136)))];
            tensor<fp16, [2, 5120, 1, 4096]> var_61464_cast_fp16 = conv(bias = up_blocks_1_attentions_2_transformer_blocks_1_ff_net_0_proj_bias_to_fp16, dilations = var_61464_dilations_0, groups = var_61464_groups_0, pad = var_61464_pad_0, pad_type = var_61464_pad_type_0, strides = var_61464_strides_0, weight = up_blocks_1_attentions_2_transformer_blocks_1_ff_net_0_proj_weight_to_fp16_palettized, x = input_817_cast_fp16)[name = tensor<string, []>("op_61464_cast_fp16")];
            tensor<int32, [2]> var_61465_split_sizes_0 = const()[name = tensor<string, []>("op_61465_split_sizes_0"), val = tensor<int32, [2]>([2560, 2560])];
            tensor<int32, []> var_61465_axis_0 = const()[name = tensor<string, []>("op_61465_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 2560, 1, 4096]> var_61465_cast_fp16_0, tensor<fp16, [2, 2560, 1, 4096]> var_61465_cast_fp16_1 = split(axis = var_61465_axis_0, split_sizes = var_61465_split_sizes_0, x = var_61464_cast_fp16)[name = tensor<string, []>("op_61465_cast_fp16")];
            tensor<string, []> var_61467_mode_0 = const()[name = tensor<string, []>("op_61467_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 2560, 1, 4096]> var_61467_cast_fp16 = gelu(mode = var_61467_mode_0, x = var_61465_cast_fp16_1)[name = tensor<string, []>("op_61467_cast_fp16")];
            tensor<fp16, [2, 2560, 1, 4096]> input_819_cast_fp16 = mul(x = var_61465_cast_fp16_0, y = var_61467_cast_fp16)[name = tensor<string, []>("input_819_cast_fp16")];
            tensor<string, []> var_61475_pad_type_0 = const()[name = tensor<string, []>("op_61475_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_61475_strides_0 = const()[name = tensor<string, []>("op_61475_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_61475_pad_0 = const()[name = tensor<string, []>("op_61475_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_61475_dilations_0 = const()[name = tensor<string, []>("op_61475_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_61475_groups_0 = const()[name = tensor<string, []>("op_61475_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 2560, 1, 1]> up_blocks_1_attentions_2_transformer_blocks_1_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(947109440))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(948338304))), name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_1_ff_net_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 2560, 1, 1])];
            tensor<fp16, [640]> up_blocks_1_attentions_2_transformer_blocks_1_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_1_ff_net_2_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(948338496)))];
            tensor<fp16, [2, 640, 1, 4096]> var_61475_cast_fp16 = conv(bias = up_blocks_1_attentions_2_transformer_blocks_1_ff_net_2_bias_to_fp16, dilations = var_61475_dilations_0, groups = var_61475_groups_0, pad = var_61475_pad_0, pad_type = var_61475_pad_type_0, strides = var_61475_strides_0, weight = up_blocks_1_attentions_2_transformer_blocks_1_ff_net_2_weight_to_fp16_palettized, x = input_819_cast_fp16)[name = tensor<string, []>("op_61475_cast_fp16")];
            tensor<fp16, [2, 640, 1, 4096]> hidden_states_573_cast_fp16 = add(x = var_61475_cast_fp16, y = inputs_cast_fp16)[name = tensor<string, []>("hidden_states_573_cast_fp16")];
            tensor<int32, [4]> var_61477 = const()[name = tensor<string, []>("op_61477"), val = tensor<int32, [4]>([2, 640, 64, 64])];
            tensor<fp16, [2, 640, 64, 64]> input_821_cast_fp16 = reshape(shape = var_61477, x = hidden_states_573_cast_fp16)[name = tensor<string, []>("input_821_cast_fp16")];
            tensor<string, []> hidden_states_575_pad_type_0 = const()[name = tensor<string, []>("hidden_states_575_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> hidden_states_575_strides_0 = const()[name = tensor<string, []>("hidden_states_575_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_575_pad_0 = const()[name = tensor<string, []>("hidden_states_575_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_575_dilations_0 = const()[name = tensor<string, []>("hidden_states_575_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_575_groups_0 = const()[name = tensor<string, []>("hidden_states_575_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_2_proj_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(948339840))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(948647104))), name = tensor<string, []>("up_blocks_1_attentions_2_proj_out_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            tensor<fp16, [640]> 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, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(948647296)))];
            tensor<fp16, [2, 640, 64, 64]> hidden_states_575_cast_fp16 = conv(bias = up_blocks_1_attentions_2_proj_out_bias_to_fp16, dilations = hidden_states_575_dilations_0, groups = hidden_states_575_groups_0, pad = hidden_states_575_pad_0, pad_type = hidden_states_575_pad_type_0, strides = hidden_states_575_strides_0, weight = up_blocks_1_attentions_2_proj_out_weight_to_fp16_palettized, x = input_821_cast_fp16)[name = tensor<string, []>("hidden_states_575_cast_fp16")];
            tensor<fp16, [2, 640, 64, 64]> input_823_cast_fp16 = add(x = hidden_states_575_cast_fp16, y = hidden_states_557_cast_fp16)[name = tensor<string, []>("input_823_cast_fp16")];
            tensor<fp32, []> input_825_scale_factor_height_0 = const()[name = tensor<string, []>("input_825_scale_factor_height_0"), val = tensor<fp32, []>(0x1p+1)];
            tensor<fp32, []> input_825_scale_factor_width_0 = const()[name = tensor<string, []>("input_825_scale_factor_width_0"), val = tensor<fp32, []>(0x1p+1)];
            tensor<fp16, [2, 640, 128, 128]> input_825_cast_fp16 = upsample_nearest_neighbor(scale_factor_height = input_825_scale_factor_height_0, scale_factor_width = input_825_scale_factor_width_0, x = input_823_cast_fp16)[name = tensor<string, []>("input_825_cast_fp16")];
            tensor<string, []> hidden_states_577_pad_type_0 = const()[name = tensor<string, []>("hidden_states_577_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_577_pad_0 = const()[name = tensor<string, []>("hidden_states_577_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> hidden_states_577_strides_0 = const()[name = tensor<string, []>("hidden_states_577_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> hidden_states_577_dilations_0 = const()[name = tensor<string, []>("hidden_states_577_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_577_groups_0 = const()[name = tensor<string, []>("hidden_states_577_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [640, 640, 3, 3]> up_blocks_1_upsamplers_0_conv_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2764800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(948648640))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(951413504))), name = tensor<string, []>("up_blocks_1_upsamplers_0_conv_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 3, 3])];
            tensor<fp16, [640]> 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, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(951413696)))];
            tensor<fp16, [2, 640, 128, 128]> hidden_states_577_cast_fp16 = conv(bias = up_blocks_1_upsamplers_0_conv_bias_to_fp16, dilations = hidden_states_577_dilations_0, groups = hidden_states_577_groups_0, pad = hidden_states_577_pad_0, pad_type = hidden_states_577_pad_type_0, strides = hidden_states_577_strides_0, weight = up_blocks_1_upsamplers_0_conv_weight_to_fp16_palettized, x = input_825_cast_fp16)[name = tensor<string, []>("hidden_states_577_cast_fp16")];
            tensor<int32, []> var_61502 = const()[name = tensor<string, []>("op_61502"), val = tensor<int32, []>(1)];
            tensor<bool, []> input_827_interleave_0 = const()[name = tensor<string, []>("input_827_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 320, 128, 128]> cast_3 = cast(dtype = cast_3_dtype_0, x = input_43_cast_fp16)[name = tensor<string, []>("cast_8")];
            tensor<fp16, [2, 960, 128, 128]> input_827_cast_fp16 = concat(axis = var_61502, interleave = input_827_interleave_0, values = (hidden_states_577_cast_fp16, cast_3))[name = tensor<string, []>("input_827_cast_fp16")];
            tensor<int32, [5]> reshape_156_shape_0 = const()[name = tensor<string, []>("reshape_156_shape_0"), val = tensor<int32, [5]>([2, 32, 30, 128, 128])];
            tensor<fp16, [2, 32, 30, 128, 128]> reshape_156_cast_fp16 = reshape(shape = reshape_156_shape_0, x = input_827_cast_fp16)[name = tensor<string, []>("reshape_156_cast_fp16")];
            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_fp16 = reduce_mean(axes = reduce_mean_117_axes_0, keep_dims = reduce_mean_117_keep_dims_0, x = reshape_156_cast_fp16)[name = tensor<string, []>("reduce_mean_117_cast_fp16")];
            tensor<fp16, [2, 32, 30, 128, 128]> sub_78_cast_fp16 = sub(x = reshape_156_cast_fp16, y = reduce_mean_117_cast_fp16)[name = tensor<string, []>("sub_78_cast_fp16")];
            tensor<fp16, [2, 32, 30, 128, 128]> square_39_cast_fp16 = square(x = sub_78_cast_fp16)[name = tensor<string, []>("square_39_cast_fp16")];
            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_fp16 = reduce_mean(axes = reduce_mean_119_axes_0, keep_dims = reduce_mean_119_keep_dims_0, x = square_39_cast_fp16)[name = tensor<string, []>("reduce_mean_119_cast_fp16")];
            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_fp16 = add(x = reduce_mean_119_cast_fp16, y = add_78_y_0_to_fp16)[name = tensor<string, []>("add_78_cast_fp16")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_39_cast_fp16 = sqrt(x = add_78_cast_fp16)[name = tensor<string, []>("sqrt_39_cast_fp16")];
            tensor<fp16, [2, 32, 30, 128, 128]> real_div_39_cast_fp16 = real_div(x = sub_78_cast_fp16, y = sqrt_39_cast_fp16)[name = tensor<string, []>("real_div_39_cast_fp16")];
            tensor<int32, [4]> reshape_157_shape_0 = const()[name = tensor<string, []>("reshape_157_shape_0"), val = tensor<int32, [4]>([2, 960, 128, 128])];
            tensor<fp16, [2, 960, 128, 128]> reshape_157_cast_fp16 = reshape(shape = reshape_157_shape_0, x = real_div_39_cast_fp16)[name = tensor<string, []>("reshape_157_cast_fp16")];
            tensor<fp16, [960]> add_79_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_79_gamma_0_to_fp16"), val = tensor<fp16, [960]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(951415040)))];
            tensor<fp16, [960]> add_79_beta_0_to_fp16 = const()[name = tensor<string, []>("add_79_beta_0_to_fp16"), val = tensor<fp16, [960]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(951417024)))];
            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, 960, 128, 128]> add_79_cast_fp16 = 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_73_mean_0_to_fp16, variance = add_73_variance_0_to_fp16, x = reshape_157_cast_fp16)[name = tensor<string, []>("add_79_cast_fp16")];
            tensor<fp16, [2, 960, 128, 128]> input_831_cast_fp16 = silu(x = add_79_cast_fp16)[name = tensor<string, []>("input_831_cast_fp16")];
            tensor<string, []> hidden_states_579_pad_type_0 = const()[name = tensor<string, []>("hidden_states_579_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_579_pad_0 = const()[name = tensor<string, []>("hidden_states_579_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> hidden_states_579_strides_0 = const()[name = tensor<string, []>("hidden_states_579_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> hidden_states_579_dilations_0 = const()[name = tensor<string, []>("hidden_states_579_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_579_groups_0 = const()[name = tensor<string, []>("hidden_states_579_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [320, 960, 3, 3]> up_blocks_2_resnets_0_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2073600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(951419008))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(953492672))), name = tensor<string, []>("up_blocks_2_resnets_0_conv1_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 960, 3, 3])];
            tensor<fp16, [320]> 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, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(953492864)))];
            tensor<fp16, [2, 320, 128, 128]> hidden_states_579_cast_fp16 = conv(bias = up_blocks_2_resnets_0_conv1_bias_to_fp16, dilations = hidden_states_579_dilations_0, groups = hidden_states_579_groups_0, pad = hidden_states_579_pad_0, pad_type = hidden_states_579_pad_type_0, strides = hidden_states_579_strides_0, weight = up_blocks_2_resnets_0_conv1_weight_to_fp16_palettized, x = input_831_cast_fp16)[name = tensor<string, []>("hidden_states_579_cast_fp16")];
            tensor<string, []> temb_29_pad_type_0 = const()[name = tensor<string, []>("temb_29_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> temb_29_strides_0 = const()[name = tensor<string, []>("temb_29_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> temb_29_pad_0 = const()[name = tensor<string, []>("temb_29_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> temb_29_dilations_0 = const()[name = tensor<string, []>("temb_29_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> temb_29_groups_0 = const()[name = tensor<string, []>("temb_29_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [320, 1280, 1, 1]> up_blocks_2_resnets_0_time_emb_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(953493568))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(953800832))), name = tensor<string, []>("up_blocks_2_resnets_0_time_emb_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 1280, 1, 1])];
            tensor<fp16, [320]> 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, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(953801024)))];
            tensor<fp16, [2, 320, 1, 1]> temb_29_cast_fp16 = conv(bias = up_blocks_2_resnets_0_time_emb_proj_bias_to_fp16, dilations = temb_29_dilations_0, groups = temb_29_groups_0, pad = temb_29_pad_0, pad_type = temb_29_pad_type_0, strides = temb_29_strides_0, weight = up_blocks_2_resnets_0_time_emb_proj_weight_to_fp16_palettized, x = cast_2)[name = tensor<string, []>("temb_29_cast_fp16")];
            tensor<fp16, [2, 320, 128, 128]> input_835_cast_fp16 = add(x = hidden_states_579_cast_fp16, y = temb_29_cast_fp16)[name = tensor<string, []>("input_835_cast_fp16")];
            tensor<int32, [5]> reshape_160_shape_0 = const()[name = tensor<string, []>("reshape_160_shape_0"), val = tensor<int32, [5]>([2, 32, 10, 128, 128])];
            tensor<fp16, [2, 32, 10, 128, 128]> reshape_160_cast_fp16 = reshape(shape = reshape_160_shape_0, x = input_835_cast_fp16)[name = tensor<string, []>("reshape_160_cast_fp16")];
            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_fp16 = reduce_mean(axes = reduce_mean_120_axes_0, keep_dims = reduce_mean_120_keep_dims_0, x = reshape_160_cast_fp16)[name = tensor<string, []>("reduce_mean_120_cast_fp16")];
            tensor<fp16, [2, 32, 10, 128, 128]> sub_80_cast_fp16 = sub(x = reshape_160_cast_fp16, y = reduce_mean_120_cast_fp16)[name = tensor<string, []>("sub_80_cast_fp16")];
            tensor<fp16, [2, 32, 10, 128, 128]> square_40_cast_fp16 = square(x = sub_80_cast_fp16)[name = tensor<string, []>("square_40_cast_fp16")];
            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_fp16 = reduce_mean(axes = reduce_mean_122_axes_0, keep_dims = reduce_mean_122_keep_dims_0, x = square_40_cast_fp16)[name = tensor<string, []>("reduce_mean_122_cast_fp16")];
            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_fp16 = add(x = reduce_mean_122_cast_fp16, y = add_80_y_0_to_fp16)[name = tensor<string, []>("add_80_cast_fp16")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_40_cast_fp16 = sqrt(x = add_80_cast_fp16)[name = tensor<string, []>("sqrt_40_cast_fp16")];
            tensor<fp16, [2, 32, 10, 128, 128]> real_div_40_cast_fp16 = real_div(x = sub_80_cast_fp16, y = sqrt_40_cast_fp16)[name = tensor<string, []>("real_div_40_cast_fp16")];
            tensor<int32, [4]> reshape_161_shape_0 = const()[name = tensor<string, []>("reshape_161_shape_0"), val = tensor<int32, [4]>([2, 320, 128, 128])];
            tensor<fp16, [2, 320, 128, 128]> reshape_161_cast_fp16 = reshape(shape = reshape_161_shape_0, x = real_div_40_cast_fp16)[name = tensor<string, []>("reshape_161_cast_fp16")];
            tensor<fp16, [320]> add_81_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_81_gamma_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(953801728)))];
            tensor<fp16, [320]> add_81_beta_0_to_fp16 = const()[name = tensor<string, []>("add_81_beta_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(953802432)))];
            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, 320, 128, 128]> add_81_cast_fp16 = 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_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_161_cast_fp16)[name = tensor<string, []>("add_81_cast_fp16")];
            tensor<fp16, [2, 320, 128, 128]> input_839_cast_fp16 = silu(x = add_81_cast_fp16)[name = tensor<string, []>("input_839_cast_fp16")];
            tensor<string, []> hidden_states_581_pad_type_0 = const()[name = tensor<string, []>("hidden_states_581_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_581_pad_0 = const()[name = tensor<string, []>("hidden_states_581_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> hidden_states_581_strides_0 = const()[name = tensor<string, []>("hidden_states_581_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> hidden_states_581_dilations_0 = const()[name = tensor<string, []>("hidden_states_581_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_581_groups_0 = const()[name = tensor<string, []>("hidden_states_581_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [320, 320, 3, 3]> up_blocks_2_resnets_0_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [691200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(953803136))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(954494400))), name = tensor<string, []>("up_blocks_2_resnets_0_conv2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 320, 3, 3])];
            tensor<fp16, [320]> 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, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(954494592)))];
            tensor<fp16, [2, 320, 128, 128]> hidden_states_581_cast_fp16 = conv(bias = up_blocks_2_resnets_0_conv2_bias_to_fp16, dilations = hidden_states_581_dilations_0, groups = hidden_states_581_groups_0, pad = hidden_states_581_pad_0, pad_type = hidden_states_581_pad_type_0, strides = hidden_states_581_strides_0, weight = up_blocks_2_resnets_0_conv2_weight_to_fp16_palettized, x = input_839_cast_fp16)[name = tensor<string, []>("hidden_states_581_cast_fp16")];
            tensor<string, []> x_17_pad_type_0 = const()[name = tensor<string, []>("x_17_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> x_17_strides_0 = const()[name = tensor<string, []>("x_17_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> x_17_pad_0 = const()[name = tensor<string, []>("x_17_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> x_17_dilations_0 = const()[name = tensor<string, []>("x_17_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> x_17_groups_0 = const()[name = tensor<string, []>("x_17_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [320, 960, 1, 1]> up_blocks_2_resnets_0_conv_shortcut_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [230400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(954495296))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(954725760))), name = tensor<string, []>("up_blocks_2_resnets_0_conv_shortcut_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 960, 1, 1])];
            tensor<fp16, [320]> 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, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(954725952)))];
            tensor<fp16, [2, 320, 128, 128]> x_17_cast_fp16 = conv(bias = up_blocks_2_resnets_0_conv_shortcut_bias_to_fp16, dilations = x_17_dilations_0, groups = x_17_groups_0, pad = x_17_pad_0, pad_type = x_17_pad_type_0, strides = x_17_strides_0, weight = up_blocks_2_resnets_0_conv_shortcut_weight_to_fp16_palettized, x = input_827_cast_fp16)[name = tensor<string, []>("x_17_cast_fp16")];
            tensor<fp16, [2, 320, 128, 128]> hidden_states_583_cast_fp16 = add(x = x_17_cast_fp16, y = hidden_states_581_cast_fp16)[name = tensor<string, []>("hidden_states_583_cast_fp16")];
            tensor<bool, []> input_841_interleave_0 = const()[name = tensor<string, []>("input_841_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 320, 128, 128]> cast_4 = cast(dtype = cast_4_dtype_0, x = input_29_cast_fp16)[name = tensor<string, []>("cast_9")];
            tensor<fp16, [2, 640, 128, 128]> input_841_cast_fp16 = concat(axis = var_61502, interleave = input_841_interleave_0, values = (hidden_states_583_cast_fp16, cast_4))[name = tensor<string, []>("input_841_cast_fp16")];
            tensor<int32, [5]> reshape_164_shape_0 = const()[name = tensor<string, []>("reshape_164_shape_0"), val = tensor<int32, [5]>([2, 32, 20, 128, 128])];
            tensor<fp16, [2, 32, 20, 128, 128]> reshape_164_cast_fp16 = reshape(shape = reshape_164_shape_0, x = input_841_cast_fp16)[name = tensor<string, []>("reshape_164_cast_fp16")];
            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_fp16 = reduce_mean(axes = reduce_mean_123_axes_0, keep_dims = reduce_mean_123_keep_dims_0, x = reshape_164_cast_fp16)[name = tensor<string, []>("reduce_mean_123_cast_fp16")];
            tensor<fp16, [2, 32, 20, 128, 128]> sub_82_cast_fp16 = sub(x = reshape_164_cast_fp16, y = reduce_mean_123_cast_fp16)[name = tensor<string, []>("sub_82_cast_fp16")];
            tensor<fp16, [2, 32, 20, 128, 128]> square_41_cast_fp16 = square(x = sub_82_cast_fp16)[name = tensor<string, []>("square_41_cast_fp16")];
            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_fp16 = reduce_mean(axes = reduce_mean_125_axes_0, keep_dims = reduce_mean_125_keep_dims_0, x = square_41_cast_fp16)[name = tensor<string, []>("reduce_mean_125_cast_fp16")];
            tensor<fp16, []> add_82_y_0_to_fp16 = const()[name = tensor<string, []>("add_82_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_82_cast_fp16 = add(x = reduce_mean_125_cast_fp16, y = add_82_y_0_to_fp16)[name = tensor<string, []>("add_82_cast_fp16")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_41_cast_fp16 = sqrt(x = add_82_cast_fp16)[name = tensor<string, []>("sqrt_41_cast_fp16")];
            tensor<fp16, [2, 32, 20, 128, 128]> real_div_41_cast_fp16 = real_div(x = sub_82_cast_fp16, y = sqrt_41_cast_fp16)[name = tensor<string, []>("real_div_41_cast_fp16")];
            tensor<int32, [4]> reshape_165_shape_0 = const()[name = tensor<string, []>("reshape_165_shape_0"), val = tensor<int32, [4]>([2, 640, 128, 128])];
            tensor<fp16, [2, 640, 128, 128]> reshape_165_cast_fp16 = reshape(shape = reshape_165_shape_0, x = real_div_41_cast_fp16)[name = tensor<string, []>("reshape_165_cast_fp16")];
            tensor<fp16, [640]> add_83_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_83_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(954726656)))];
            tensor<fp16, [640]> add_83_beta_0_to_fp16 = const()[name = tensor<string, []>("add_83_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(954728000)))];
            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, 640, 128, 128]> add_83_cast_fp16 = 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_11_mean_0_to_fp16, variance = add_11_variance_0_to_fp16, x = reshape_165_cast_fp16)[name = tensor<string, []>("add_83_cast_fp16")];
            tensor<fp16, [2, 640, 128, 128]> input_845_cast_fp16 = silu(x = add_83_cast_fp16)[name = tensor<string, []>("input_845_cast_fp16")];
            tensor<string, []> hidden_states_585_pad_type_0 = const()[name = tensor<string, []>("hidden_states_585_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_585_pad_0 = const()[name = tensor<string, []>("hidden_states_585_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> hidden_states_585_strides_0 = const()[name = tensor<string, []>("hidden_states_585_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> hidden_states_585_dilations_0 = const()[name = tensor<string, []>("hidden_states_585_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_585_groups_0 = const()[name = tensor<string, []>("hidden_states_585_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [320, 640, 3, 3]> up_blocks_2_resnets_1_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1382400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(954729344))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(956111808))), name = tensor<string, []>("up_blocks_2_resnets_1_conv1_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 640, 3, 3])];
            tensor<fp16, [320]> 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, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(956112000)))];
            tensor<fp16, [2, 320, 128, 128]> hidden_states_585_cast_fp16 = conv(bias = up_blocks_2_resnets_1_conv1_bias_to_fp16, dilations = hidden_states_585_dilations_0, groups = hidden_states_585_groups_0, pad = hidden_states_585_pad_0, pad_type = hidden_states_585_pad_type_0, strides = hidden_states_585_strides_0, weight = up_blocks_2_resnets_1_conv1_weight_to_fp16_palettized, x = input_845_cast_fp16)[name = tensor<string, []>("hidden_states_585_cast_fp16")];
            tensor<string, []> temb_31_pad_type_0 = const()[name = tensor<string, []>("temb_31_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> temb_31_strides_0 = const()[name = tensor<string, []>("temb_31_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> temb_31_pad_0 = const()[name = tensor<string, []>("temb_31_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> temb_31_dilations_0 = const()[name = tensor<string, []>("temb_31_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> temb_31_groups_0 = const()[name = tensor<string, []>("temb_31_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [320, 1280, 1, 1]> up_blocks_2_resnets_1_time_emb_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(956112704))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(956419968))), name = tensor<string, []>("up_blocks_2_resnets_1_time_emb_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 1280, 1, 1])];
            tensor<fp16, [320]> 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, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(956420160)))];
            tensor<fp16, [2, 320, 1, 1]> temb_31_cast_fp16 = conv(bias = up_blocks_2_resnets_1_time_emb_proj_bias_to_fp16, dilations = temb_31_dilations_0, groups = temb_31_groups_0, pad = temb_31_pad_0, pad_type = temb_31_pad_type_0, strides = temb_31_strides_0, weight = up_blocks_2_resnets_1_time_emb_proj_weight_to_fp16_palettized, x = cast_2)[name = tensor<string, []>("temb_31_cast_fp16")];
            tensor<fp16, [2, 320, 128, 128]> input_849_cast_fp16 = add(x = hidden_states_585_cast_fp16, y = temb_31_cast_fp16)[name = tensor<string, []>("input_849_cast_fp16")];
            tensor<int32, [5]> reshape_168_shape_0 = const()[name = tensor<string, []>("reshape_168_shape_0"), val = tensor<int32, [5]>([2, 32, 10, 128, 128])];
            tensor<fp16, [2, 32, 10, 128, 128]> reshape_168_cast_fp16 = reshape(shape = reshape_168_shape_0, x = input_849_cast_fp16)[name = tensor<string, []>("reshape_168_cast_fp16")];
            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_fp16 = reduce_mean(axes = reduce_mean_126_axes_0, keep_dims = reduce_mean_126_keep_dims_0, x = reshape_168_cast_fp16)[name = tensor<string, []>("reduce_mean_126_cast_fp16")];
            tensor<fp16, [2, 32, 10, 128, 128]> sub_84_cast_fp16 = sub(x = reshape_168_cast_fp16, y = reduce_mean_126_cast_fp16)[name = tensor<string, []>("sub_84_cast_fp16")];
            tensor<fp16, [2, 32, 10, 128, 128]> square_42_cast_fp16 = square(x = sub_84_cast_fp16)[name = tensor<string, []>("square_42_cast_fp16")];
            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_fp16 = reduce_mean(axes = reduce_mean_128_axes_0, keep_dims = reduce_mean_128_keep_dims_0, x = square_42_cast_fp16)[name = tensor<string, []>("reduce_mean_128_cast_fp16")];
            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_fp16 = add(x = reduce_mean_128_cast_fp16, y = add_84_y_0_to_fp16)[name = tensor<string, []>("add_84_cast_fp16")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_42_cast_fp16 = sqrt(x = add_84_cast_fp16)[name = tensor<string, []>("sqrt_42_cast_fp16")];
            tensor<fp16, [2, 32, 10, 128, 128]> real_div_42_cast_fp16 = real_div(x = sub_84_cast_fp16, y = sqrt_42_cast_fp16)[name = tensor<string, []>("real_div_42_cast_fp16")];
            tensor<int32, [4]> reshape_169_shape_0 = const()[name = tensor<string, []>("reshape_169_shape_0"), val = tensor<int32, [4]>([2, 320, 128, 128])];
            tensor<fp16, [2, 320, 128, 128]> reshape_169_cast_fp16 = reshape(shape = reshape_169_shape_0, x = real_div_42_cast_fp16)[name = tensor<string, []>("reshape_169_cast_fp16")];
            tensor<fp16, [320]> add_85_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_85_gamma_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(956420864)))];
            tensor<fp16, [320]> add_85_beta_0_to_fp16 = const()[name = tensor<string, []>("add_85_beta_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(956421568)))];
            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, 320, 128, 128]> add_85_cast_fp16 = 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_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_169_cast_fp16)[name = tensor<string, []>("add_85_cast_fp16")];
            tensor<fp16, [2, 320, 128, 128]> input_853_cast_fp16 = silu(x = add_85_cast_fp16)[name = tensor<string, []>("input_853_cast_fp16")];
            tensor<string, []> hidden_states_587_pad_type_0 = const()[name = tensor<string, []>("hidden_states_587_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_587_pad_0 = const()[name = tensor<string, []>("hidden_states_587_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> hidden_states_587_strides_0 = const()[name = tensor<string, []>("hidden_states_587_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> hidden_states_587_dilations_0 = const()[name = tensor<string, []>("hidden_states_587_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_587_groups_0 = const()[name = tensor<string, []>("hidden_states_587_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [320, 320, 3, 3]> up_blocks_2_resnets_1_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [691200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(956422272))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(957113536))), name = tensor<string, []>("up_blocks_2_resnets_1_conv2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 320, 3, 3])];
            tensor<fp16, [320]> 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, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(957113728)))];
            tensor<fp16, [2, 320, 128, 128]> hidden_states_587_cast_fp16 = conv(bias = up_blocks_2_resnets_1_conv2_bias_to_fp16, dilations = hidden_states_587_dilations_0, groups = hidden_states_587_groups_0, pad = hidden_states_587_pad_0, pad_type = hidden_states_587_pad_type_0, strides = hidden_states_587_strides_0, weight = up_blocks_2_resnets_1_conv2_weight_to_fp16_palettized, x = input_853_cast_fp16)[name = tensor<string, []>("hidden_states_587_cast_fp16")];
            tensor<string, []> x_19_pad_type_0 = const()[name = tensor<string, []>("x_19_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> x_19_strides_0 = const()[name = tensor<string, []>("x_19_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> x_19_pad_0 = const()[name = tensor<string, []>("x_19_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> x_19_dilations_0 = const()[name = tensor<string, []>("x_19_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> x_19_groups_0 = const()[name = tensor<string, []>("x_19_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [320, 640, 1, 1]> up_blocks_2_resnets_1_conv_shortcut_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [153600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(957114432))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(957268096))), name = tensor<string, []>("up_blocks_2_resnets_1_conv_shortcut_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 640, 1, 1])];
            tensor<fp16, [320]> 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, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(957268288)))];
            tensor<fp16, [2, 320, 128, 128]> x_19_cast_fp16 = conv(bias = up_blocks_2_resnets_1_conv_shortcut_bias_to_fp16, dilations = x_19_dilations_0, groups = x_19_groups_0, pad = x_19_pad_0, pad_type = x_19_pad_type_0, strides = x_19_strides_0, weight = up_blocks_2_resnets_1_conv_shortcut_weight_to_fp16_palettized, x = input_841_cast_fp16)[name = tensor<string, []>("x_19_cast_fp16")];
            tensor<fp16, [2, 320, 128, 128]> hidden_states_589_cast_fp16 = add(x = x_19_cast_fp16, y = hidden_states_587_cast_fp16)[name = tensor<string, []>("hidden_states_589_cast_fp16")];
            tensor<bool, []> input_855_interleave_0 = const()[name = tensor<string, []>("input_855_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 320, 128, 128]> cast_7 = cast(dtype = cast_7_dtype_0, x = input_13_cast_fp16)[name = tensor<string, []>("cast_11")];
            tensor<fp16, [2, 640, 128, 128]> input_855_cast_fp16 = concat(axis = var_61502, interleave = input_855_interleave_0, values = (hidden_states_589_cast_fp16, cast_7))[name = tensor<string, []>("input_855_cast_fp16")];
            tensor<int32, [5]> reshape_172_shape_0 = const()[name = tensor<string, []>("reshape_172_shape_0"), val = tensor<int32, [5]>([2, 32, 20, 128, 128])];
            tensor<fp16, [2, 32, 20, 128, 128]> reshape_172_cast_fp16 = reshape(shape = reshape_172_shape_0, x = input_855_cast_fp16)[name = tensor<string, []>("reshape_172_cast_fp16")];
            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_fp16 = reduce_mean(axes = reduce_mean_129_axes_0, keep_dims = reduce_mean_129_keep_dims_0, x = reshape_172_cast_fp16)[name = tensor<string, []>("reduce_mean_129_cast_fp16")];
            tensor<fp16, [2, 32, 20, 128, 128]> sub_86_cast_fp16 = sub(x = reshape_172_cast_fp16, y = reduce_mean_129_cast_fp16)[name = tensor<string, []>("sub_86_cast_fp16")];
            tensor<fp16, [2, 32, 20, 128, 128]> square_43_cast_fp16 = square(x = sub_86_cast_fp16)[name = tensor<string, []>("square_43_cast_fp16")];
            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_fp16 = reduce_mean(axes = reduce_mean_131_axes_0, keep_dims = reduce_mean_131_keep_dims_0, x = square_43_cast_fp16)[name = tensor<string, []>("reduce_mean_131_cast_fp16")];
            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_fp16 = add(x = reduce_mean_131_cast_fp16, y = add_86_y_0_to_fp16)[name = tensor<string, []>("add_86_cast_fp16")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_43_cast_fp16 = sqrt(x = add_86_cast_fp16)[name = tensor<string, []>("sqrt_43_cast_fp16")];
            tensor<fp16, [2, 32, 20, 128, 128]> real_div_43_cast_fp16 = real_div(x = sub_86_cast_fp16, y = sqrt_43_cast_fp16)[name = tensor<string, []>("real_div_43_cast_fp16")];
            tensor<int32, [4]> reshape_173_shape_0 = const()[name = tensor<string, []>("reshape_173_shape_0"), val = tensor<int32, [4]>([2, 640, 128, 128])];
            tensor<fp16, [2, 640, 128, 128]> reshape_173_cast_fp16 = reshape(shape = reshape_173_shape_0, x = real_div_43_cast_fp16)[name = tensor<string, []>("reshape_173_cast_fp16")];
            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, []>(957268992)))];
            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, []>(957270336)))];
            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, 128, 128]> add_87_cast_fp16 = 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_11_mean_0_to_fp16, variance = add_11_variance_0_to_fp16, x = reshape_173_cast_fp16)[name = tensor<string, []>("add_87_cast_fp16")];
            tensor<fp16, [2, 640, 128, 128]> input_859_cast_fp16 = silu(x = add_87_cast_fp16)[name = tensor<string, []>("input_859_cast_fp16")];
            tensor<string, []> hidden_states_591_pad_type_0 = const()[name = tensor<string, []>("hidden_states_591_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_591_pad_0 = const()[name = tensor<string, []>("hidden_states_591_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> hidden_states_591_strides_0 = const()[name = tensor<string, []>("hidden_states_591_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> hidden_states_591_dilations_0 = const()[name = tensor<string, []>("hidden_states_591_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_591_groups_0 = const()[name = tensor<string, []>("hidden_states_591_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [320, 640, 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, []>(957271680))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(958654144))), name = tensor<string, []>("up_blocks_2_resnets_2_conv1_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 640, 3, 3])];
            tensor<fp16, [320]> 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, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(958654336)))];
            tensor<fp16, [2, 320, 128, 128]> hidden_states_591_cast_fp16 = conv(bias = up_blocks_2_resnets_2_conv1_bias_to_fp16, dilations = hidden_states_591_dilations_0, groups = hidden_states_591_groups_0, pad = hidden_states_591_pad_0, pad_type = hidden_states_591_pad_type_0, strides = hidden_states_591_strides_0, weight = up_blocks_2_resnets_2_conv1_weight_to_fp16_palettized, x = input_859_cast_fp16)[name = tensor<string, []>("hidden_states_591_cast_fp16")];
            tensor<string, []> temb_pad_type_0 = const()[name = tensor<string, []>("temb_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> temb_strides_0 = const()[name = tensor<string, []>("temb_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> temb_pad_0 = const()[name = tensor<string, []>("temb_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> temb_dilations_0 = const()[name = tensor<string, []>("temb_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> temb_groups_0 = const()[name = tensor<string, []>("temb_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [320, 1280, 1, 1]> up_blocks_2_resnets_2_time_emb_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(958655040))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(958962304))), name = tensor<string, []>("up_blocks_2_resnets_2_time_emb_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 1280, 1, 1])];
            tensor<fp16, [320]> 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, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(958962496)))];
            tensor<fp16, [2, 320, 1, 1]> temb_cast_fp16 = conv(bias = up_blocks_2_resnets_2_time_emb_proj_bias_to_fp16, dilations = temb_dilations_0, groups = temb_groups_0, pad = temb_pad_0, pad_type = temb_pad_type_0, strides = temb_strides_0, weight = up_blocks_2_resnets_2_time_emb_proj_weight_to_fp16_palettized, x = cast_2)[name = tensor<string, []>("temb_cast_fp16")];
            tensor<fp16, [2, 320, 128, 128]> input_863_cast_fp16 = add(x = hidden_states_591_cast_fp16, y = temb_cast_fp16)[name = tensor<string, []>("input_863_cast_fp16")];
            tensor<int32, [5]> reshape_176_shape_0 = const()[name = tensor<string, []>("reshape_176_shape_0"), val = tensor<int32, [5]>([2, 32, 10, 128, 128])];
            tensor<fp16, [2, 32, 10, 128, 128]> reshape_176_cast_fp16 = reshape(shape = reshape_176_shape_0, x = input_863_cast_fp16)[name = tensor<string, []>("reshape_176_cast_fp16")];
            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_fp16 = reduce_mean(axes = reduce_mean_132_axes_0, keep_dims = reduce_mean_132_keep_dims_0, x = reshape_176_cast_fp16)[name = tensor<string, []>("reduce_mean_132_cast_fp16")];
            tensor<fp16, [2, 32, 10, 128, 128]> sub_88_cast_fp16 = sub(x = reshape_176_cast_fp16, y = reduce_mean_132_cast_fp16)[name = tensor<string, []>("sub_88_cast_fp16")];
            tensor<fp16, [2, 32, 10, 128, 128]> square_44_cast_fp16 = square(x = sub_88_cast_fp16)[name = tensor<string, []>("square_44_cast_fp16")];
            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_fp16 = reduce_mean(axes = reduce_mean_134_axes_0, keep_dims = reduce_mean_134_keep_dims_0, x = square_44_cast_fp16)[name = tensor<string, []>("reduce_mean_134_cast_fp16")];
            tensor<fp16, []> add_88_y_0_to_fp16 = const()[name = tensor<string, []>("add_88_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_88_cast_fp16 = add(x = reduce_mean_134_cast_fp16, y = add_88_y_0_to_fp16)[name = tensor<string, []>("add_88_cast_fp16")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_44_cast_fp16 = sqrt(x = add_88_cast_fp16)[name = tensor<string, []>("sqrt_44_cast_fp16")];
            tensor<fp16, [2, 32, 10, 128, 128]> real_div_44_cast_fp16 = real_div(x = sub_88_cast_fp16, y = sqrt_44_cast_fp16)[name = tensor<string, []>("real_div_44_cast_fp16")];
            tensor<int32, [4]> reshape_177_shape_0 = const()[name = tensor<string, []>("reshape_177_shape_0"), val = tensor<int32, [4]>([2, 320, 128, 128])];
            tensor<fp16, [2, 320, 128, 128]> reshape_177_cast_fp16 = reshape(shape = reshape_177_shape_0, x = real_div_44_cast_fp16)[name = tensor<string, []>("reshape_177_cast_fp16")];
            tensor<fp16, [320]> add_89_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_89_gamma_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(958963200)))];
            tensor<fp16, [320]> add_89_beta_0_to_fp16 = const()[name = tensor<string, []>("add_89_beta_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(958963904)))];
            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, 320, 128, 128]> add_89_cast_fp16 = 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_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_177_cast_fp16)[name = tensor<string, []>("add_89_cast_fp16")];
            tensor<fp16, [2, 320, 128, 128]> input_867_cast_fp16 = silu(x = add_89_cast_fp16)[name = tensor<string, []>("input_867_cast_fp16")];
            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]>([1, 1, 1, 1])];
            tensor<int32, [2]> hidden_states_strides_0 = const()[name = tensor<string, []>("hidden_states_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> hidden_states_dilations_0 = const()[name = tensor<string, []>("hidden_states_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_groups_0 = const()[name = tensor<string, []>("hidden_states_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [320, 320, 3, 3]> up_blocks_2_resnets_2_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [691200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(958964608))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(959655872))), name = tensor<string, []>("up_blocks_2_resnets_2_conv2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 320, 3, 3])];
            tensor<fp16, [320]> 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, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(959656064)))];
            tensor<fp16, [2, 320, 128, 128]> hidden_states_cast_fp16 = conv(bias = up_blocks_2_resnets_2_conv2_bias_to_fp16, dilations = hidden_states_dilations_0, groups = hidden_states_groups_0, pad = hidden_states_pad_0, pad_type = hidden_states_pad_type_0, strides = hidden_states_strides_0, weight = up_blocks_2_resnets_2_conv2_weight_to_fp16_palettized, x = input_867_cast_fp16)[name = tensor<string, []>("hidden_states_cast_fp16")];
            tensor<string, []> x_pad_type_0 = const()[name = tensor<string, []>("x_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> x_strides_0 = const()[name = tensor<string, []>("x_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> x_pad_0 = const()[name = tensor<string, []>("x_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> x_dilations_0 = const()[name = tensor<string, []>("x_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> x_groups_0 = const()[name = tensor<string, []>("x_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [320, 640, 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, []>(959656768))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(959810432))), name = tensor<string, []>("up_blocks_2_resnets_2_conv_shortcut_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 640, 1, 1])];
            tensor<fp16, [320]> 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, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(959810624)))];
            tensor<fp16, [2, 320, 128, 128]> x_cast_fp16 = conv(bias = up_blocks_2_resnets_2_conv_shortcut_bias_to_fp16, dilations = x_dilations_0, groups = x_groups_0, pad = x_pad_0, pad_type = x_pad_type_0, strides = x_strides_0, weight = up_blocks_2_resnets_2_conv_shortcut_weight_to_fp16_palettized, x = input_855_cast_fp16)[name = tensor<string, []>("x_cast_fp16")];
            tensor<fp16, [2, 320, 128, 128]> input_869_cast_fp16 = add(x = x_cast_fp16, y = hidden_states_cast_fp16)[name = tensor<string, []>("input_869_cast_fp16")];
            tensor<int32, [5]> reshape_180_shape_0 = const()[name = tensor<string, []>("reshape_180_shape_0"), val = tensor<int32, [5]>([2, 32, 10, 128, 128])];
            tensor<fp16, [2, 32, 10, 128, 128]> reshape_180_cast_fp16 = reshape(shape = reshape_180_shape_0, x = input_869_cast_fp16)[name = tensor<string, []>("reshape_180_cast_fp16")];
            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_fp16 = reduce_mean(axes = reduce_mean_135_axes_0, keep_dims = reduce_mean_135_keep_dims_0, x = reshape_180_cast_fp16)[name = tensor<string, []>("reduce_mean_135_cast_fp16")];
            tensor<fp16, [2, 32, 10, 128, 128]> sub_90_cast_fp16 = sub(x = reshape_180_cast_fp16, y = reduce_mean_135_cast_fp16)[name = tensor<string, []>("sub_90_cast_fp16")];
            tensor<fp16, [2, 32, 10, 128, 128]> square_45_cast_fp16 = square(x = sub_90_cast_fp16)[name = tensor<string, []>("square_45_cast_fp16")];
            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_fp16 = reduce_mean(axes = reduce_mean_137_axes_0, keep_dims = reduce_mean_137_keep_dims_0, x = square_45_cast_fp16)[name = tensor<string, []>("reduce_mean_137_cast_fp16")];
            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_fp16 = add(x = reduce_mean_137_cast_fp16, y = add_90_y_0_to_fp16)[name = tensor<string, []>("add_90_cast_fp16")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_45_cast_fp16 = sqrt(x = add_90_cast_fp16)[name = tensor<string, []>("sqrt_45_cast_fp16")];
            tensor<fp16, [2, 32, 10, 128, 128]> real_div_45_cast_fp16 = real_div(x = sub_90_cast_fp16, y = sqrt_45_cast_fp16)[name = tensor<string, []>("real_div_45_cast_fp16")];
            tensor<int32, [4]> reshape_181_shape_0 = const()[name = tensor<string, []>("reshape_181_shape_0"), val = tensor<int32, [4]>([2, 320, 128, 128])];
            tensor<fp16, [2, 320, 128, 128]> reshape_181_cast_fp16 = reshape(shape = reshape_181_shape_0, x = real_div_45_cast_fp16)[name = tensor<string, []>("reshape_181_cast_fp16")];
            tensor<fp16, [320]> add_91_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_91_gamma_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(959811328)))];
            tensor<fp16, [320]> add_91_beta_0_to_fp16 = const()[name = tensor<string, []>("add_91_beta_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(959812032)))];
            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, 320, 128, 128]> add_91_cast_fp16 = 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_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_181_cast_fp16)[name = tensor<string, []>("add_91_cast_fp16")];
            tensor<fp16, [2, 320, 128, 128]> input_cast_fp16 = silu(x = add_91_cast_fp16)[name = tensor<string, []>("input_cast_fp16")];
            tensor<string, []> var_61667_pad_type_0 = const()[name = tensor<string, []>("op_61667_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_61667_pad_0 = const()[name = tensor<string, []>("op_61667_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> var_61667_strides_0 = const()[name = tensor<string, []>("op_61667_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_61667_dilations_0 = const()[name = tensor<string, []>("op_61667_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_61667_groups_0 = const()[name = tensor<string, []>("op_61667_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [4, 320, 3, 3]> conv_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [8640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(959812736))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(959821440))), 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]>([0x1.48p-6, -0x1.4ep-10, 0x1.96cp-10, -0x1.024p-6])];
            tensor<fp16, [2, 4, 128, 128]> var_61667_cast_fp16 = conv(bias = conv_out_bias_to_fp16, dilations = var_61667_dilations_0, groups = var_61667_groups_0, pad = var_61667_pad_0, pad_type = var_61667_pad_type_0, strides = var_61667_strides_0, weight = conv_out_weight_to_fp16_palettized, x = input_cast_fp16)[name = tensor<string, []>("op_61667_cast_fp16")];
            tensor<string, []> var_61667_cast_fp16_to_fp32_dtype_0 = const()[name = tensor<string, []>("op_61667_cast_fp16_to_fp32_dtype_0"), val = tensor<string, []>("fp32")];
            tensor<fp32, [2, 4, 128, 128]> noise_pred = cast(dtype = var_61667_cast_fp16_to_fp32_dtype_0, x = var_61667_cast_fp16)[name = tensor<string, []>("cast_0")];
        } -> (noise_pred);
}