program(1.3)
[buildInfo = dict<string, string>({{"coremlc-component-MIL", "3600.16.1"}, {"coremlc-version", "3600.22.1"}})]
{
    func main<ios18>(tensor<fp16, [1, 768, 1, 77]> encoder_hidden_states, tensor<fp16, [1, 9, 64, 64]> sample, tensor<fp16, [1]> timestep) {
            int32 var_20 = const()[name = string("op_20"), val = int32(-1)];
            tensor<int32, [1]> var_37_axes_0 = const()[name = string("op_37_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1]> var_37_cast_fp16 = expand_dims(axes = var_37_axes_0, x = timestep)[name = string("op_37_cast_fp16")];
            tensor<fp16, [1, 160]> var_39_to_fp16 = const()[name = string("op_39_to_fp16"), val = tensor<fp16, [1, 160]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))];
            tensor<fp16, [1, 160]> emb_3_cast_fp16 = mul(x = var_37_cast_fp16, y = var_39_to_fp16)[name = string("emb_3_cast_fp16")];
            tensor<fp16, [1, 160]> var_44_cast_fp16 = sin(x = emb_3_cast_fp16)[name = string("op_44_cast_fp16")];
            tensor<fp16, [1, 160]> var_45_cast_fp16 = cos(x = emb_3_cast_fp16)[name = string("op_45_cast_fp16")];
            bool emb_interleave_0 = const()[name = string("emb_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 320]> emb_cast_fp16 = concat(axis = var_20, interleave = emb_interleave_0, values = (var_44_cast_fp16, var_45_cast_fp16))[name = string("emb_cast_fp16")];
            tensor<int32, [2]> var_49_begin_0 = const()[name = string("op_49_begin_0"), val = tensor<int32, [2]>([0, 160])];
            tensor<int32, [2]> var_49_end_0 = const()[name = string("op_49_end_0"), val = tensor<int32, [2]>([1, 320])];
            tensor<bool, [2]> var_49_end_mask_0 = const()[name = string("op_49_end_mask_0"), val = tensor<bool, [2]>([true, true])];
            tensor<fp16, [1, 160]> var_49_cast_fp16 = slice_by_index(begin = var_49_begin_0, end = var_49_end_0, end_mask = var_49_end_mask_0, x = emb_cast_fp16)[name = string("op_49_cast_fp16")];
            tensor<int32, [2]> var_51_begin_0 = const()[name = string("op_51_begin_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [2]> var_51_end_0 = const()[name = string("op_51_end_0"), val = tensor<int32, [2]>([1, 160])];
            tensor<bool, [2]> var_51_end_mask_0 = const()[name = string("op_51_end_mask_0"), val = tensor<bool, [2]>([true, false])];
            tensor<fp16, [1, 160]> var_51_cast_fp16 = slice_by_index(begin = var_51_begin_0, end = var_51_end_0, end_mask = var_51_end_mask_0, x = emb_cast_fp16)[name = string("op_51_cast_fp16")];
            bool sample_interleave_0 = const()[name = string("sample_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 320]> sample_cast_fp16 = concat(axis = var_20, interleave = sample_interleave_0, values = (var_49_cast_fp16, var_51_cast_fp16))[name = string("sample_cast_fp16")];
            tensor<int32, [1]> var_61_axes_0 = const()[name = string("op_61_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 320, 1]> var_61_cast_fp16 = expand_dims(axes = var_61_axes_0, x = sample_cast_fp16)[name = string("op_61_cast_fp16")];
            tensor<int32, [1]> input_1_axes_0 = const()[name = string("input_1_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 320, 1, 1]> input_1_cast_fp16 = expand_dims(axes = input_1_axes_0, x = var_61_cast_fp16)[name = string("input_1_cast_fp16")];
            string input_3_pad_type_0 = const()[name = string("input_3_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_3_strides_0 = const()[name = string("input_3_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_3_pad_0 = const()[name = string("input_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_3_dilations_0 = const()[name = string("input_3_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_3_groups_0 = const()[name = string("input_3_groups_0"), val = int32(1)];
            tensor<fp16, [1280, 320, 1, 1]> time_embedding_linear_1_weight_to_fp16 = const()[name = string("time_embedding_linear_1_weight_to_fp16"), val = tensor<fp16, [1280, 320, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(448)))];
            tensor<fp16, [1280]> time_embedding_linear_1_bias_to_fp16 = const()[name = string("time_embedding_linear_1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(819712)))];
            tensor<fp16, [1, 1280, 1, 1]> input_3_cast_fp16 = conv(bias = time_embedding_linear_1_bias_to_fp16, dilations = input_3_dilations_0, groups = input_3_groups_0, pad = input_3_pad_0, pad_type = input_3_pad_type_0, strides = input_3_strides_0, weight = time_embedding_linear_1_weight_to_fp16, x = input_1_cast_fp16)[name = string("input_3_cast_fp16")];
            tensor<fp16, [1, 1280, 1, 1]> input_5_cast_fp16 = silu(x = input_3_cast_fp16)[name = string("input_5_cast_fp16")];
            string input_13_pad_type_0 = const()[name = string("input_13_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_13_strides_0 = const()[name = string("input_13_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_13_pad_0 = const()[name = string("input_13_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_13_dilations_0 = const()[name = string("input_13_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_13_groups_0 = const()[name = string("input_13_groups_0"), val = int32(1)];
            tensor<fp16, [1280, 1280, 1, 1]> time_embedding_linear_2_weight_to_fp16 = const()[name = string("time_embedding_linear_2_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(822336)))];
            tensor<fp16, [1280]> time_embedding_linear_2_bias_to_fp16 = const()[name = string("time_embedding_linear_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4099200)))];
            tensor<fp16, [1, 1280, 1, 1]> input_13_cast_fp16 = conv(bias = time_embedding_linear_2_bias_to_fp16, dilations = input_13_dilations_0, groups = input_13_groups_0, pad = input_13_pad_0, pad_type = input_13_pad_type_0, strides = input_13_strides_0, weight = time_embedding_linear_2_weight_to_fp16, x = input_5_cast_fp16)[name = string("input_13_cast_fp16")];
            string input_7_pad_type_0 = const()[name = string("input_7_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> input_7_pad_0 = const()[name = string("input_7_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> input_7_strides_0 = const()[name = string("input_7_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> input_7_dilations_0 = const()[name = string("input_7_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_7_groups_0 = const()[name = string("input_7_groups_0"), val = int32(1)];
            tensor<fp16, [320, 9, 3, 3]> conv_in_weight_to_fp16 = const()[name = string("conv_in_weight_to_fp16"), val = tensor<fp16, [320, 9, 3, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4101824)))];
            tensor<fp16, [320]> conv_in_bias_to_fp16 = const()[name = string("conv_in_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4153728)))];
            tensor<fp16, [1, 320, 64, 64]> input_7_cast_fp16_1 = conv(bias = conv_in_bias_to_fp16, dilations = input_7_dilations_0, groups = input_7_groups_0, pad = input_7_pad_0, pad_type = input_7_pad_type_0, strides = input_7_strides_0, weight = conv_in_weight_to_fp16, x = sample)[name = string("input_7_cast_fp16")];
            int32 var_110 = const()[name = string("op_110"), val = int32(1)];
            tensor<int32, [5]> reshape_0_shape_0 = const()[name = string("reshape_0_shape_0"), val = tensor<int32, [5]>([1, 32, 10, 64, 64])];
            tensor<fp16, [1, 32, 10, 64, 64]> reshape_0_cast_fp16 = reshape(shape = reshape_0_shape_0, x = input_7_cast_fp16_1)[name = string("reshape_0_cast_fp16")];
            tensor<int32, [3]> reduce_mean_0_axes_0 = const()[name = string("reduce_mean_0_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            bool reduce_mean_0_keep_dims_0 = const()[name = string("reduce_mean_0_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 1, 1, 1]> reduce_mean_0_cast_fp16 = reduce_mean(axes = reduce_mean_0_axes_0, keep_dims = reduce_mean_0_keep_dims_0, x = reshape_0_cast_fp16)[name = string("reduce_mean_0_cast_fp16")];
            tensor<fp16, [1, 32, 10, 64, 64]> sub_0_cast_fp16 = sub(x = reshape_0_cast_fp16, y = reduce_mean_0_cast_fp16)[name = string("sub_0_cast_fp16")];
            tensor<fp16, [1, 32, 10, 64, 64]> square_0_cast_fp16 = square(x = sub_0_cast_fp16)[name = string("square_0_cast_fp16")];
            tensor<int32, [3]> reduce_mean_2_axes_0 = const()[name = string("reduce_mean_2_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            bool reduce_mean_2_keep_dims_0 = const()[name = string("reduce_mean_2_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 1, 1, 1]> reduce_mean_2_cast_fp16 = reduce_mean(axes = reduce_mean_2_axes_0, keep_dims = reduce_mean_2_keep_dims_0, x = square_0_cast_fp16)[name = string("reduce_mean_2_cast_fp16")];
            fp16 add_0_y_0_to_fp16 = const()[name = string("add_0_y_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 32, 1, 1, 1]> add_0_cast_fp16 = add(x = reduce_mean_2_cast_fp16, y = add_0_y_0_to_fp16)[name = string("add_0_cast_fp16")];
            tensor<fp16, [1, 32, 1, 1, 1]> sqrt_0_cast_fp16 = sqrt(x = add_0_cast_fp16)[name = string("sqrt_0_cast_fp16")];
            tensor<fp16, [1, 32, 10, 64, 64]> real_div_0_cast_fp16 = real_div(x = sub_0_cast_fp16, y = sqrt_0_cast_fp16)[name = string("real_div_0_cast_fp16")];
            tensor<int32, [4]> reshape_1_shape_0 = const()[name = string("reshape_1_shape_0"), val = tensor<int32, [4]>([1, 320, 64, 64])];
            tensor<fp16, [1, 320, 64, 64]> reshape_1_cast_fp16 = reshape(shape = reshape_1_shape_0, x = real_div_0_cast_fp16)[name = string("reshape_1_cast_fp16")];
            tensor<fp16, [320]> add_1_mean_0_to_fp16 = const()[name = string("add_1_mean_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4154432)))];
            tensor<fp16, [320]> add_1_variance_0_to_fp16 = const()[name = string("add_1_variance_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4155136)))];
            tensor<fp16, [320]> add_1_gamma_0_to_fp16 = const()[name = string("add_1_gamma_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4155840)))];
            tensor<fp16, [320]> add_1_beta_0_to_fp16 = const()[name = string("add_1_beta_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4156544)))];
            fp16 add_1_epsilon_0_to_fp16 = const()[name = string("add_1_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 320, 64, 64]> add_1_cast_fp16 = batch_norm(beta = add_1_beta_0_to_fp16, epsilon = add_1_epsilon_0_to_fp16, gamma = add_1_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_1_cast_fp16)[name = string("add_1_cast_fp16")];
            tensor<fp16, [1, 320, 64, 64]> input_11_cast_fp16 = silu(x = add_1_cast_fp16)[name = string("input_11_cast_fp16")];
            string hidden_states_1_pad_type_0 = const()[name = string("hidden_states_1_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> hidden_states_1_pad_0 = const()[name = string("hidden_states_1_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> hidden_states_1_strides_0 = const()[name = string("hidden_states_1_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> hidden_states_1_dilations_0 = const()[name = string("hidden_states_1_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_1_groups_0 = const()[name = string("hidden_states_1_groups_0"), val = int32(1)];
            tensor<fp16, [320, 320, 3, 3]> down_blocks_0_resnets_0_conv1_weight_to_fp16 = const()[name = string("down_blocks_0_resnets_0_conv1_weight_to_fp16"), val = tensor<fp16, [320, 320, 3, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4157248)))];
            tensor<fp16, [320]> down_blocks_0_resnets_0_conv1_bias_to_fp16 = const()[name = string("down_blocks_0_resnets_0_conv1_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6000512)))];
            tensor<fp16, [1, 320, 64, 64]> hidden_states_1_cast_fp16 = conv(bias = down_blocks_0_resnets_0_conv1_bias_to_fp16, dilations = hidden_states_1_dilations_0, groups = hidden_states_1_groups_0, pad = hidden_states_1_pad_0, pad_type = hidden_states_1_pad_type_0, strides = hidden_states_1_strides_0, weight = down_blocks_0_resnets_0_conv1_weight_to_fp16, x = input_11_cast_fp16)[name = string("hidden_states_1_cast_fp16")];
            tensor<fp16, [1, 1280, 1, 1]> input_15_cast_fp16_1 = silu(x = input_13_cast_fp16)[name = string("input_15_cast_fp16")];
            string temb_1_pad_type_0 = const()[name = string("temb_1_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> temb_1_strides_0 = const()[name = string("temb_1_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> temb_1_pad_0 = const()[name = string("temb_1_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> temb_1_dilations_0 = const()[name = string("temb_1_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 temb_1_groups_0 = const()[name = string("temb_1_groups_0"), val = int32(1)];
            tensor<fp16, [320, 1280, 1, 1]> down_blocks_0_resnets_0_time_emb_proj_weight_to_fp16 = const()[name = string("down_blocks_0_resnets_0_time_emb_proj_weight_to_fp16"), val = tensor<fp16, [320, 1280, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6001216)))];
            tensor<fp16, [320]> down_blocks_0_resnets_0_time_emb_proj_bias_to_fp16 = const()[name = string("down_blocks_0_resnets_0_time_emb_proj_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6820480)))];
            tensor<fp16, [1, 320, 1, 1]> temb_1_cast_fp16 = conv(bias = down_blocks_0_resnets_0_time_emb_proj_bias_to_fp16, dilations = temb_1_dilations_0, groups = temb_1_groups_0, pad = temb_1_pad_0, pad_type = temb_1_pad_type_0, strides = temb_1_strides_0, weight = down_blocks_0_resnets_0_time_emb_proj_weight_to_fp16, x = input_15_cast_fp16_1)[name = string("temb_1_cast_fp16")];
            tensor<fp16, [1, 320, 64, 64]> input_17_cast_fp16 = add(x = hidden_states_1_cast_fp16, y = temb_1_cast_fp16)[name = string("input_17_cast_fp16")];
            tensor<int32, [5]> reshape_4_shape_0 = const()[name = string("reshape_4_shape_0"), val = tensor<int32, [5]>([1, 32, 10, 64, 64])];
            tensor<fp16, [1, 32, 10, 64, 64]> reshape_4_cast_fp16 = reshape(shape = reshape_4_shape_0, x = input_17_cast_fp16)[name = string("reshape_4_cast_fp16")];
            tensor<int32, [3]> reduce_mean_3_axes_0 = const()[name = string("reduce_mean_3_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            bool reduce_mean_3_keep_dims_0 = const()[name = string("reduce_mean_3_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 1, 1, 1]> reduce_mean_3_cast_fp16 = reduce_mean(axes = reduce_mean_3_axes_0, keep_dims = reduce_mean_3_keep_dims_0, x = reshape_4_cast_fp16)[name = string("reduce_mean_3_cast_fp16")];
            tensor<fp16, [1, 32, 10, 64, 64]> sub_2_cast_fp16 = sub(x = reshape_4_cast_fp16, y = reduce_mean_3_cast_fp16)[name = string("sub_2_cast_fp16")];
            tensor<fp16, [1, 32, 10, 64, 64]> square_1_cast_fp16 = square(x = sub_2_cast_fp16)[name = string("square_1_cast_fp16")];
            tensor<int32, [3]> reduce_mean_5_axes_0 = const()[name = string("reduce_mean_5_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            bool reduce_mean_5_keep_dims_0 = const()[name = string("reduce_mean_5_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 1, 1, 1]> reduce_mean_5_cast_fp16 = reduce_mean(axes = reduce_mean_5_axes_0, keep_dims = reduce_mean_5_keep_dims_0, x = square_1_cast_fp16)[name = string("reduce_mean_5_cast_fp16")];
            fp16 add_2_y_0_to_fp16 = const()[name = string("add_2_y_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 32, 1, 1, 1]> add_2_cast_fp16 = add(x = reduce_mean_5_cast_fp16, y = add_2_y_0_to_fp16)[name = string("add_2_cast_fp16")];
            tensor<fp16, [1, 32, 1, 1, 1]> sqrt_1_cast_fp16 = sqrt(x = add_2_cast_fp16)[name = string("sqrt_1_cast_fp16")];
            tensor<fp16, [1, 32, 10, 64, 64]> real_div_1_cast_fp16 = real_div(x = sub_2_cast_fp16, y = sqrt_1_cast_fp16)[name = string("real_div_1_cast_fp16")];
            tensor<int32, [4]> reshape_5_shape_0 = const()[name = string("reshape_5_shape_0"), val = tensor<int32, [4]>([1, 320, 64, 64])];
            tensor<fp16, [1, 320, 64, 64]> reshape_5_cast_fp16 = reshape(shape = reshape_5_shape_0, x = real_div_1_cast_fp16)[name = string("reshape_5_cast_fp16")];
            tensor<fp16, [320]> add_3_gamma_0_to_fp16 = const()[name = string("add_3_gamma_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6821184)))];
            tensor<fp16, [320]> add_3_beta_0_to_fp16 = const()[name = string("add_3_beta_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6821888)))];
            fp16 add_3_epsilon_0_to_fp16 = const()[name = string("add_3_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 320, 64, 64]> add_3_cast_fp16 = batch_norm(beta = add_3_beta_0_to_fp16, epsilon = add_3_epsilon_0_to_fp16, gamma = add_3_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_5_cast_fp16)[name = string("add_3_cast_fp16")];
            tensor<fp16, [1, 320, 64, 64]> input_21_cast_fp16 = silu(x = add_3_cast_fp16)[name = string("input_21_cast_fp16")];
            string hidden_states_3_pad_type_0 = const()[name = string("hidden_states_3_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> hidden_states_3_pad_0 = const()[name = string("hidden_states_3_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> hidden_states_3_strides_0 = const()[name = string("hidden_states_3_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> hidden_states_3_dilations_0 = const()[name = string("hidden_states_3_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_3_groups_0 = const()[name = string("hidden_states_3_groups_0"), val = int32(1)];
            tensor<fp16, [320, 320, 3, 3]> down_blocks_0_resnets_0_conv2_weight_to_fp16 = const()[name = string("down_blocks_0_resnets_0_conv2_weight_to_fp16"), val = tensor<fp16, [320, 320, 3, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6822592)))];
            tensor<fp16, [320]> down_blocks_0_resnets_0_conv2_bias_to_fp16 = const()[name = string("down_blocks_0_resnets_0_conv2_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8665856)))];
            tensor<fp16, [1, 320, 64, 64]> hidden_states_3_cast_fp16 = conv(bias = down_blocks_0_resnets_0_conv2_bias_to_fp16, dilations = hidden_states_3_dilations_0, groups = hidden_states_3_groups_0, pad = hidden_states_3_pad_0, pad_type = hidden_states_3_pad_type_0, strides = hidden_states_3_strides_0, weight = down_blocks_0_resnets_0_conv2_weight_to_fp16, x = input_21_cast_fp16)[name = string("hidden_states_3_cast_fp16")];
            tensor<fp16, [1, 320, 64, 64]> hidden_states_5_cast_fp16 = add(x = input_7_cast_fp16_1, y = hidden_states_3_cast_fp16)[name = string("hidden_states_5_cast_fp16")];
            tensor<int32, [5]> reshape_8_shape_0 = const()[name = string("reshape_8_shape_0"), val = tensor<int32, [5]>([1, 32, 10, 64, 64])];
            tensor<fp16, [1, 32, 10, 64, 64]> reshape_8_cast_fp16 = reshape(shape = reshape_8_shape_0, x = hidden_states_5_cast_fp16)[name = string("reshape_8_cast_fp16")];
            tensor<int32, [3]> reduce_mean_6_axes_0 = const()[name = string("reduce_mean_6_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            bool reduce_mean_6_keep_dims_0 = const()[name = string("reduce_mean_6_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 1, 1, 1]> reduce_mean_6_cast_fp16 = reduce_mean(axes = reduce_mean_6_axes_0, keep_dims = reduce_mean_6_keep_dims_0, x = reshape_8_cast_fp16)[name = string("reduce_mean_6_cast_fp16")];
            tensor<fp16, [1, 32, 10, 64, 64]> sub_4_cast_fp16 = sub(x = reshape_8_cast_fp16, y = reduce_mean_6_cast_fp16)[name = string("sub_4_cast_fp16")];
            tensor<fp16, [1, 32, 10, 64, 64]> square_2_cast_fp16 = square(x = sub_4_cast_fp16)[name = string("square_2_cast_fp16")];
            tensor<int32, [3]> reduce_mean_8_axes_0 = const()[name = string("reduce_mean_8_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            bool reduce_mean_8_keep_dims_0 = const()[name = string("reduce_mean_8_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 1, 1, 1]> reduce_mean_8_cast_fp16 = reduce_mean(axes = reduce_mean_8_axes_0, keep_dims = reduce_mean_8_keep_dims_0, x = square_2_cast_fp16)[name = string("reduce_mean_8_cast_fp16")];
            fp16 add_4_y_0_to_fp16 = const()[name = string("add_4_y_0_to_fp16"), val = fp16(0x1.1p-20)];
            tensor<fp16, [1, 32, 1, 1, 1]> add_4_cast_fp16 = add(x = reduce_mean_8_cast_fp16, y = add_4_y_0_to_fp16)[name = string("add_4_cast_fp16")];
            tensor<fp16, [1, 32, 1, 1, 1]> sqrt_2_cast_fp16 = sqrt(x = add_4_cast_fp16)[name = string("sqrt_2_cast_fp16")];
            tensor<fp16, [1, 32, 10, 64, 64]> real_div_2_cast_fp16 = real_div(x = sub_4_cast_fp16, y = sqrt_2_cast_fp16)[name = string("real_div_2_cast_fp16")];
            tensor<int32, [4]> reshape_9_shape_0 = const()[name = string("reshape_9_shape_0"), val = tensor<int32, [4]>([1, 320, 64, 64])];
            tensor<fp16, [1, 320, 64, 64]> reshape_9_cast_fp16 = reshape(shape = reshape_9_shape_0, x = real_div_2_cast_fp16)[name = string("reshape_9_cast_fp16")];
            tensor<fp16, [320]> add_5_gamma_0_to_fp16 = const()[name = string("add_5_gamma_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8666560)))];
            tensor<fp16, [320]> add_5_beta_0_to_fp16 = const()[name = string("add_5_beta_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8667264)))];
            fp16 add_5_epsilon_0_to_fp16 = const()[name = string("add_5_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 320, 64, 64]> add_5_cast_fp16 = batch_norm(beta = add_5_beta_0_to_fp16, epsilon = add_5_epsilon_0_to_fp16, gamma = add_5_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_9_cast_fp16)[name = string("add_5_cast_fp16")];
            string hidden_states_7_pad_type_0 = const()[name = string("hidden_states_7_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> hidden_states_7_strides_0 = const()[name = string("hidden_states_7_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_7_pad_0 = const()[name = string("hidden_states_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_7_dilations_0 = const()[name = string("hidden_states_7_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_7_groups_0 = const()[name = string("hidden_states_7_groups_0"), val = int32(1)];
            tensor<fp16, [320, 320, 1, 1]> down_blocks_0_attentions_0_proj_in_weight_to_fp16 = const()[name = string("down_blocks_0_attentions_0_proj_in_weight_to_fp16"), val = tensor<fp16, [320, 320, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8667968)))];
            tensor<fp16, [320]> down_blocks_0_attentions_0_proj_in_bias_to_fp16 = const()[name = string("down_blocks_0_attentions_0_proj_in_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8872832)))];
            tensor<fp16, [1, 320, 64, 64]> hidden_states_7_cast_fp16 = conv(bias = down_blocks_0_attentions_0_proj_in_bias_to_fp16, dilations = hidden_states_7_dilations_0, groups = hidden_states_7_groups_0, pad = hidden_states_7_pad_0, pad_type = hidden_states_7_pad_type_0, strides = hidden_states_7_strides_0, weight = down_blocks_0_attentions_0_proj_in_weight_to_fp16, x = add_5_cast_fp16)[name = string("hidden_states_7_cast_fp16")];
            tensor<int32, [4]> var_177 = const()[name = string("op_177"), val = tensor<int32, [4]>([1, 320, 1, 4096])];
            tensor<fp16, [1, 320, 1, 4096]> inputs_1_cast_fp16 = reshape(shape = var_177, x = hidden_states_7_cast_fp16)[name = string("inputs_1_cast_fp16")];
            tensor<int32, [1]> hidden_states_9_axes_0 = const()[name = string("hidden_states_9_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [320]> hidden_states_9_gamma_0_to_fp16 = const()[name = string("hidden_states_9_gamma_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8873536)))];
            tensor<fp16, [320]> hidden_states_9_beta_0_to_fp16 = const()[name = string("hidden_states_9_beta_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8874240)))];
            fp16 var_193_to_fp16 = const()[name = string("op_193_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 320, 1, 4096]> hidden_states_9_cast_fp16 = layer_norm(axes = hidden_states_9_axes_0, beta = hidden_states_9_beta_0_to_fp16, epsilon = var_193_to_fp16, gamma = hidden_states_9_gamma_0_to_fp16, x = inputs_1_cast_fp16)[name = string("hidden_states_9_cast_fp16")];
            string q_1_pad_type_0 = const()[name = string("q_1_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> q_1_strides_0 = const()[name = string("q_1_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_1_pad_0 = const()[name = string("q_1_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_1_dilations_0 = const()[name = string("q_1_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 q_1_groups_0 = const()[name = string("q_1_groups_0"), val = int32(1)];
            tensor<fp16, [320, 320, 1, 1]> down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = string("down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [320, 320, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8874944)))];
            tensor<fp16, [1, 320, 1, 4096]> q_1_cast_fp16 = conv(dilations = q_1_dilations_0, groups = q_1_groups_0, pad = q_1_pad_0, pad_type = q_1_pad_type_0, strides = q_1_strides_0, weight = down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_9_cast_fp16)[name = string("q_1_cast_fp16")];
            string k_1_pad_type_0 = const()[name = string("k_1_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> k_1_strides_0 = const()[name = string("k_1_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_1_pad_0 = const()[name = string("k_1_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_1_dilations_0 = const()[name = string("k_1_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 k_1_groups_0 = const()[name = string("k_1_groups_0"), val = int32(1)];
            tensor<fp16, [320, 320, 1, 1]> down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = string("down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [320, 320, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9079808)))];
            tensor<fp16, [1, 320, 1, 4096]> k_1_cast_fp16 = conv(dilations = k_1_dilations_0, groups = k_1_groups_0, pad = k_1_pad_0, pad_type = k_1_pad_type_0, strides = k_1_strides_0, weight = down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_9_cast_fp16)[name = string("k_1_cast_fp16")];
            string v_1_pad_type_0 = const()[name = string("v_1_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> v_1_strides_0 = const()[name = string("v_1_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_1_pad_0 = const()[name = string("v_1_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_1_dilations_0 = const()[name = string("v_1_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 v_1_groups_0 = const()[name = string("v_1_groups_0"), val = int32(1)];
            tensor<fp16, [320, 320, 1, 1]> down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = string("down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [320, 320, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9284672)))];
            tensor<fp16, [1, 320, 1, 4096]> v_1_cast_fp16 = conv(dilations = v_1_dilations_0, groups = v_1_groups_0, pad = v_1_pad_0, pad_type = v_1_pad_type_0, strides = v_1_strides_0, weight = down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_9_cast_fp16)[name = string("v_1_cast_fp16")];
            tensor<int32, [4]> var_226_begin_0 = const()[name = string("op_226_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_226_end_0 = const()[name = string("op_226_end_0"), val = tensor<int32, [4]>([1, 40, 1, 4096])];
            tensor<bool, [4]> var_226_end_mask_0 = const()[name = string("op_226_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 40, 1, 4096]> var_226_cast_fp16 = slice_by_index(begin = var_226_begin_0, end = var_226_end_0, end_mask = var_226_end_mask_0, x = q_1_cast_fp16)[name = string("op_226_cast_fp16")];
            tensor<int32, [4]> var_230_begin_0 = const()[name = string("op_230_begin_0"), val = tensor<int32, [4]>([0, 40, 0, 0])];
            tensor<int32, [4]> var_230_end_0 = const()[name = string("op_230_end_0"), val = tensor<int32, [4]>([1, 80, 1, 4096])];
            tensor<bool, [4]> var_230_end_mask_0 = const()[name = string("op_230_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 40, 1, 4096]> var_230_cast_fp16 = slice_by_index(begin = var_230_begin_0, end = var_230_end_0, end_mask = var_230_end_mask_0, x = q_1_cast_fp16)[name = string("op_230_cast_fp16")];
            tensor<int32, [4]> var_234_begin_0 = const()[name = string("op_234_begin_0"), val = tensor<int32, [4]>([0, 80, 0, 0])];
            tensor<int32, [4]> var_234_end_0 = const()[name = string("op_234_end_0"), val = tensor<int32, [4]>([1, 120, 1, 4096])];
            tensor<bool, [4]> var_234_end_mask_0 = const()[name = string("op_234_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 40, 1, 4096]> var_234_cast_fp16 = slice_by_index(begin = var_234_begin_0, end = var_234_end_0, end_mask = var_234_end_mask_0, x = q_1_cast_fp16)[name = string("op_234_cast_fp16")];
            tensor<int32, [4]> var_238_begin_0 = const()[name = string("op_238_begin_0"), val = tensor<int32, [4]>([0, 120, 0, 0])];
            tensor<int32, [4]> var_238_end_0 = const()[name = string("op_238_end_0"), val = tensor<int32, [4]>([1, 160, 1, 4096])];
            tensor<bool, [4]> var_238_end_mask_0 = const()[name = string("op_238_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 40, 1, 4096]> var_238_cast_fp16 = slice_by_index(begin = var_238_begin_0, end = var_238_end_0, end_mask = var_238_end_mask_0, x = q_1_cast_fp16)[name = string("op_238_cast_fp16")];
            tensor<int32, [4]> var_242_begin_0 = const()[name = string("op_242_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_242_end_0 = const()[name = string("op_242_end_0"), val = tensor<int32, [4]>([1, 200, 1, 4096])];
            tensor<bool, [4]> var_242_end_mask_0 = const()[name = string("op_242_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 40, 1, 4096]> var_242_cast_fp16 = slice_by_index(begin = var_242_begin_0, end = var_242_end_0, end_mask = var_242_end_mask_0, x = q_1_cast_fp16)[name = string("op_242_cast_fp16")];
            tensor<int32, [4]> var_246_begin_0 = const()[name = string("op_246_begin_0"), val = tensor<int32, [4]>([0, 200, 0, 0])];
            tensor<int32, [4]> var_246_end_0 = const()[name = string("op_246_end_0"), val = tensor<int32, [4]>([1, 240, 1, 4096])];
            tensor<bool, [4]> var_246_end_mask_0 = const()[name = string("op_246_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 40, 1, 4096]> var_246_cast_fp16 = slice_by_index(begin = var_246_begin_0, end = var_246_end_0, end_mask = var_246_end_mask_0, x = q_1_cast_fp16)[name = string("op_246_cast_fp16")];
            tensor<int32, [4]> var_250_begin_0 = const()[name = string("op_250_begin_0"), val = tensor<int32, [4]>([0, 240, 0, 0])];
            tensor<int32, [4]> var_250_end_0 = const()[name = string("op_250_end_0"), val = tensor<int32, [4]>([1, 280, 1, 4096])];
            tensor<bool, [4]> var_250_end_mask_0 = const()[name = string("op_250_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 40, 1, 4096]> var_250_cast_fp16 = slice_by_index(begin = var_250_begin_0, end = var_250_end_0, end_mask = var_250_end_mask_0, x = q_1_cast_fp16)[name = string("op_250_cast_fp16")];
            tensor<int32, [4]> var_254_begin_0 = const()[name = string("op_254_begin_0"), val = tensor<int32, [4]>([0, 280, 0, 0])];
            tensor<int32, [4]> var_254_end_0 = const()[name = string("op_254_end_0"), val = tensor<int32, [4]>([1, 1, 1, 4096])];
            tensor<bool, [4]> var_254_end_mask_0 = const()[name = string("op_254_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 40, 1, 4096]> var_254_cast_fp16 = slice_by_index(begin = var_254_begin_0, end = var_254_end_0, end_mask = var_254_end_mask_0, x = q_1_cast_fp16)[name = string("op_254_cast_fp16")];
            tensor<int32, [4]> k_3_perm_0 = const()[name = string("k_3_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_261_begin_0 = const()[name = string("op_261_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_261_end_0 = const()[name = string("op_261_end_0"), val = tensor<int32, [4]>([1, 4096, 1, 40])];
            tensor<bool, [4]> var_261_end_mask_0 = const()[name = string("op_261_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4096, 1, 320]> k_3_cast_fp16 = transpose(perm = k_3_perm_0, x = k_1_cast_fp16)[name = string("transpose_7")];
            tensor<fp16, [1, 4096, 1, 40]> var_261_cast_fp16 = slice_by_index(begin = var_261_begin_0, end = var_261_end_0, end_mask = var_261_end_mask_0, x = k_3_cast_fp16)[name = string("op_261_cast_fp16")];
            tensor<int32, [4]> var_265_begin_0 = const()[name = string("op_265_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 40])];
            tensor<int32, [4]> var_265_end_0 = const()[name = string("op_265_end_0"), val = tensor<int32, [4]>([1, 4096, 1, 80])];
            tensor<bool, [4]> var_265_end_mask_0 = const()[name = string("op_265_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4096, 1, 40]> var_265_cast_fp16 = slice_by_index(begin = var_265_begin_0, end = var_265_end_0, end_mask = var_265_end_mask_0, x = k_3_cast_fp16)[name = string("op_265_cast_fp16")];
            tensor<int32, [4]> var_269_begin_0 = const()[name = string("op_269_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 80])];
            tensor<int32, [4]> var_269_end_0 = const()[name = string("op_269_end_0"), val = tensor<int32, [4]>([1, 4096, 1, 120])];
            tensor<bool, [4]> var_269_end_mask_0 = const()[name = string("op_269_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4096, 1, 40]> var_269_cast_fp16 = slice_by_index(begin = var_269_begin_0, end = var_269_end_0, end_mask = var_269_end_mask_0, x = k_3_cast_fp16)[name = string("op_269_cast_fp16")];
            tensor<int32, [4]> var_273_begin_0 = const()[name = string("op_273_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 120])];
            tensor<int32, [4]> var_273_end_0 = const()[name = string("op_273_end_0"), val = tensor<int32, [4]>([1, 4096, 1, 160])];
            tensor<bool, [4]> var_273_end_mask_0 = const()[name = string("op_273_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4096, 1, 40]> var_273_cast_fp16 = slice_by_index(begin = var_273_begin_0, end = var_273_end_0, end_mask = var_273_end_mask_0, x = k_3_cast_fp16)[name = string("op_273_cast_fp16")];
            tensor<int32, [4]> var_277_begin_0 = const()[name = string("op_277_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 160])];
            tensor<int32, [4]> var_277_end_0 = const()[name = string("op_277_end_0"), val = tensor<int32, [4]>([1, 4096, 1, 200])];
            tensor<bool, [4]> var_277_end_mask_0 = const()[name = string("op_277_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4096, 1, 40]> var_277_cast_fp16 = slice_by_index(begin = var_277_begin_0, end = var_277_end_0, end_mask = var_277_end_mask_0, x = k_3_cast_fp16)[name = string("op_277_cast_fp16")];
            tensor<int32, [4]> var_281_begin_0 = const()[name = string("op_281_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 200])];
            tensor<int32, [4]> var_281_end_0 = const()[name = string("op_281_end_0"), val = tensor<int32, [4]>([1, 4096, 1, 240])];
            tensor<bool, [4]> var_281_end_mask_0 = const()[name = string("op_281_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4096, 1, 40]> var_281_cast_fp16 = slice_by_index(begin = var_281_begin_0, end = var_281_end_0, end_mask = var_281_end_mask_0, x = k_3_cast_fp16)[name = string("op_281_cast_fp16")];
            tensor<int32, [4]> var_285_begin_0 = const()[name = string("op_285_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 240])];
            tensor<int32, [4]> var_285_end_0 = const()[name = string("op_285_end_0"), val = tensor<int32, [4]>([1, 4096, 1, 280])];
            tensor<bool, [4]> var_285_end_mask_0 = const()[name = string("op_285_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4096, 1, 40]> var_285_cast_fp16 = slice_by_index(begin = var_285_begin_0, end = var_285_end_0, end_mask = var_285_end_mask_0, x = k_3_cast_fp16)[name = string("op_285_cast_fp16")];
            tensor<int32, [4]> var_289_begin_0 = const()[name = string("op_289_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 280])];
            tensor<int32, [4]> var_289_end_0 = const()[name = string("op_289_end_0"), val = tensor<int32, [4]>([1, 4096, 1, 1])];
            tensor<bool, [4]> var_289_end_mask_0 = const()[name = string("op_289_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4096, 1, 40]> var_289_cast_fp16 = slice_by_index(begin = var_289_begin_0, end = var_289_end_0, end_mask = var_289_end_mask_0, x = k_3_cast_fp16)[name = string("op_289_cast_fp16")];
            tensor<int32, [4]> var_291_begin_0 = const()[name = string("op_291_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_291_end_0 = const()[name = string("op_291_end_0"), val = tensor<int32, [4]>([1, 40, 1, 4096])];
            tensor<bool, [4]> var_291_end_mask_0 = const()[name = string("op_291_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 40, 1, 4096]> var_291_cast_fp16 = slice_by_index(begin = var_291_begin_0, end = var_291_end_0, end_mask = var_291_end_mask_0, x = v_1_cast_fp16)[name = string("op_291_cast_fp16")];
            tensor<int32, [4]> var_295_begin_0 = const()[name = string("op_295_begin_0"), val = tensor<int32, [4]>([0, 40, 0, 0])];
            tensor<int32, [4]> var_295_end_0 = const()[name = string("op_295_end_0"), val = tensor<int32, [4]>([1, 80, 1, 4096])];
            tensor<bool, [4]> var_295_end_mask_0 = const()[name = string("op_295_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 40, 1, 4096]> var_295_cast_fp16 = slice_by_index(begin = var_295_begin_0, end = var_295_end_0, end_mask = var_295_end_mask_0, x = v_1_cast_fp16)[name = string("op_295_cast_fp16")];
            tensor<int32, [4]> var_299_begin_0 = const()[name = string("op_299_begin_0"), val = tensor<int32, [4]>([0, 80, 0, 0])];
            tensor<int32, [4]> var_299_end_0 = const()[name = string("op_299_end_0"), val = tensor<int32, [4]>([1, 120, 1, 4096])];
            tensor<bool, [4]> var_299_end_mask_0 = const()[name = string("op_299_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 40, 1, 4096]> var_299_cast_fp16 = slice_by_index(begin = var_299_begin_0, end = var_299_end_0, end_mask = var_299_end_mask_0, x = v_1_cast_fp16)[name = string("op_299_cast_fp16")];
            tensor<int32, [4]> var_303_begin_0 = const()[name = string("op_303_begin_0"), val = tensor<int32, [4]>([0, 120, 0, 0])];
            tensor<int32, [4]> var_303_end_0 = const()[name = string("op_303_end_0"), val = tensor<int32, [4]>([1, 160, 1, 4096])];
            tensor<bool, [4]> var_303_end_mask_0 = const()[name = string("op_303_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 40, 1, 4096]> var_303_cast_fp16 = slice_by_index(begin = var_303_begin_0, end = var_303_end_0, end_mask = var_303_end_mask_0, x = v_1_cast_fp16)[name = string("op_303_cast_fp16")];
            tensor<int32, [4]> var_307_begin_0 = const()[name = string("op_307_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_307_end_0 = const()[name = string("op_307_end_0"), val = tensor<int32, [4]>([1, 200, 1, 4096])];
            tensor<bool, [4]> var_307_end_mask_0 = const()[name = string("op_307_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 40, 1, 4096]> var_307_cast_fp16 = slice_by_index(begin = var_307_begin_0, end = var_307_end_0, end_mask = var_307_end_mask_0, x = v_1_cast_fp16)[name = string("op_307_cast_fp16")];
            tensor<int32, [4]> var_311_begin_0 = const()[name = string("op_311_begin_0"), val = tensor<int32, [4]>([0, 200, 0, 0])];
            tensor<int32, [4]> var_311_end_0 = const()[name = string("op_311_end_0"), val = tensor<int32, [4]>([1, 240, 1, 4096])];
            tensor<bool, [4]> var_311_end_mask_0 = const()[name = string("op_311_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 40, 1, 4096]> var_311_cast_fp16 = slice_by_index(begin = var_311_begin_0, end = var_311_end_0, end_mask = var_311_end_mask_0, x = v_1_cast_fp16)[name = string("op_311_cast_fp16")];
            tensor<int32, [4]> var_315_begin_0 = const()[name = string("op_315_begin_0"), val = tensor<int32, [4]>([0, 240, 0, 0])];
            tensor<int32, [4]> var_315_end_0 = const()[name = string("op_315_end_0"), val = tensor<int32, [4]>([1, 280, 1, 4096])];
            tensor<bool, [4]> var_315_end_mask_0 = const()[name = string("op_315_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 40, 1, 4096]> var_315_cast_fp16 = slice_by_index(begin = var_315_begin_0, end = var_315_end_0, end_mask = var_315_end_mask_0, x = v_1_cast_fp16)[name = string("op_315_cast_fp16")];
            tensor<int32, [4]> var_319_begin_0 = const()[name = string("op_319_begin_0"), val = tensor<int32, [4]>([0, 280, 0, 0])];
            tensor<int32, [4]> var_319_end_0 = const()[name = string("op_319_end_0"), val = tensor<int32, [4]>([1, 1, 1, 4096])];
            tensor<bool, [4]> var_319_end_mask_0 = const()[name = string("op_319_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 40, 1, 4096]> var_319_cast_fp16 = slice_by_index(begin = var_319_begin_0, end = var_319_end_0, end_mask = var_319_end_mask_0, x = v_1_cast_fp16)[name = string("op_319_cast_fp16")];
            string var_323_equation_0 = const()[name = string("op_323_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 4096, 1, 4096]> var_323_cast_fp16 = einsum(equation = var_323_equation_0, values = (var_261_cast_fp16, var_226_cast_fp16))[name = string("op_323_cast_fp16")];
            fp16 var_324_to_fp16 = const()[name = string("op_324_to_fp16"), val = fp16(0x1.43cp-3)];
            tensor<fp16, [1, 4096, 1, 4096]> aw_1_cast_fp16 = mul(x = var_323_cast_fp16, y = var_324_to_fp16)[name = string("aw_1_cast_fp16")];
            string var_327_equation_0 = const()[name = string("op_327_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 4096, 1, 4096]> var_327_cast_fp16 = einsum(equation = var_327_equation_0, values = (var_265_cast_fp16, var_230_cast_fp16))[name = string("op_327_cast_fp16")];
            fp16 var_328_to_fp16 = const()[name = string("op_328_to_fp16"), val = fp16(0x1.43cp-3)];
            tensor<fp16, [1, 4096, 1, 4096]> aw_3_cast_fp16 = mul(x = var_327_cast_fp16, y = var_328_to_fp16)[name = string("aw_3_cast_fp16")];
            string var_331_equation_0 = const()[name = string("op_331_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 4096, 1, 4096]> var_331_cast_fp16 = einsum(equation = var_331_equation_0, values = (var_269_cast_fp16, var_234_cast_fp16))[name = string("op_331_cast_fp16")];
            fp16 var_332_to_fp16 = const()[name = string("op_332_to_fp16"), val = fp16(0x1.43cp-3)];
            tensor<fp16, [1, 4096, 1, 4096]> aw_5_cast_fp16 = mul(x = var_331_cast_fp16, y = var_332_to_fp16)[name = string("aw_5_cast_fp16")];
            string var_335_equation_0 = const()[name = string("op_335_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 4096, 1, 4096]> var_335_cast_fp16 = einsum(equation = var_335_equation_0, values = (var_273_cast_fp16, var_238_cast_fp16))[name = string("op_335_cast_fp16")];
            fp16 var_336_to_fp16 = const()[name = string("op_336_to_fp16"), val = fp16(0x1.43cp-3)];
            tensor<fp16, [1, 4096, 1, 4096]> aw_7_cast_fp16 = mul(x = var_335_cast_fp16, y = var_336_to_fp16)[name = string("aw_7_cast_fp16")];
            string var_339_equation_0 = const()[name = string("op_339_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 4096, 1, 4096]> var_339_cast_fp16 = einsum(equation = var_339_equation_0, values = (var_277_cast_fp16, var_242_cast_fp16))[name = string("op_339_cast_fp16")];
            fp16 var_340_to_fp16 = const()[name = string("op_340_to_fp16"), val = fp16(0x1.43cp-3)];
            tensor<fp16, [1, 4096, 1, 4096]> aw_9_cast_fp16 = mul(x = var_339_cast_fp16, y = var_340_to_fp16)[name = string("aw_9_cast_fp16")];
            string var_343_equation_0 = const()[name = string("op_343_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 4096, 1, 4096]> var_343_cast_fp16 = einsum(equation = var_343_equation_0, values = (var_281_cast_fp16, var_246_cast_fp16))[name = string("op_343_cast_fp16")];
            fp16 var_344_to_fp16 = const()[name = string("op_344_to_fp16"), val = fp16(0x1.43cp-3)];
            tensor<fp16, [1, 4096, 1, 4096]> aw_11_cast_fp16 = mul(x = var_343_cast_fp16, y = var_344_to_fp16)[name = string("aw_11_cast_fp16")];
            string var_347_equation_0 = const()[name = string("op_347_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 4096, 1, 4096]> var_347_cast_fp16 = einsum(equation = var_347_equation_0, values = (var_285_cast_fp16, var_250_cast_fp16))[name = string("op_347_cast_fp16")];
            fp16 var_348_to_fp16 = const()[name = string("op_348_to_fp16"), val = fp16(0x1.43cp-3)];
            tensor<fp16, [1, 4096, 1, 4096]> aw_13_cast_fp16 = mul(x = var_347_cast_fp16, y = var_348_to_fp16)[name = string("aw_13_cast_fp16")];
            string var_351_equation_0 = const()[name = string("op_351_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 4096, 1, 4096]> var_351_cast_fp16 = einsum(equation = var_351_equation_0, values = (var_289_cast_fp16, var_254_cast_fp16))[name = string("op_351_cast_fp16")];
            fp16 var_352_to_fp16 = const()[name = string("op_352_to_fp16"), val = fp16(0x1.43cp-3)];
            tensor<fp16, [1, 4096, 1, 4096]> aw_15_cast_fp16 = mul(x = var_351_cast_fp16, y = var_352_to_fp16)[name = string("aw_15_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 4096]> var_354_cast_fp16 = softmax(axis = var_110, x = aw_1_cast_fp16)[name = string("op_354_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 4096]> var_355_cast_fp16 = softmax(axis = var_110, x = aw_3_cast_fp16)[name = string("op_355_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 4096]> var_356_cast_fp16 = softmax(axis = var_110, x = aw_5_cast_fp16)[name = string("op_356_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 4096]> var_357_cast_fp16 = softmax(axis = var_110, x = aw_7_cast_fp16)[name = string("op_357_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 4096]> var_358_cast_fp16 = softmax(axis = var_110, x = aw_9_cast_fp16)[name = string("op_358_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 4096]> var_359_cast_fp16 = softmax(axis = var_110, x = aw_11_cast_fp16)[name = string("op_359_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 4096]> var_360_cast_fp16 = softmax(axis = var_110, x = aw_13_cast_fp16)[name = string("op_360_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 4096]> var_361_cast_fp16 = softmax(axis = var_110, x = aw_15_cast_fp16)[name = string("op_361_cast_fp16")];
            string var_363_equation_0 = const()[name = string("op_363_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 40, 1, 4096]> var_363_cast_fp16 = einsum(equation = var_363_equation_0, values = (var_291_cast_fp16, var_354_cast_fp16))[name = string("op_363_cast_fp16")];
            string var_365_equation_0 = const()[name = string("op_365_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 40, 1, 4096]> var_365_cast_fp16 = einsum(equation = var_365_equation_0, values = (var_295_cast_fp16, var_355_cast_fp16))[name = string("op_365_cast_fp16")];
            string var_367_equation_0 = const()[name = string("op_367_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 40, 1, 4096]> var_367_cast_fp16 = einsum(equation = var_367_equation_0, values = (var_299_cast_fp16, var_356_cast_fp16))[name = string("op_367_cast_fp16")];
            string var_369_equation_0 = const()[name = string("op_369_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 40, 1, 4096]> var_369_cast_fp16 = einsum(equation = var_369_equation_0, values = (var_303_cast_fp16, var_357_cast_fp16))[name = string("op_369_cast_fp16")];
            string var_371_equation_0 = const()[name = string("op_371_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 40, 1, 4096]> var_371_cast_fp16 = einsum(equation = var_371_equation_0, values = (var_307_cast_fp16, var_358_cast_fp16))[name = string("op_371_cast_fp16")];
            string var_373_equation_0 = const()[name = string("op_373_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 40, 1, 4096]> var_373_cast_fp16 = einsum(equation = var_373_equation_0, values = (var_311_cast_fp16, var_359_cast_fp16))[name = string("op_373_cast_fp16")];
            string var_375_equation_0 = const()[name = string("op_375_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 40, 1, 4096]> var_375_cast_fp16 = einsum(equation = var_375_equation_0, values = (var_315_cast_fp16, var_360_cast_fp16))[name = string("op_375_cast_fp16")];
            string var_377_equation_0 = const()[name = string("op_377_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 40, 1, 4096]> var_377_cast_fp16 = einsum(equation = var_377_equation_0, values = (var_319_cast_fp16, var_361_cast_fp16))[name = string("op_377_cast_fp16")];
            bool input_25_interleave_0 = const()[name = string("input_25_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 320, 1, 4096]> input_25_cast_fp16 = concat(axis = var_110, interleave = input_25_interleave_0, values = (var_363_cast_fp16, var_365_cast_fp16, var_367_cast_fp16, var_369_cast_fp16, var_371_cast_fp16, var_373_cast_fp16, var_375_cast_fp16, var_377_cast_fp16))[name = string("input_25_cast_fp16")];
            string var_387_pad_type_0 = const()[name = string("op_387_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_387_strides_0 = const()[name = string("op_387_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_387_pad_0 = const()[name = string("op_387_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_387_dilations_0 = const()[name = string("op_387_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_387_groups_0 = const()[name = string("op_387_groups_0"), val = int32(1)];
            tensor<fp16, [320, 320, 1, 1]> down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = string("down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [320, 320, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9489536)))];
            tensor<fp16, [320]> down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = string("down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9694400)))];
            tensor<fp16, [1, 320, 1, 4096]> var_387_cast_fp16 = conv(bias = down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_387_dilations_0, groups = var_387_groups_0, pad = var_387_pad_0, pad_type = var_387_pad_type_0, strides = var_387_strides_0, weight = down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_25_cast_fp16)[name = string("op_387_cast_fp16")];
            tensor<fp16, [1, 320, 1, 4096]> inputs_3_cast_fp16 = add(x = var_387_cast_fp16, y = inputs_1_cast_fp16)[name = string("inputs_3_cast_fp16")];
            tensor<int32, [1]> hidden_states_11_axes_0 = const()[name = string("hidden_states_11_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [320]> hidden_states_11_gamma_0_to_fp16 = const()[name = string("hidden_states_11_gamma_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9695104)))];
            tensor<fp16, [320]> hidden_states_11_beta_0_to_fp16 = const()[name = string("hidden_states_11_beta_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9695808)))];
            fp16 var_397_to_fp16 = const()[name = string("op_397_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 320, 1, 4096]> hidden_states_11_cast_fp16 = layer_norm(axes = hidden_states_11_axes_0, beta = hidden_states_11_beta_0_to_fp16, epsilon = var_397_to_fp16, gamma = hidden_states_11_gamma_0_to_fp16, x = inputs_3_cast_fp16)[name = string("hidden_states_11_cast_fp16")];
            string q_3_pad_type_0 = const()[name = string("q_3_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> q_3_strides_0 = const()[name = string("q_3_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_3_pad_0 = const()[name = string("q_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_3_dilations_0 = const()[name = string("q_3_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 q_3_groups_0 = const()[name = string("q_3_groups_0"), val = int32(1)];
            tensor<fp16, [320, 320, 1, 1]> down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = string("down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [320, 320, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9696512)))];
            tensor<fp16, [1, 320, 1, 4096]> q_3_cast_fp16 = conv(dilations = q_3_dilations_0, groups = q_3_groups_0, pad = q_3_pad_0, pad_type = q_3_pad_type_0, strides = q_3_strides_0, weight = down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_11_cast_fp16)[name = string("q_3_cast_fp16")];
            string k_5_pad_type_0 = const()[name = string("k_5_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> k_5_strides_0 = const()[name = string("k_5_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_5_pad_0 = const()[name = string("k_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_5_dilations_0 = const()[name = string("k_5_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 k_5_groups_0 = const()[name = string("k_5_groups_0"), val = int32(1)];
            tensor<fp16, [320, 768, 1, 1]> down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = string("down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [320, 768, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9901376)))];
            tensor<fp16, [1, 320, 1, 77]> k_5_cast_fp16 = conv(dilations = k_5_dilations_0, groups = k_5_groups_0, pad = k_5_pad_0, pad_type = k_5_pad_type_0, strides = k_5_strides_0, weight = down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = string("k_5_cast_fp16")];
            string v_3_pad_type_0 = const()[name = string("v_3_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> v_3_strides_0 = const()[name = string("v_3_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_3_pad_0 = const()[name = string("v_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_3_dilations_0 = const()[name = string("v_3_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 v_3_groups_0 = const()[name = string("v_3_groups_0"), val = int32(1)];
            tensor<fp16, [320, 768, 1, 1]> down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = string("down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [320, 768, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10392960)))];
            tensor<fp16, [1, 320, 1, 77]> v_3_cast_fp16 = conv(dilations = v_3_dilations_0, groups = v_3_groups_0, pad = v_3_pad_0, pad_type = v_3_pad_type_0, strides = v_3_strides_0, weight = down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = string("v_3_cast_fp16")];
            tensor<int32, [4]> var_430_begin_0 = const()[name = string("op_430_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_430_end_0 = const()[name = string("op_430_end_0"), val = tensor<int32, [4]>([1, 40, 1, 4096])];
            tensor<bool, [4]> var_430_end_mask_0 = const()[name = string("op_430_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 40, 1, 4096]> var_430_cast_fp16 = slice_by_index(begin = var_430_begin_0, end = var_430_end_0, end_mask = var_430_end_mask_0, x = q_3_cast_fp16)[name = string("op_430_cast_fp16")];
            tensor<int32, [4]> var_434_begin_0 = const()[name = string("op_434_begin_0"), val = tensor<int32, [4]>([0, 40, 0, 0])];
            tensor<int32, [4]> var_434_end_0 = const()[name = string("op_434_end_0"), val = tensor<int32, [4]>([1, 80, 1, 4096])];
            tensor<bool, [4]> var_434_end_mask_0 = const()[name = string("op_434_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 40, 1, 4096]> var_434_cast_fp16 = slice_by_index(begin = var_434_begin_0, end = var_434_end_0, end_mask = var_434_end_mask_0, x = q_3_cast_fp16)[name = string("op_434_cast_fp16")];
            tensor<int32, [4]> var_438_begin_0 = const()[name = string("op_438_begin_0"), val = tensor<int32, [4]>([0, 80, 0, 0])];
            tensor<int32, [4]> var_438_end_0 = const()[name = string("op_438_end_0"), val = tensor<int32, [4]>([1, 120, 1, 4096])];
            tensor<bool, [4]> var_438_end_mask_0 = const()[name = string("op_438_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 40, 1, 4096]> var_438_cast_fp16 = slice_by_index(begin = var_438_begin_0, end = var_438_end_0, end_mask = var_438_end_mask_0, x = q_3_cast_fp16)[name = string("op_438_cast_fp16")];
            tensor<int32, [4]> var_442_begin_0 = const()[name = string("op_442_begin_0"), val = tensor<int32, [4]>([0, 120, 0, 0])];
            tensor<int32, [4]> var_442_end_0 = const()[name = string("op_442_end_0"), val = tensor<int32, [4]>([1, 160, 1, 4096])];
            tensor<bool, [4]> var_442_end_mask_0 = const()[name = string("op_442_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 40, 1, 4096]> var_442_cast_fp16 = slice_by_index(begin = var_442_begin_0, end = var_442_end_0, end_mask = var_442_end_mask_0, x = q_3_cast_fp16)[name = string("op_442_cast_fp16")];
            tensor<int32, [4]> var_446_begin_0 = const()[name = string("op_446_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_446_end_0 = const()[name = string("op_446_end_0"), val = tensor<int32, [4]>([1, 200, 1, 4096])];
            tensor<bool, [4]> var_446_end_mask_0 = const()[name = string("op_446_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 40, 1, 4096]> var_446_cast_fp16 = slice_by_index(begin = var_446_begin_0, end = var_446_end_0, end_mask = var_446_end_mask_0, x = q_3_cast_fp16)[name = string("op_446_cast_fp16")];
            tensor<int32, [4]> var_450_begin_0 = const()[name = string("op_450_begin_0"), val = tensor<int32, [4]>([0, 200, 0, 0])];
            tensor<int32, [4]> var_450_end_0 = const()[name = string("op_450_end_0"), val = tensor<int32, [4]>([1, 240, 1, 4096])];
            tensor<bool, [4]> var_450_end_mask_0 = const()[name = string("op_450_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 40, 1, 4096]> var_450_cast_fp16 = slice_by_index(begin = var_450_begin_0, end = var_450_end_0, end_mask = var_450_end_mask_0, x = q_3_cast_fp16)[name = string("op_450_cast_fp16")];
            tensor<int32, [4]> var_454_begin_0 = const()[name = string("op_454_begin_0"), val = tensor<int32, [4]>([0, 240, 0, 0])];
            tensor<int32, [4]> var_454_end_0 = const()[name = string("op_454_end_0"), val = tensor<int32, [4]>([1, 280, 1, 4096])];
            tensor<bool, [4]> var_454_end_mask_0 = const()[name = string("op_454_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 40, 1, 4096]> var_454_cast_fp16 = slice_by_index(begin = var_454_begin_0, end = var_454_end_0, end_mask = var_454_end_mask_0, x = q_3_cast_fp16)[name = string("op_454_cast_fp16")];
            tensor<int32, [4]> var_458_begin_0 = const()[name = string("op_458_begin_0"), val = tensor<int32, [4]>([0, 280, 0, 0])];
            tensor<int32, [4]> var_458_end_0 = const()[name = string("op_458_end_0"), val = tensor<int32, [4]>([1, 1, 1, 4096])];
            tensor<bool, [4]> var_458_end_mask_0 = const()[name = string("op_458_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 40, 1, 4096]> var_458_cast_fp16 = slice_by_index(begin = var_458_begin_0, end = var_458_end_0, end_mask = var_458_end_mask_0, x = q_3_cast_fp16)[name = string("op_458_cast_fp16")];
            tensor<int32, [4]> k_7_perm_0 = const()[name = string("k_7_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_465_begin_0 = const()[name = string("op_465_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_465_end_0 = const()[name = string("op_465_end_0"), val = tensor<int32, [4]>([1, 77, 1, 40])];
            tensor<bool, [4]> var_465_end_mask_0 = const()[name = string("op_465_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 77, 1, 320]> k_7_cast_fp16 = transpose(perm = k_7_perm_0, x = k_5_cast_fp16)[name = string("transpose_6")];
            tensor<fp16, [1, 77, 1, 40]> var_465_cast_fp16 = slice_by_index(begin = var_465_begin_0, end = var_465_end_0, end_mask = var_465_end_mask_0, x = k_7_cast_fp16)[name = string("op_465_cast_fp16")];
            tensor<int32, [4]> var_469_begin_0 = const()[name = string("op_469_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 40])];
            tensor<int32, [4]> var_469_end_0 = const()[name = string("op_469_end_0"), val = tensor<int32, [4]>([1, 77, 1, 80])];
            tensor<bool, [4]> var_469_end_mask_0 = const()[name = string("op_469_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 77, 1, 40]> var_469_cast_fp16 = slice_by_index(begin = var_469_begin_0, end = var_469_end_0, end_mask = var_469_end_mask_0, x = k_7_cast_fp16)[name = string("op_469_cast_fp16")];
            tensor<int32, [4]> var_473_begin_0 = const()[name = string("op_473_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 80])];
            tensor<int32, [4]> var_473_end_0 = const()[name = string("op_473_end_0"), val = tensor<int32, [4]>([1, 77, 1, 120])];
            tensor<bool, [4]> var_473_end_mask_0 = const()[name = string("op_473_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 77, 1, 40]> var_473_cast_fp16 = slice_by_index(begin = var_473_begin_0, end = var_473_end_0, end_mask = var_473_end_mask_0, x = k_7_cast_fp16)[name = string("op_473_cast_fp16")];
            tensor<int32, [4]> var_477_begin_0 = const()[name = string("op_477_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 120])];
            tensor<int32, [4]> var_477_end_0 = const()[name = string("op_477_end_0"), val = tensor<int32, [4]>([1, 77, 1, 160])];
            tensor<bool, [4]> var_477_end_mask_0 = const()[name = string("op_477_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 77, 1, 40]> var_477_cast_fp16 = slice_by_index(begin = var_477_begin_0, end = var_477_end_0, end_mask = var_477_end_mask_0, x = k_7_cast_fp16)[name = string("op_477_cast_fp16")];
            tensor<int32, [4]> var_481_begin_0 = const()[name = string("op_481_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 160])];
            tensor<int32, [4]> var_481_end_0 = const()[name = string("op_481_end_0"), val = tensor<int32, [4]>([1, 77, 1, 200])];
            tensor<bool, [4]> var_481_end_mask_0 = const()[name = string("op_481_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 77, 1, 40]> var_481_cast_fp16 = slice_by_index(begin = var_481_begin_0, end = var_481_end_0, end_mask = var_481_end_mask_0, x = k_7_cast_fp16)[name = string("op_481_cast_fp16")];
            tensor<int32, [4]> var_485_begin_0 = const()[name = string("op_485_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 200])];
            tensor<int32, [4]> var_485_end_0 = const()[name = string("op_485_end_0"), val = tensor<int32, [4]>([1, 77, 1, 240])];
            tensor<bool, [4]> var_485_end_mask_0 = const()[name = string("op_485_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 77, 1, 40]> var_485_cast_fp16 = slice_by_index(begin = var_485_begin_0, end = var_485_end_0, end_mask = var_485_end_mask_0, x = k_7_cast_fp16)[name = string("op_485_cast_fp16")];
            tensor<int32, [4]> var_489_begin_0 = const()[name = string("op_489_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 240])];
            tensor<int32, [4]> var_489_end_0 = const()[name = string("op_489_end_0"), val = tensor<int32, [4]>([1, 77, 1, 280])];
            tensor<bool, [4]> var_489_end_mask_0 = const()[name = string("op_489_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 77, 1, 40]> var_489_cast_fp16 = slice_by_index(begin = var_489_begin_0, end = var_489_end_0, end_mask = var_489_end_mask_0, x = k_7_cast_fp16)[name = string("op_489_cast_fp16")];
            tensor<int32, [4]> var_493_begin_0 = const()[name = string("op_493_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 280])];
            tensor<int32, [4]> var_493_end_0 = const()[name = string("op_493_end_0"), val = tensor<int32, [4]>([1, 77, 1, 1])];
            tensor<bool, [4]> var_493_end_mask_0 = const()[name = string("op_493_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 77, 1, 40]> var_493_cast_fp16 = slice_by_index(begin = var_493_begin_0, end = var_493_end_0, end_mask = var_493_end_mask_0, x = k_7_cast_fp16)[name = string("op_493_cast_fp16")];
            tensor<int32, [4]> var_495_begin_0 = const()[name = string("op_495_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_495_end_0 = const()[name = string("op_495_end_0"), val = tensor<int32, [4]>([1, 40, 1, 77])];
            tensor<bool, [4]> var_495_end_mask_0 = const()[name = string("op_495_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 40, 1, 77]> var_495_cast_fp16 = slice_by_index(begin = var_495_begin_0, end = var_495_end_0, end_mask = var_495_end_mask_0, x = v_3_cast_fp16)[name = string("op_495_cast_fp16")];
            tensor<int32, [4]> var_499_begin_0 = const()[name = string("op_499_begin_0"), val = tensor<int32, [4]>([0, 40, 0, 0])];
            tensor<int32, [4]> var_499_end_0 = const()[name = string("op_499_end_0"), val = tensor<int32, [4]>([1, 80, 1, 77])];
            tensor<bool, [4]> var_499_end_mask_0 = const()[name = string("op_499_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 40, 1, 77]> var_499_cast_fp16 = slice_by_index(begin = var_499_begin_0, end = var_499_end_0, end_mask = var_499_end_mask_0, x = v_3_cast_fp16)[name = string("op_499_cast_fp16")];
            tensor<int32, [4]> var_503_begin_0 = const()[name = string("op_503_begin_0"), val = tensor<int32, [4]>([0, 80, 0, 0])];
            tensor<int32, [4]> var_503_end_0 = const()[name = string("op_503_end_0"), val = tensor<int32, [4]>([1, 120, 1, 77])];
            tensor<bool, [4]> var_503_end_mask_0 = const()[name = string("op_503_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 40, 1, 77]> var_503_cast_fp16 = slice_by_index(begin = var_503_begin_0, end = var_503_end_0, end_mask = var_503_end_mask_0, x = v_3_cast_fp16)[name = string("op_503_cast_fp16")];
            tensor<int32, [4]> var_507_begin_0 = const()[name = string("op_507_begin_0"), val = tensor<int32, [4]>([0, 120, 0, 0])];
            tensor<int32, [4]> var_507_end_0 = const()[name = string("op_507_end_0"), val = tensor<int32, [4]>([1, 160, 1, 77])];
            tensor<bool, [4]> var_507_end_mask_0 = const()[name = string("op_507_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 40, 1, 77]> var_507_cast_fp16 = slice_by_index(begin = var_507_begin_0, end = var_507_end_0, end_mask = var_507_end_mask_0, x = v_3_cast_fp16)[name = string("op_507_cast_fp16")];
            tensor<int32, [4]> var_511_begin_0 = const()[name = string("op_511_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_511_end_0 = const()[name = string("op_511_end_0"), val = tensor<int32, [4]>([1, 200, 1, 77])];
            tensor<bool, [4]> var_511_end_mask_0 = const()[name = string("op_511_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 40, 1, 77]> var_511_cast_fp16 = slice_by_index(begin = var_511_begin_0, end = var_511_end_0, end_mask = var_511_end_mask_0, x = v_3_cast_fp16)[name = string("op_511_cast_fp16")];
            tensor<int32, [4]> var_515_begin_0 = const()[name = string("op_515_begin_0"), val = tensor<int32, [4]>([0, 200, 0, 0])];
            tensor<int32, [4]> var_515_end_0 = const()[name = string("op_515_end_0"), val = tensor<int32, [4]>([1, 240, 1, 77])];
            tensor<bool, [4]> var_515_end_mask_0 = const()[name = string("op_515_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 40, 1, 77]> var_515_cast_fp16 = slice_by_index(begin = var_515_begin_0, end = var_515_end_0, end_mask = var_515_end_mask_0, x = v_3_cast_fp16)[name = string("op_515_cast_fp16")];
            tensor<int32, [4]> var_519_begin_0 = const()[name = string("op_519_begin_0"), val = tensor<int32, [4]>([0, 240, 0, 0])];
            tensor<int32, [4]> var_519_end_0 = const()[name = string("op_519_end_0"), val = tensor<int32, [4]>([1, 280, 1, 77])];
            tensor<bool, [4]> var_519_end_mask_0 = const()[name = string("op_519_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 40, 1, 77]> var_519_cast_fp16 = slice_by_index(begin = var_519_begin_0, end = var_519_end_0, end_mask = var_519_end_mask_0, x = v_3_cast_fp16)[name = string("op_519_cast_fp16")];
            tensor<int32, [4]> var_523_begin_0 = const()[name = string("op_523_begin_0"), val = tensor<int32, [4]>([0, 280, 0, 0])];
            tensor<int32, [4]> var_523_end_0 = const()[name = string("op_523_end_0"), val = tensor<int32, [4]>([1, 1, 1, 77])];
            tensor<bool, [4]> var_523_end_mask_0 = const()[name = string("op_523_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 40, 1, 77]> var_523_cast_fp16 = slice_by_index(begin = var_523_begin_0, end = var_523_end_0, end_mask = var_523_end_mask_0, x = v_3_cast_fp16)[name = string("op_523_cast_fp16")];
            string var_527_equation_0 = const()[name = string("op_527_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 77, 1, 4096]> var_527_cast_fp16 = einsum(equation = var_527_equation_0, values = (var_465_cast_fp16, var_430_cast_fp16))[name = string("op_527_cast_fp16")];
            fp16 var_528_to_fp16 = const()[name = string("op_528_to_fp16"), val = fp16(0x1.43cp-3)];
            tensor<fp16, [1, 77, 1, 4096]> aw_17_cast_fp16 = mul(x = var_527_cast_fp16, y = var_528_to_fp16)[name = string("aw_17_cast_fp16")];
            string var_531_equation_0 = const()[name = string("op_531_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 77, 1, 4096]> var_531_cast_fp16 = einsum(equation = var_531_equation_0, values = (var_469_cast_fp16, var_434_cast_fp16))[name = string("op_531_cast_fp16")];
            fp16 var_532_to_fp16 = const()[name = string("op_532_to_fp16"), val = fp16(0x1.43cp-3)];
            tensor<fp16, [1, 77, 1, 4096]> aw_19_cast_fp16 = mul(x = var_531_cast_fp16, y = var_532_to_fp16)[name = string("aw_19_cast_fp16")];
            string var_535_equation_0 = const()[name = string("op_535_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 77, 1, 4096]> var_535_cast_fp16 = einsum(equation = var_535_equation_0, values = (var_473_cast_fp16, var_438_cast_fp16))[name = string("op_535_cast_fp16")];
            fp16 var_536_to_fp16 = const()[name = string("op_536_to_fp16"), val = fp16(0x1.43cp-3)];
            tensor<fp16, [1, 77, 1, 4096]> aw_21_cast_fp16 = mul(x = var_535_cast_fp16, y = var_536_to_fp16)[name = string("aw_21_cast_fp16")];
            string var_539_equation_0 = const()[name = string("op_539_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 77, 1, 4096]> var_539_cast_fp16 = einsum(equation = var_539_equation_0, values = (var_477_cast_fp16, var_442_cast_fp16))[name = string("op_539_cast_fp16")];
            fp16 var_540_to_fp16 = const()[name = string("op_540_to_fp16"), val = fp16(0x1.43cp-3)];
            tensor<fp16, [1, 77, 1, 4096]> aw_23_cast_fp16 = mul(x = var_539_cast_fp16, y = var_540_to_fp16)[name = string("aw_23_cast_fp16")];
            string var_543_equation_0 = const()[name = string("op_543_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 77, 1, 4096]> var_543_cast_fp16 = einsum(equation = var_543_equation_0, values = (var_481_cast_fp16, var_446_cast_fp16))[name = string("op_543_cast_fp16")];
            fp16 var_544_to_fp16 = const()[name = string("op_544_to_fp16"), val = fp16(0x1.43cp-3)];
            tensor<fp16, [1, 77, 1, 4096]> aw_25_cast_fp16 = mul(x = var_543_cast_fp16, y = var_544_to_fp16)[name = string("aw_25_cast_fp16")];
            string var_547_equation_0 = const()[name = string("op_547_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 77, 1, 4096]> var_547_cast_fp16 = einsum(equation = var_547_equation_0, values = (var_485_cast_fp16, var_450_cast_fp16))[name = string("op_547_cast_fp16")];
            fp16 var_548_to_fp16 = const()[name = string("op_548_to_fp16"), val = fp16(0x1.43cp-3)];
            tensor<fp16, [1, 77, 1, 4096]> aw_27_cast_fp16 = mul(x = var_547_cast_fp16, y = var_548_to_fp16)[name = string("aw_27_cast_fp16")];
            string var_551_equation_0 = const()[name = string("op_551_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 77, 1, 4096]> var_551_cast_fp16 = einsum(equation = var_551_equation_0, values = (var_489_cast_fp16, var_454_cast_fp16))[name = string("op_551_cast_fp16")];
            fp16 var_552_to_fp16 = const()[name = string("op_552_to_fp16"), val = fp16(0x1.43cp-3)];
            tensor<fp16, [1, 77, 1, 4096]> aw_29_cast_fp16 = mul(x = var_551_cast_fp16, y = var_552_to_fp16)[name = string("aw_29_cast_fp16")];
            string var_555_equation_0 = const()[name = string("op_555_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 77, 1, 4096]> var_555_cast_fp16 = einsum(equation = var_555_equation_0, values = (var_493_cast_fp16, var_458_cast_fp16))[name = string("op_555_cast_fp16")];
            fp16 var_556_to_fp16 = const()[name = string("op_556_to_fp16"), val = fp16(0x1.43cp-3)];
            tensor<fp16, [1, 77, 1, 4096]> aw_31_cast_fp16 = mul(x = var_555_cast_fp16, y = var_556_to_fp16)[name = string("aw_31_cast_fp16")];
            tensor<fp16, [1, 77, 1, 4096]> var_558_cast_fp16 = softmax(axis = var_110, x = aw_17_cast_fp16)[name = string("op_558_cast_fp16")];
            tensor<fp16, [1, 77, 1, 4096]> var_559_cast_fp16 = softmax(axis = var_110, x = aw_19_cast_fp16)[name = string("op_559_cast_fp16")];
            tensor<fp16, [1, 77, 1, 4096]> var_560_cast_fp16 = softmax(axis = var_110, x = aw_21_cast_fp16)[name = string("op_560_cast_fp16")];
            tensor<fp16, [1, 77, 1, 4096]> var_561_cast_fp16 = softmax(axis = var_110, x = aw_23_cast_fp16)[name = string("op_561_cast_fp16")];
            tensor<fp16, [1, 77, 1, 4096]> var_562_cast_fp16 = softmax(axis = var_110, x = aw_25_cast_fp16)[name = string("op_562_cast_fp16")];
            tensor<fp16, [1, 77, 1, 4096]> var_563_cast_fp16 = softmax(axis = var_110, x = aw_27_cast_fp16)[name = string("op_563_cast_fp16")];
            tensor<fp16, [1, 77, 1, 4096]> var_564_cast_fp16 = softmax(axis = var_110, x = aw_29_cast_fp16)[name = string("op_564_cast_fp16")];
            tensor<fp16, [1, 77, 1, 4096]> var_565_cast_fp16 = softmax(axis = var_110, x = aw_31_cast_fp16)[name = string("op_565_cast_fp16")];
            string var_567_equation_0 = const()[name = string("op_567_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 40, 1, 4096]> var_567_cast_fp16 = einsum(equation = var_567_equation_0, values = (var_495_cast_fp16, var_558_cast_fp16))[name = string("op_567_cast_fp16")];
            string var_569_equation_0 = const()[name = string("op_569_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 40, 1, 4096]> var_569_cast_fp16 = einsum(equation = var_569_equation_0, values = (var_499_cast_fp16, var_559_cast_fp16))[name = string("op_569_cast_fp16")];
            string var_571_equation_0 = const()[name = string("op_571_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 40, 1, 4096]> var_571_cast_fp16 = einsum(equation = var_571_equation_0, values = (var_503_cast_fp16, var_560_cast_fp16))[name = string("op_571_cast_fp16")];
            string var_573_equation_0 = const()[name = string("op_573_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 40, 1, 4096]> var_573_cast_fp16 = einsum(equation = var_573_equation_0, values = (var_507_cast_fp16, var_561_cast_fp16))[name = string("op_573_cast_fp16")];
            string var_575_equation_0 = const()[name = string("op_575_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 40, 1, 4096]> var_575_cast_fp16 = einsum(equation = var_575_equation_0, values = (var_511_cast_fp16, var_562_cast_fp16))[name = string("op_575_cast_fp16")];
            string var_577_equation_0 = const()[name = string("op_577_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 40, 1, 4096]> var_577_cast_fp16 = einsum(equation = var_577_equation_0, values = (var_515_cast_fp16, var_563_cast_fp16))[name = string("op_577_cast_fp16")];
            string var_579_equation_0 = const()[name = string("op_579_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 40, 1, 4096]> var_579_cast_fp16 = einsum(equation = var_579_equation_0, values = (var_519_cast_fp16, var_564_cast_fp16))[name = string("op_579_cast_fp16")];
            string var_581_equation_0 = const()[name = string("op_581_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 40, 1, 4096]> var_581_cast_fp16 = einsum(equation = var_581_equation_0, values = (var_523_cast_fp16, var_565_cast_fp16))[name = string("op_581_cast_fp16")];
            bool input_27_interleave_0 = const()[name = string("input_27_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 320, 1, 4096]> input_27_cast_fp16 = concat(axis = var_110, interleave = input_27_interleave_0, values = (var_567_cast_fp16, var_569_cast_fp16, var_571_cast_fp16, var_573_cast_fp16, var_575_cast_fp16, var_577_cast_fp16, var_579_cast_fp16, var_581_cast_fp16))[name = string("input_27_cast_fp16")];
            string var_591_pad_type_0 = const()[name = string("op_591_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_591_strides_0 = const()[name = string("op_591_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_591_pad_0 = const()[name = string("op_591_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_591_dilations_0 = const()[name = string("op_591_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_591_groups_0 = const()[name = string("op_591_groups_0"), val = int32(1)];
            tensor<fp16, [320, 320, 1, 1]> down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = string("down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [320, 320, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10884544)))];
            tensor<fp16, [320]> down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = string("down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11089408)))];
            tensor<fp16, [1, 320, 1, 4096]> var_591_cast_fp16 = conv(bias = down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_591_dilations_0, groups = var_591_groups_0, pad = var_591_pad_0, pad_type = var_591_pad_type_0, strides = var_591_strides_0, weight = down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_27_cast_fp16)[name = string("op_591_cast_fp16")];
            tensor<fp16, [1, 320, 1, 4096]> inputs_5_cast_fp16 = add(x = var_591_cast_fp16, y = inputs_3_cast_fp16)[name = string("inputs_5_cast_fp16")];
            tensor<int32, [1]> input_29_axes_0 = const()[name = string("input_29_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [320]> input_29_gamma_0_to_fp16 = const()[name = string("input_29_gamma_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11090112)))];
            tensor<fp16, [320]> input_29_beta_0_to_fp16 = const()[name = string("input_29_beta_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11090816)))];
            fp16 var_601_to_fp16 = const()[name = string("op_601_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 320, 1, 4096]> input_29_cast_fp16 = layer_norm(axes = input_29_axes_0, beta = input_29_beta_0_to_fp16, epsilon = var_601_to_fp16, gamma = input_29_gamma_0_to_fp16, x = inputs_5_cast_fp16)[name = string("input_29_cast_fp16")];
            string var_621_pad_type_0 = const()[name = string("op_621_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_621_strides_0 = const()[name = string("op_621_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_621_pad_0 = const()[name = string("op_621_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_621_dilations_0 = const()[name = string("op_621_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_621_groups_0 = const()[name = string("op_621_groups_0"), val = int32(1)];
            tensor<fp16, [2560, 320, 1, 1]> down_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = string("down_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [2560, 320, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11091520)))];
            tensor<fp16, [2560]> down_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = string("down_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12729984)))];
            tensor<fp16, [1, 2560, 1, 4096]> var_621_cast_fp16 = conv(bias = down_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_621_dilations_0, groups = var_621_groups_0, pad = var_621_pad_0, pad_type = var_621_pad_type_0, strides = var_621_strides_0, weight = down_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_29_cast_fp16)[name = string("op_621_cast_fp16")];
            tensor<int32, [2]> var_622_split_sizes_0 = const()[name = string("op_622_split_sizes_0"), val = tensor<int32, [2]>([1280, 1280])];
            int32 var_622_axis_0 = const()[name = string("op_622_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1280, 1, 4096]> var_622_cast_fp16_0, tensor<fp16, [1, 1280, 1, 4096]> var_622_cast_fp16_1 = split(axis = var_622_axis_0, split_sizes = var_622_split_sizes_0, x = var_621_cast_fp16)[name = string("op_622_cast_fp16")];
            string var_624_mode_0 = const()[name = string("op_624_mode_0"), val = string("EXACT")];
            tensor<fp16, [1, 1280, 1, 4096]> var_624_cast_fp16 = gelu(mode = var_624_mode_0, x = var_622_cast_fp16_1)[name = string("op_624_cast_fp16")];
            tensor<fp16, [1, 1280, 1, 4096]> input_31_cast_fp16 = mul(x = var_622_cast_fp16_0, y = var_624_cast_fp16)[name = string("input_31_cast_fp16")];
            string var_632_pad_type_0 = const()[name = string("op_632_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_632_strides_0 = const()[name = string("op_632_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_632_pad_0 = const()[name = string("op_632_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_632_dilations_0 = const()[name = string("op_632_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_632_groups_0 = const()[name = string("op_632_groups_0"), val = int32(1)];
            tensor<fp16, [320, 1280, 1, 1]> down_blocks_0_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = string("down_blocks_0_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor<fp16, [320, 1280, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12735168)))];
            tensor<fp16, [320]> down_blocks_0_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = string("down_blocks_0_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13554432)))];
            tensor<fp16, [1, 320, 1, 4096]> var_632_cast_fp16 = conv(bias = down_blocks_0_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_632_dilations_0, groups = var_632_groups_0, pad = var_632_pad_0, pad_type = var_632_pad_type_0, strides = var_632_strides_0, weight = down_blocks_0_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_31_cast_fp16)[name = string("op_632_cast_fp16")];
            tensor<fp16, [1, 320, 1, 4096]> hidden_states_15_cast_fp16 = add(x = var_632_cast_fp16, y = inputs_5_cast_fp16)[name = string("hidden_states_15_cast_fp16")];
            tensor<int32, [4]> var_634 = const()[name = string("op_634"), val = tensor<int32, [4]>([1, 320, 64, 64])];
            tensor<fp16, [1, 320, 64, 64]> input_33_cast_fp16 = reshape(shape = var_634, x = hidden_states_15_cast_fp16)[name = string("input_33_cast_fp16")];
            string hidden_states_17_pad_type_0 = const()[name = string("hidden_states_17_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> hidden_states_17_strides_0 = const()[name = string("hidden_states_17_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_17_pad_0 = const()[name = string("hidden_states_17_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_17_dilations_0 = const()[name = string("hidden_states_17_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_17_groups_0 = const()[name = string("hidden_states_17_groups_0"), val = int32(1)];
            tensor<fp16, [320, 320, 1, 1]> down_blocks_0_attentions_0_proj_out_weight_to_fp16 = const()[name = string("down_blocks_0_attentions_0_proj_out_weight_to_fp16"), val = tensor<fp16, [320, 320, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13555136)))];
            tensor<fp16, [320]> down_blocks_0_attentions_0_proj_out_bias_to_fp16 = const()[name = string("down_blocks_0_attentions_0_proj_out_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13760000)))];
            tensor<fp16, [1, 320, 64, 64]> hidden_states_17_cast_fp16 = conv(bias = down_blocks_0_attentions_0_proj_out_bias_to_fp16, dilations = hidden_states_17_dilations_0, groups = hidden_states_17_groups_0, pad = hidden_states_17_pad_0, pad_type = hidden_states_17_pad_type_0, strides = hidden_states_17_strides_0, weight = down_blocks_0_attentions_0_proj_out_weight_to_fp16, x = input_33_cast_fp16)[name = string("hidden_states_17_cast_fp16")];
            tensor<fp16, [1, 320, 64, 64]> input_35_cast_fp16_1 = add(x = hidden_states_17_cast_fp16, y = hidden_states_5_cast_fp16)[name = string("input_35_cast_fp16")];
            string input_37_pad_type_0 = const()[name = string("input_37_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> input_37_pad_0 = const()[name = string("input_37_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> input_37_strides_0 = const()[name = string("input_37_strides_0"), val = tensor<int32, [2]>([2, 2])];
            tensor<int32, [2]> input_37_dilations_0 = const()[name = string("input_37_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_37_groups_0 = const()[name = string("input_37_groups_0"), val = int32(1)];
            tensor<fp16, [320, 320, 3, 3]> down_blocks_0_downsamplers_0_conv_weight_to_fp16 = const()[name = string("down_blocks_0_downsamplers_0_conv_weight_to_fp16"), val = tensor<fp16, [320, 320, 3, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13760704)))];
            tensor<fp16, [320]> down_blocks_0_downsamplers_0_conv_bias_to_fp16 = const()[name = string("down_blocks_0_downsamplers_0_conv_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15603968)))];
            tensor<fp16, [1, 320, 32, 32]> input_37_cast_fp16_1 = conv(bias = down_blocks_0_downsamplers_0_conv_bias_to_fp16, dilations = input_37_dilations_0, groups = input_37_groups_0, pad = input_37_pad_0, pad_type = input_37_pad_type_0, strides = input_37_strides_0, weight = down_blocks_0_downsamplers_0_conv_weight_to_fp16, x = input_35_cast_fp16_1)[name = string("input_37_cast_fp16")];
            int32 var_676 = const()[name = string("op_676"), val = int32(1)];
            tensor<int32, [5]> reshape_12_shape_0 = const()[name = string("reshape_12_shape_0"), val = tensor<int32, [5]>([1, 32, 10, 32, 32])];
            tensor<fp16, [1, 32, 10, 32, 32]> reshape_12_cast_fp16 = reshape(shape = reshape_12_shape_0, x = input_37_cast_fp16_1)[name = string("reshape_12_cast_fp16")];
            tensor<int32, [3]> reduce_mean_9_axes_0 = const()[name = string("reduce_mean_9_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            bool reduce_mean_9_keep_dims_0 = const()[name = string("reduce_mean_9_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 1, 1, 1]> reduce_mean_9_cast_fp16 = reduce_mean(axes = reduce_mean_9_axes_0, keep_dims = reduce_mean_9_keep_dims_0, x = reshape_12_cast_fp16)[name = string("reduce_mean_9_cast_fp16")];
            tensor<fp16, [1, 32, 10, 32, 32]> sub_6_cast_fp16 = sub(x = reshape_12_cast_fp16, y = reduce_mean_9_cast_fp16)[name = string("sub_6_cast_fp16")];
            tensor<fp16, [1, 32, 10, 32, 32]> square_3_cast_fp16 = square(x = sub_6_cast_fp16)[name = string("square_3_cast_fp16")];
            tensor<int32, [3]> reduce_mean_11_axes_0 = const()[name = string("reduce_mean_11_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            bool reduce_mean_11_keep_dims_0 = const()[name = string("reduce_mean_11_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 1, 1, 1]> reduce_mean_11_cast_fp16 = reduce_mean(axes = reduce_mean_11_axes_0, keep_dims = reduce_mean_11_keep_dims_0, x = square_3_cast_fp16)[name = string("reduce_mean_11_cast_fp16")];
            fp16 add_6_y_0_to_fp16 = const()[name = string("add_6_y_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 32, 1, 1, 1]> add_6_cast_fp16 = add(x = reduce_mean_11_cast_fp16, y = add_6_y_0_to_fp16)[name = string("add_6_cast_fp16")];
            tensor<fp16, [1, 32, 1, 1, 1]> sqrt_3_cast_fp16 = sqrt(x = add_6_cast_fp16)[name = string("sqrt_3_cast_fp16")];
            tensor<fp16, [1, 32, 10, 32, 32]> real_div_3_cast_fp16 = real_div(x = sub_6_cast_fp16, y = sqrt_3_cast_fp16)[name = string("real_div_3_cast_fp16")];
            tensor<int32, [4]> reshape_13_shape_0 = const()[name = string("reshape_13_shape_0"), val = tensor<int32, [4]>([1, 320, 32, 32])];
            tensor<fp16, [1, 320, 32, 32]> reshape_13_cast_fp16 = reshape(shape = reshape_13_shape_0, x = real_div_3_cast_fp16)[name = string("reshape_13_cast_fp16")];
            tensor<fp16, [320]> add_7_gamma_0_to_fp16 = const()[name = string("add_7_gamma_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15604672)))];
            tensor<fp16, [320]> add_7_beta_0_to_fp16 = const()[name = string("add_7_beta_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15605376)))];
            fp16 add_7_epsilon_0_to_fp16 = const()[name = string("add_7_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 320, 32, 32]> add_7_cast_fp16 = batch_norm(beta = add_7_beta_0_to_fp16, epsilon = add_7_epsilon_0_to_fp16, gamma = add_7_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_13_cast_fp16)[name = string("add_7_cast_fp16")];
            tensor<fp16, [1, 320, 32, 32]> input_41_cast_fp16 = silu(x = add_7_cast_fp16)[name = string("input_41_cast_fp16")];
            string hidden_states_19_pad_type_0 = const()[name = string("hidden_states_19_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> hidden_states_19_pad_0 = const()[name = string("hidden_states_19_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> hidden_states_19_strides_0 = const()[name = string("hidden_states_19_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> hidden_states_19_dilations_0 = const()[name = string("hidden_states_19_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_19_groups_0 = const()[name = string("hidden_states_19_groups_0"), val = int32(1)];
            tensor<fp16, [640, 320, 3, 3]> down_blocks_1_resnets_0_conv1_weight_to_fp16 = const()[name = string("down_blocks_1_resnets_0_conv1_weight_to_fp16"), val = tensor<fp16, [640, 320, 3, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15606080)))];
            tensor<fp16, [640]> down_blocks_1_resnets_0_conv1_bias_to_fp16 = const()[name = string("down_blocks_1_resnets_0_conv1_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19292544)))];
            tensor<fp16, [1, 640, 32, 32]> hidden_states_19_cast_fp16 = conv(bias = down_blocks_1_resnets_0_conv1_bias_to_fp16, dilations = hidden_states_19_dilations_0, groups = hidden_states_19_groups_0, pad = hidden_states_19_pad_0, pad_type = hidden_states_19_pad_type_0, strides = hidden_states_19_strides_0, weight = down_blocks_1_resnets_0_conv1_weight_to_fp16, x = input_41_cast_fp16)[name = string("hidden_states_19_cast_fp16")];
            string temb_3_pad_type_0 = const()[name = string("temb_3_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> temb_3_strides_0 = const()[name = string("temb_3_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> temb_3_pad_0 = const()[name = string("temb_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> temb_3_dilations_0 = const()[name = string("temb_3_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 temb_3_groups_0 = const()[name = string("temb_3_groups_0"), val = int32(1)];
            tensor<fp16, [640, 1280, 1, 1]> down_blocks_1_resnets_0_time_emb_proj_weight_to_fp16 = const()[name = string("down_blocks_1_resnets_0_time_emb_proj_weight_to_fp16"), val = tensor<fp16, [640, 1280, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19293888)))];
            tensor<fp16, [640]> down_blocks_1_resnets_0_time_emb_proj_bias_to_fp16 = const()[name = string("down_blocks_1_resnets_0_time_emb_proj_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20932352)))];
            tensor<fp16, [1, 640, 1, 1]> temb_3_cast_fp16 = conv(bias = down_blocks_1_resnets_0_time_emb_proj_bias_to_fp16, dilations = temb_3_dilations_0, groups = temb_3_groups_0, pad = temb_3_pad_0, pad_type = temb_3_pad_type_0, strides = temb_3_strides_0, weight = down_blocks_1_resnets_0_time_emb_proj_weight_to_fp16, x = input_15_cast_fp16_1)[name = string("temb_3_cast_fp16")];
            tensor<fp16, [1, 640, 32, 32]> input_45_cast_fp16 = add(x = hidden_states_19_cast_fp16, y = temb_3_cast_fp16)[name = string("input_45_cast_fp16")];
            tensor<int32, [5]> reshape_16_shape_0 = const()[name = string("reshape_16_shape_0"), val = tensor<int32, [5]>([1, 32, 20, 32, 32])];
            tensor<fp16, [1, 32, 20, 32, 32]> reshape_16_cast_fp16 = reshape(shape = reshape_16_shape_0, x = input_45_cast_fp16)[name = string("reshape_16_cast_fp16")];
            tensor<int32, [3]> reduce_mean_12_axes_0 = const()[name = string("reduce_mean_12_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            bool reduce_mean_12_keep_dims_0 = const()[name = string("reduce_mean_12_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 1, 1, 1]> reduce_mean_12_cast_fp16 = reduce_mean(axes = reduce_mean_12_axes_0, keep_dims = reduce_mean_12_keep_dims_0, x = reshape_16_cast_fp16)[name = string("reduce_mean_12_cast_fp16")];
            tensor<fp16, [1, 32, 20, 32, 32]> sub_8_cast_fp16 = sub(x = reshape_16_cast_fp16, y = reduce_mean_12_cast_fp16)[name = string("sub_8_cast_fp16")];
            tensor<fp16, [1, 32, 20, 32, 32]> square_4_cast_fp16 = square(x = sub_8_cast_fp16)[name = string("square_4_cast_fp16")];
            tensor<int32, [3]> reduce_mean_14_axes_0 = const()[name = string("reduce_mean_14_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            bool reduce_mean_14_keep_dims_0 = const()[name = string("reduce_mean_14_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 1, 1, 1]> reduce_mean_14_cast_fp16 = reduce_mean(axes = reduce_mean_14_axes_0, keep_dims = reduce_mean_14_keep_dims_0, x = square_4_cast_fp16)[name = string("reduce_mean_14_cast_fp16")];
            fp16 add_8_y_0_to_fp16 = const()[name = string("add_8_y_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 32, 1, 1, 1]> add_8_cast_fp16 = add(x = reduce_mean_14_cast_fp16, y = add_8_y_0_to_fp16)[name = string("add_8_cast_fp16")];
            tensor<fp16, [1, 32, 1, 1, 1]> sqrt_4_cast_fp16 = sqrt(x = add_8_cast_fp16)[name = string("sqrt_4_cast_fp16")];
            tensor<fp16, [1, 32, 20, 32, 32]> real_div_4_cast_fp16 = real_div(x = sub_8_cast_fp16, y = sqrt_4_cast_fp16)[name = string("real_div_4_cast_fp16")];
            tensor<int32, [4]> reshape_17_shape_0 = const()[name = string("reshape_17_shape_0"), val = tensor<int32, [4]>([1, 640, 32, 32])];
            tensor<fp16, [1, 640, 32, 32]> reshape_17_cast_fp16 = reshape(shape = reshape_17_shape_0, x = real_div_4_cast_fp16)[name = string("reshape_17_cast_fp16")];
            tensor<fp16, [640]> add_9_mean_0_to_fp16 = const()[name = string("add_9_mean_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20933696)))];
            tensor<fp16, [640]> add_9_variance_0_to_fp16 = const()[name = string("add_9_variance_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20935040)))];
            tensor<fp16, [640]> add_9_gamma_0_to_fp16 = const()[name = string("add_9_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20936384)))];
            tensor<fp16, [640]> add_9_beta_0_to_fp16 = const()[name = string("add_9_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20937728)))];
            fp16 add_9_epsilon_0_to_fp16 = const()[name = string("add_9_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 640, 32, 32]> add_9_cast_fp16 = batch_norm(beta = add_9_beta_0_to_fp16, epsilon = add_9_epsilon_0_to_fp16, gamma = add_9_gamma_0_to_fp16, mean = add_9_mean_0_to_fp16, variance = add_9_variance_0_to_fp16, x = reshape_17_cast_fp16)[name = string("add_9_cast_fp16")];
            tensor<fp16, [1, 640, 32, 32]> input_49_cast_fp16 = silu(x = add_9_cast_fp16)[name = string("input_49_cast_fp16")];
            string hidden_states_21_pad_type_0 = const()[name = string("hidden_states_21_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> hidden_states_21_pad_0 = const()[name = string("hidden_states_21_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> hidden_states_21_strides_0 = const()[name = string("hidden_states_21_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> hidden_states_21_dilations_0 = const()[name = string("hidden_states_21_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_21_groups_0 = const()[name = string("hidden_states_21_groups_0"), val = int32(1)];
            tensor<fp16, [640, 640, 3, 3]> down_blocks_1_resnets_0_conv2_weight_to_fp16 = const()[name = string("down_blocks_1_resnets_0_conv2_weight_to_fp16"), val = tensor<fp16, [640, 640, 3, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20939072)))];
            tensor<fp16, [640]> down_blocks_1_resnets_0_conv2_bias_to_fp16 = const()[name = string("down_blocks_1_resnets_0_conv2_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28311936)))];
            tensor<fp16, [1, 640, 32, 32]> hidden_states_21_cast_fp16 = conv(bias = down_blocks_1_resnets_0_conv2_bias_to_fp16, dilations = hidden_states_21_dilations_0, groups = hidden_states_21_groups_0, pad = hidden_states_21_pad_0, pad_type = hidden_states_21_pad_type_0, strides = hidden_states_21_strides_0, weight = down_blocks_1_resnets_0_conv2_weight_to_fp16, x = input_49_cast_fp16)[name = string("hidden_states_21_cast_fp16")];
            string x_1_pad_type_0 = const()[name = string("x_1_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> x_1_strides_0 = const()[name = string("x_1_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> x_1_pad_0 = const()[name = string("x_1_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> x_1_dilations_0 = const()[name = string("x_1_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 x_1_groups_0 = const()[name = string("x_1_groups_0"), val = int32(1)];
            tensor<fp16, [640, 320, 1, 1]> down_blocks_1_resnets_0_conv_shortcut_weight_to_fp16 = const()[name = string("down_blocks_1_resnets_0_conv_shortcut_weight_to_fp16"), val = tensor<fp16, [640, 320, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28313280)))];
            tensor<fp16, [640]> down_blocks_1_resnets_0_conv_shortcut_bias_to_fp16 = const()[name = string("down_blocks_1_resnets_0_conv_shortcut_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28722944)))];
            tensor<fp16, [1, 640, 32, 32]> x_1_cast_fp16 = conv(bias = down_blocks_1_resnets_0_conv_shortcut_bias_to_fp16, dilations = x_1_dilations_0, groups = x_1_groups_0, pad = x_1_pad_0, pad_type = x_1_pad_type_0, strides = x_1_strides_0, weight = down_blocks_1_resnets_0_conv_shortcut_weight_to_fp16, x = input_37_cast_fp16_1)[name = string("x_1_cast_fp16")];
            tensor<fp16, [1, 640, 32, 32]> hidden_states_23_cast_fp16 = add(x = x_1_cast_fp16, y = hidden_states_21_cast_fp16)[name = string("hidden_states_23_cast_fp16")];
            tensor<int32, [5]> reshape_20_shape_0 = const()[name = string("reshape_20_shape_0"), val = tensor<int32, [5]>([1, 32, 20, 32, 32])];
            tensor<fp16, [1, 32, 20, 32, 32]> reshape_20_cast_fp16 = reshape(shape = reshape_20_shape_0, x = hidden_states_23_cast_fp16)[name = string("reshape_20_cast_fp16")];
            tensor<int32, [3]> reduce_mean_15_axes_0 = const()[name = string("reduce_mean_15_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            bool reduce_mean_15_keep_dims_0 = const()[name = string("reduce_mean_15_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 1, 1, 1]> reduce_mean_15_cast_fp16 = reduce_mean(axes = reduce_mean_15_axes_0, keep_dims = reduce_mean_15_keep_dims_0, x = reshape_20_cast_fp16)[name = string("reduce_mean_15_cast_fp16")];
            tensor<fp16, [1, 32, 20, 32, 32]> sub_10_cast_fp16 = sub(x = reshape_20_cast_fp16, y = reduce_mean_15_cast_fp16)[name = string("sub_10_cast_fp16")];
            tensor<fp16, [1, 32, 20, 32, 32]> square_5_cast_fp16 = square(x = sub_10_cast_fp16)[name = string("square_5_cast_fp16")];
            tensor<int32, [3]> reduce_mean_17_axes_0 = const()[name = string("reduce_mean_17_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            bool reduce_mean_17_keep_dims_0 = const()[name = string("reduce_mean_17_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 1, 1, 1]> reduce_mean_17_cast_fp16 = reduce_mean(axes = reduce_mean_17_axes_0, keep_dims = reduce_mean_17_keep_dims_0, x = square_5_cast_fp16)[name = string("reduce_mean_17_cast_fp16")];
            fp16 add_10_y_0_to_fp16 = const()[name = string("add_10_y_0_to_fp16"), val = fp16(0x1.1p-20)];
            tensor<fp16, [1, 32, 1, 1, 1]> add_10_cast_fp16 = add(x = reduce_mean_17_cast_fp16, y = add_10_y_0_to_fp16)[name = string("add_10_cast_fp16")];
            tensor<fp16, [1, 32, 1, 1, 1]> sqrt_5_cast_fp16 = sqrt(x = add_10_cast_fp16)[name = string("sqrt_5_cast_fp16")];
            tensor<fp16, [1, 32, 20, 32, 32]> real_div_5_cast_fp16 = real_div(x = sub_10_cast_fp16, y = sqrt_5_cast_fp16)[name = string("real_div_5_cast_fp16")];
            tensor<int32, [4]> reshape_21_shape_0 = const()[name = string("reshape_21_shape_0"), val = tensor<int32, [4]>([1, 640, 32, 32])];
            tensor<fp16, [1, 640, 32, 32]> reshape_21_cast_fp16 = reshape(shape = reshape_21_shape_0, x = real_div_5_cast_fp16)[name = string("reshape_21_cast_fp16")];
            tensor<fp16, [640]> add_11_gamma_0_to_fp16 = const()[name = string("add_11_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28724288)))];
            tensor<fp16, [640]> add_11_beta_0_to_fp16 = const()[name = string("add_11_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28725632)))];
            fp16 add_11_epsilon_0_to_fp16 = const()[name = string("add_11_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 640, 32, 32]> add_11_cast_fp16 = batch_norm(beta = add_11_beta_0_to_fp16, epsilon = add_11_epsilon_0_to_fp16, gamma = add_11_gamma_0_to_fp16, mean = add_9_mean_0_to_fp16, variance = add_9_variance_0_to_fp16, x = reshape_21_cast_fp16)[name = string("add_11_cast_fp16")];
            string hidden_states_25_pad_type_0 = const()[name = string("hidden_states_25_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> hidden_states_25_strides_0 = const()[name = string("hidden_states_25_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_25_pad_0 = const()[name = string("hidden_states_25_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_25_dilations_0 = const()[name = string("hidden_states_25_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_25_groups_0 = const()[name = string("hidden_states_25_groups_0"), val = int32(1)];
            tensor<fp16, [640, 640, 1, 1]> down_blocks_1_attentions_0_proj_in_weight_to_fp16 = const()[name = string("down_blocks_1_attentions_0_proj_in_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28726976)))];
            tensor<fp16, [640]> down_blocks_1_attentions_0_proj_in_bias_to_fp16 = const()[name = string("down_blocks_1_attentions_0_proj_in_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29546240)))];
            tensor<fp16, [1, 640, 32, 32]> hidden_states_25_cast_fp16 = conv(bias = down_blocks_1_attentions_0_proj_in_bias_to_fp16, dilations = hidden_states_25_dilations_0, groups = hidden_states_25_groups_0, pad = hidden_states_25_pad_0, pad_type = hidden_states_25_pad_type_0, strides = hidden_states_25_strides_0, weight = down_blocks_1_attentions_0_proj_in_weight_to_fp16, x = add_11_cast_fp16)[name = string("hidden_states_25_cast_fp16")];
            tensor<int32, [4]> var_751 = const()[name = string("op_751"), val = tensor<int32, [4]>([1, 640, 1, 1024])];
            tensor<fp16, [1, 640, 1, 1024]> inputs_7_cast_fp16 = reshape(shape = var_751, x = hidden_states_25_cast_fp16)[name = string("inputs_7_cast_fp16")];
            tensor<int32, [1]> hidden_states_27_axes_0 = const()[name = string("hidden_states_27_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [640]> hidden_states_27_gamma_0_to_fp16 = const()[name = string("hidden_states_27_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29547584)))];
            tensor<fp16, [640]> hidden_states_27_beta_0_to_fp16 = const()[name = string("hidden_states_27_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29548928)))];
            fp16 var_767_to_fp16 = const()[name = string("op_767_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 640, 1, 1024]> hidden_states_27_cast_fp16 = layer_norm(axes = hidden_states_27_axes_0, beta = hidden_states_27_beta_0_to_fp16, epsilon = var_767_to_fp16, gamma = hidden_states_27_gamma_0_to_fp16, x = inputs_7_cast_fp16)[name = string("hidden_states_27_cast_fp16")];
            string q_5_pad_type_0 = const()[name = string("q_5_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> q_5_strides_0 = const()[name = string("q_5_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_5_pad_0 = const()[name = string("q_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_5_dilations_0 = const()[name = string("q_5_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 q_5_groups_0 = const()[name = string("q_5_groups_0"), val = int32(1)];
            tensor<fp16, [640, 640, 1, 1]> down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = string("down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29550272)))];
            tensor<fp16, [1, 640, 1, 1024]> q_5_cast_fp16 = conv(dilations = q_5_dilations_0, groups = q_5_groups_0, pad = q_5_pad_0, pad_type = q_5_pad_type_0, strides = q_5_strides_0, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_27_cast_fp16)[name = string("q_5_cast_fp16")];
            string k_9_pad_type_0 = const()[name = string("k_9_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> k_9_strides_0 = const()[name = string("k_9_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_9_pad_0 = const()[name = string("k_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_9_dilations_0 = const()[name = string("k_9_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 k_9_groups_0 = const()[name = string("k_9_groups_0"), val = int32(1)];
            tensor<fp16, [640, 640, 1, 1]> down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = string("down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30369536)))];
            tensor<fp16, [1, 640, 1, 1024]> k_9_cast_fp16 = conv(dilations = k_9_dilations_0, groups = k_9_groups_0, pad = k_9_pad_0, pad_type = k_9_pad_type_0, strides = k_9_strides_0, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_27_cast_fp16)[name = string("k_9_cast_fp16")];
            string v_5_pad_type_0 = const()[name = string("v_5_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> v_5_strides_0 = const()[name = string("v_5_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_5_pad_0 = const()[name = string("v_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_5_dilations_0 = const()[name = string("v_5_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 v_5_groups_0 = const()[name = string("v_5_groups_0"), val = int32(1)];
            tensor<fp16, [640, 640, 1, 1]> down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = string("down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31188800)))];
            tensor<fp16, [1, 640, 1, 1024]> v_5_cast_fp16 = conv(dilations = v_5_dilations_0, groups = v_5_groups_0, pad = v_5_pad_0, pad_type = v_5_pad_type_0, strides = v_5_strides_0, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_27_cast_fp16)[name = string("v_5_cast_fp16")];
            tensor<int32, [4]> var_800_begin_0 = const()[name = string("op_800_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_800_end_0 = const()[name = string("op_800_end_0"), val = tensor<int32, [4]>([1, 80, 1, 1024])];
            tensor<bool, [4]> var_800_end_mask_0 = const()[name = string("op_800_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 80, 1, 1024]> var_800_cast_fp16 = slice_by_index(begin = var_800_begin_0, end = var_800_end_0, end_mask = var_800_end_mask_0, x = q_5_cast_fp16)[name = string("op_800_cast_fp16")];
            tensor<int32, [4]> var_804_begin_0 = const()[name = string("op_804_begin_0"), val = tensor<int32, [4]>([0, 80, 0, 0])];
            tensor<int32, [4]> var_804_end_0 = const()[name = string("op_804_end_0"), val = tensor<int32, [4]>([1, 160, 1, 1024])];
            tensor<bool, [4]> var_804_end_mask_0 = const()[name = string("op_804_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 80, 1, 1024]> var_804_cast_fp16 = slice_by_index(begin = var_804_begin_0, end = var_804_end_0, end_mask = var_804_end_mask_0, x = q_5_cast_fp16)[name = string("op_804_cast_fp16")];
            tensor<int32, [4]> var_808_begin_0 = const()[name = string("op_808_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_808_end_0 = const()[name = string("op_808_end_0"), val = tensor<int32, [4]>([1, 240, 1, 1024])];
            tensor<bool, [4]> var_808_end_mask_0 = const()[name = string("op_808_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 80, 1, 1024]> var_808_cast_fp16 = slice_by_index(begin = var_808_begin_0, end = var_808_end_0, end_mask = var_808_end_mask_0, x = q_5_cast_fp16)[name = string("op_808_cast_fp16")];
            tensor<int32, [4]> var_812_begin_0 = const()[name = string("op_812_begin_0"), val = tensor<int32, [4]>([0, 240, 0, 0])];
            tensor<int32, [4]> var_812_end_0 = const()[name = string("op_812_end_0"), val = tensor<int32, [4]>([1, 320, 1, 1024])];
            tensor<bool, [4]> var_812_end_mask_0 = const()[name = string("op_812_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 80, 1, 1024]> var_812_cast_fp16 = slice_by_index(begin = var_812_begin_0, end = var_812_end_0, end_mask = var_812_end_mask_0, x = q_5_cast_fp16)[name = string("op_812_cast_fp16")];
            tensor<int32, [4]> var_816_begin_0 = const()[name = string("op_816_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_816_end_0 = const()[name = string("op_816_end_0"), val = tensor<int32, [4]>([1, 400, 1, 1024])];
            tensor<bool, [4]> var_816_end_mask_0 = const()[name = string("op_816_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 80, 1, 1024]> var_816_cast_fp16 = slice_by_index(begin = var_816_begin_0, end = var_816_end_0, end_mask = var_816_end_mask_0, x = q_5_cast_fp16)[name = string("op_816_cast_fp16")];
            tensor<int32, [4]> var_820_begin_0 = const()[name = string("op_820_begin_0"), val = tensor<int32, [4]>([0, 400, 0, 0])];
            tensor<int32, [4]> var_820_end_0 = const()[name = string("op_820_end_0"), val = tensor<int32, [4]>([1, 480, 1, 1024])];
            tensor<bool, [4]> var_820_end_mask_0 = const()[name = string("op_820_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 80, 1, 1024]> var_820_cast_fp16 = slice_by_index(begin = var_820_begin_0, end = var_820_end_0, end_mask = var_820_end_mask_0, x = q_5_cast_fp16)[name = string("op_820_cast_fp16")];
            tensor<int32, [4]> var_824_begin_0 = const()[name = string("op_824_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_824_end_0 = const()[name = string("op_824_end_0"), val = tensor<int32, [4]>([1, 560, 1, 1024])];
            tensor<bool, [4]> var_824_end_mask_0 = const()[name = string("op_824_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 80, 1, 1024]> var_824_cast_fp16 = slice_by_index(begin = var_824_begin_0, end = var_824_end_0, end_mask = var_824_end_mask_0, x = q_5_cast_fp16)[name = string("op_824_cast_fp16")];
            tensor<int32, [4]> var_828_begin_0 = const()[name = string("op_828_begin_0"), val = tensor<int32, [4]>([0, 560, 0, 0])];
            tensor<int32, [4]> var_828_end_0 = const()[name = string("op_828_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1024])];
            tensor<bool, [4]> var_828_end_mask_0 = const()[name = string("op_828_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 80, 1, 1024]> var_828_cast_fp16 = slice_by_index(begin = var_828_begin_0, end = var_828_end_0, end_mask = var_828_end_mask_0, x = q_5_cast_fp16)[name = string("op_828_cast_fp16")];
            tensor<int32, [4]> k_11_perm_0 = const()[name = string("k_11_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_835_begin_0 = const()[name = string("op_835_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_835_end_0 = const()[name = string("op_835_end_0"), val = tensor<int32, [4]>([1, 1024, 1, 80])];
            tensor<bool, [4]> var_835_end_mask_0 = const()[name = string("op_835_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 1024, 1, 640]> k_11_cast_fp16 = transpose(perm = k_11_perm_0, x = k_9_cast_fp16)[name = string("transpose_5")];
            tensor<fp16, [1, 1024, 1, 80]> var_835_cast_fp16 = slice_by_index(begin = var_835_begin_0, end = var_835_end_0, end_mask = var_835_end_mask_0, x = k_11_cast_fp16)[name = string("op_835_cast_fp16")];
            tensor<int32, [4]> var_839_begin_0 = const()[name = string("op_839_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 80])];
            tensor<int32, [4]> var_839_end_0 = const()[name = string("op_839_end_0"), val = tensor<int32, [4]>([1, 1024, 1, 160])];
            tensor<bool, [4]> var_839_end_mask_0 = const()[name = string("op_839_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 1024, 1, 80]> var_839_cast_fp16 = slice_by_index(begin = var_839_begin_0, end = var_839_end_0, end_mask = var_839_end_mask_0, x = k_11_cast_fp16)[name = string("op_839_cast_fp16")];
            tensor<int32, [4]> var_843_begin_0 = const()[name = string("op_843_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 160])];
            tensor<int32, [4]> var_843_end_0 = const()[name = string("op_843_end_0"), val = tensor<int32, [4]>([1, 1024, 1, 240])];
            tensor<bool, [4]> var_843_end_mask_0 = const()[name = string("op_843_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 1024, 1, 80]> var_843_cast_fp16 = slice_by_index(begin = var_843_begin_0, end = var_843_end_0, end_mask = var_843_end_mask_0, x = k_11_cast_fp16)[name = string("op_843_cast_fp16")];
            tensor<int32, [4]> var_847_begin_0 = const()[name = string("op_847_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 240])];
            tensor<int32, [4]> var_847_end_0 = const()[name = string("op_847_end_0"), val = tensor<int32, [4]>([1, 1024, 1, 320])];
            tensor<bool, [4]> var_847_end_mask_0 = const()[name = string("op_847_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 1024, 1, 80]> var_847_cast_fp16 = slice_by_index(begin = var_847_begin_0, end = var_847_end_0, end_mask = var_847_end_mask_0, x = k_11_cast_fp16)[name = string("op_847_cast_fp16")];
            tensor<int32, [4]> var_851_begin_0 = const()[name = string("op_851_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_851_end_0 = const()[name = string("op_851_end_0"), val = tensor<int32, [4]>([1, 1024, 1, 400])];
            tensor<bool, [4]> var_851_end_mask_0 = const()[name = string("op_851_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 1024, 1, 80]> var_851_cast_fp16 = slice_by_index(begin = var_851_begin_0, end = var_851_end_0, end_mask = var_851_end_mask_0, x = k_11_cast_fp16)[name = string("op_851_cast_fp16")];
            tensor<int32, [4]> var_855_begin_0 = const()[name = string("op_855_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 400])];
            tensor<int32, [4]> var_855_end_0 = const()[name = string("op_855_end_0"), val = tensor<int32, [4]>([1, 1024, 1, 480])];
            tensor<bool, [4]> var_855_end_mask_0 = const()[name = string("op_855_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 1024, 1, 80]> var_855_cast_fp16 = slice_by_index(begin = var_855_begin_0, end = var_855_end_0, end_mask = var_855_end_mask_0, x = k_11_cast_fp16)[name = string("op_855_cast_fp16")];
            tensor<int32, [4]> var_859_begin_0 = const()[name = string("op_859_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 480])];
            tensor<int32, [4]> var_859_end_0 = const()[name = string("op_859_end_0"), val = tensor<int32, [4]>([1, 1024, 1, 560])];
            tensor<bool, [4]> var_859_end_mask_0 = const()[name = string("op_859_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 1024, 1, 80]> var_859_cast_fp16 = slice_by_index(begin = var_859_begin_0, end = var_859_end_0, end_mask = var_859_end_mask_0, x = k_11_cast_fp16)[name = string("op_859_cast_fp16")];
            tensor<int32, [4]> var_863_begin_0 = const()[name = string("op_863_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 560])];
            tensor<int32, [4]> var_863_end_0 = const()[name = string("op_863_end_0"), val = tensor<int32, [4]>([1, 1024, 1, 1])];
            tensor<bool, [4]> var_863_end_mask_0 = const()[name = string("op_863_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 1024, 1, 80]> var_863_cast_fp16 = slice_by_index(begin = var_863_begin_0, end = var_863_end_0, end_mask = var_863_end_mask_0, x = k_11_cast_fp16)[name = string("op_863_cast_fp16")];
            tensor<int32, [4]> var_865_begin_0 = const()[name = string("op_865_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_865_end_0 = const()[name = string("op_865_end_0"), val = tensor<int32, [4]>([1, 80, 1, 1024])];
            tensor<bool, [4]> var_865_end_mask_0 = const()[name = string("op_865_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 80, 1, 1024]> var_865_cast_fp16 = slice_by_index(begin = var_865_begin_0, end = var_865_end_0, end_mask = var_865_end_mask_0, x = v_5_cast_fp16)[name = string("op_865_cast_fp16")];
            tensor<int32, [4]> var_869_begin_0 = const()[name = string("op_869_begin_0"), val = tensor<int32, [4]>([0, 80, 0, 0])];
            tensor<int32, [4]> var_869_end_0 = const()[name = string("op_869_end_0"), val = tensor<int32, [4]>([1, 160, 1, 1024])];
            tensor<bool, [4]> var_869_end_mask_0 = const()[name = string("op_869_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 80, 1, 1024]> var_869_cast_fp16 = slice_by_index(begin = var_869_begin_0, end = var_869_end_0, end_mask = var_869_end_mask_0, x = v_5_cast_fp16)[name = string("op_869_cast_fp16")];
            tensor<int32, [4]> var_873_begin_0 = const()[name = string("op_873_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_873_end_0 = const()[name = string("op_873_end_0"), val = tensor<int32, [4]>([1, 240, 1, 1024])];
            tensor<bool, [4]> var_873_end_mask_0 = const()[name = string("op_873_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 80, 1, 1024]> var_873_cast_fp16 = slice_by_index(begin = var_873_begin_0, end = var_873_end_0, end_mask = var_873_end_mask_0, x = v_5_cast_fp16)[name = string("op_873_cast_fp16")];
            tensor<int32, [4]> var_877_begin_0 = const()[name = string("op_877_begin_0"), val = tensor<int32, [4]>([0, 240, 0, 0])];
            tensor<int32, [4]> var_877_end_0 = const()[name = string("op_877_end_0"), val = tensor<int32, [4]>([1, 320, 1, 1024])];
            tensor<bool, [4]> var_877_end_mask_0 = const()[name = string("op_877_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 80, 1, 1024]> var_877_cast_fp16 = slice_by_index(begin = var_877_begin_0, end = var_877_end_0, end_mask = var_877_end_mask_0, x = v_5_cast_fp16)[name = string("op_877_cast_fp16")];
            tensor<int32, [4]> var_881_begin_0 = const()[name = string("op_881_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_881_end_0 = const()[name = string("op_881_end_0"), val = tensor<int32, [4]>([1, 400, 1, 1024])];
            tensor<bool, [4]> var_881_end_mask_0 = const()[name = string("op_881_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 80, 1, 1024]> var_881_cast_fp16 = slice_by_index(begin = var_881_begin_0, end = var_881_end_0, end_mask = var_881_end_mask_0, x = v_5_cast_fp16)[name = string("op_881_cast_fp16")];
            tensor<int32, [4]> var_885_begin_0 = const()[name = string("op_885_begin_0"), val = tensor<int32, [4]>([0, 400, 0, 0])];
            tensor<int32, [4]> var_885_end_0 = const()[name = string("op_885_end_0"), val = tensor<int32, [4]>([1, 480, 1, 1024])];
            tensor<bool, [4]> var_885_end_mask_0 = const()[name = string("op_885_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 80, 1, 1024]> var_885_cast_fp16 = slice_by_index(begin = var_885_begin_0, end = var_885_end_0, end_mask = var_885_end_mask_0, x = v_5_cast_fp16)[name = string("op_885_cast_fp16")];
            tensor<int32, [4]> var_889_begin_0 = const()[name = string("op_889_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_889_end_0 = const()[name = string("op_889_end_0"), val = tensor<int32, [4]>([1, 560, 1, 1024])];
            tensor<bool, [4]> var_889_end_mask_0 = const()[name = string("op_889_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 80, 1, 1024]> var_889_cast_fp16 = slice_by_index(begin = var_889_begin_0, end = var_889_end_0, end_mask = var_889_end_mask_0, x = v_5_cast_fp16)[name = string("op_889_cast_fp16")];
            tensor<int32, [4]> var_893_begin_0 = const()[name = string("op_893_begin_0"), val = tensor<int32, [4]>([0, 560, 0, 0])];
            tensor<int32, [4]> var_893_end_0 = const()[name = string("op_893_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1024])];
            tensor<bool, [4]> var_893_end_mask_0 = const()[name = string("op_893_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 80, 1, 1024]> var_893_cast_fp16 = slice_by_index(begin = var_893_begin_0, end = var_893_end_0, end_mask = var_893_end_mask_0, x = v_5_cast_fp16)[name = string("op_893_cast_fp16")];
            string var_897_equation_0 = const()[name = string("op_897_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1024, 1, 1024]> var_897_cast_fp16 = einsum(equation = var_897_equation_0, values = (var_835_cast_fp16, var_800_cast_fp16))[name = string("op_897_cast_fp16")];
            fp16 var_898_to_fp16 = const()[name = string("op_898_to_fp16"), val = fp16(0x1.cap-4)];
            tensor<fp16, [1, 1024, 1, 1024]> aw_33_cast_fp16 = mul(x = var_897_cast_fp16, y = var_898_to_fp16)[name = string("aw_33_cast_fp16")];
            string var_901_equation_0 = const()[name = string("op_901_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1024, 1, 1024]> var_901_cast_fp16 = einsum(equation = var_901_equation_0, values = (var_839_cast_fp16, var_804_cast_fp16))[name = string("op_901_cast_fp16")];
            fp16 var_902_to_fp16 = const()[name = string("op_902_to_fp16"), val = fp16(0x1.cap-4)];
            tensor<fp16, [1, 1024, 1, 1024]> aw_35_cast_fp16 = mul(x = var_901_cast_fp16, y = var_902_to_fp16)[name = string("aw_35_cast_fp16")];
            string var_905_equation_0 = const()[name = string("op_905_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1024, 1, 1024]> var_905_cast_fp16 = einsum(equation = var_905_equation_0, values = (var_843_cast_fp16, var_808_cast_fp16))[name = string("op_905_cast_fp16")];
            fp16 var_906_to_fp16 = const()[name = string("op_906_to_fp16"), val = fp16(0x1.cap-4)];
            tensor<fp16, [1, 1024, 1, 1024]> aw_37_cast_fp16 = mul(x = var_905_cast_fp16, y = var_906_to_fp16)[name = string("aw_37_cast_fp16")];
            string var_909_equation_0 = const()[name = string("op_909_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1024, 1, 1024]> var_909_cast_fp16 = einsum(equation = var_909_equation_0, values = (var_847_cast_fp16, var_812_cast_fp16))[name = string("op_909_cast_fp16")];
            fp16 var_910_to_fp16 = const()[name = string("op_910_to_fp16"), val = fp16(0x1.cap-4)];
            tensor<fp16, [1, 1024, 1, 1024]> aw_39_cast_fp16 = mul(x = var_909_cast_fp16, y = var_910_to_fp16)[name = string("aw_39_cast_fp16")];
            string var_913_equation_0 = const()[name = string("op_913_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1024, 1, 1024]> var_913_cast_fp16 = einsum(equation = var_913_equation_0, values = (var_851_cast_fp16, var_816_cast_fp16))[name = string("op_913_cast_fp16")];
            fp16 var_914_to_fp16 = const()[name = string("op_914_to_fp16"), val = fp16(0x1.cap-4)];
            tensor<fp16, [1, 1024, 1, 1024]> aw_41_cast_fp16 = mul(x = var_913_cast_fp16, y = var_914_to_fp16)[name = string("aw_41_cast_fp16")];
            string var_917_equation_0 = const()[name = string("op_917_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1024, 1, 1024]> var_917_cast_fp16 = einsum(equation = var_917_equation_0, values = (var_855_cast_fp16, var_820_cast_fp16))[name = string("op_917_cast_fp16")];
            fp16 var_918_to_fp16 = const()[name = string("op_918_to_fp16"), val = fp16(0x1.cap-4)];
            tensor<fp16, [1, 1024, 1, 1024]> aw_43_cast_fp16 = mul(x = var_917_cast_fp16, y = var_918_to_fp16)[name = string("aw_43_cast_fp16")];
            string var_921_equation_0 = const()[name = string("op_921_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1024, 1, 1024]> var_921_cast_fp16 = einsum(equation = var_921_equation_0, values = (var_859_cast_fp16, var_824_cast_fp16))[name = string("op_921_cast_fp16")];
            fp16 var_922_to_fp16 = const()[name = string("op_922_to_fp16"), val = fp16(0x1.cap-4)];
            tensor<fp16, [1, 1024, 1, 1024]> aw_45_cast_fp16 = mul(x = var_921_cast_fp16, y = var_922_to_fp16)[name = string("aw_45_cast_fp16")];
            string var_925_equation_0 = const()[name = string("op_925_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1024, 1, 1024]> var_925_cast_fp16 = einsum(equation = var_925_equation_0, values = (var_863_cast_fp16, var_828_cast_fp16))[name = string("op_925_cast_fp16")];
            fp16 var_926_to_fp16 = const()[name = string("op_926_to_fp16"), val = fp16(0x1.cap-4)];
            tensor<fp16, [1, 1024, 1, 1024]> aw_47_cast_fp16 = mul(x = var_925_cast_fp16, y = var_926_to_fp16)[name = string("aw_47_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1024]> var_928_cast_fp16 = softmax(axis = var_676, x = aw_33_cast_fp16)[name = string("op_928_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1024]> var_929_cast_fp16 = softmax(axis = var_676, x = aw_35_cast_fp16)[name = string("op_929_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1024]> var_930_cast_fp16 = softmax(axis = var_676, x = aw_37_cast_fp16)[name = string("op_930_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1024]> var_931_cast_fp16 = softmax(axis = var_676, x = aw_39_cast_fp16)[name = string("op_931_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1024]> var_932_cast_fp16 = softmax(axis = var_676, x = aw_41_cast_fp16)[name = string("op_932_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1024]> var_933_cast_fp16 = softmax(axis = var_676, x = aw_43_cast_fp16)[name = string("op_933_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1024]> var_934_cast_fp16 = softmax(axis = var_676, x = aw_45_cast_fp16)[name = string("op_934_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1024]> var_935_cast_fp16 = softmax(axis = var_676, x = aw_47_cast_fp16)[name = string("op_935_cast_fp16")];
            string var_937_equation_0 = const()[name = string("op_937_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 80, 1, 1024]> var_937_cast_fp16 = einsum(equation = var_937_equation_0, values = (var_865_cast_fp16, var_928_cast_fp16))[name = string("op_937_cast_fp16")];
            string var_939_equation_0 = const()[name = string("op_939_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 80, 1, 1024]> var_939_cast_fp16 = einsum(equation = var_939_equation_0, values = (var_869_cast_fp16, var_929_cast_fp16))[name = string("op_939_cast_fp16")];
            string var_941_equation_0 = const()[name = string("op_941_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 80, 1, 1024]> var_941_cast_fp16 = einsum(equation = var_941_equation_0, values = (var_873_cast_fp16, var_930_cast_fp16))[name = string("op_941_cast_fp16")];
            string var_943_equation_0 = const()[name = string("op_943_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 80, 1, 1024]> var_943_cast_fp16 = einsum(equation = var_943_equation_0, values = (var_877_cast_fp16, var_931_cast_fp16))[name = string("op_943_cast_fp16")];
            string var_945_equation_0 = const()[name = string("op_945_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 80, 1, 1024]> var_945_cast_fp16 = einsum(equation = var_945_equation_0, values = (var_881_cast_fp16, var_932_cast_fp16))[name = string("op_945_cast_fp16")];
            string var_947_equation_0 = const()[name = string("op_947_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 80, 1, 1024]> var_947_cast_fp16 = einsum(equation = var_947_equation_0, values = (var_885_cast_fp16, var_933_cast_fp16))[name = string("op_947_cast_fp16")];
            string var_949_equation_0 = const()[name = string("op_949_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 80, 1, 1024]> var_949_cast_fp16 = einsum(equation = var_949_equation_0, values = (var_889_cast_fp16, var_934_cast_fp16))[name = string("op_949_cast_fp16")];
            string var_951_equation_0 = const()[name = string("op_951_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 80, 1, 1024]> var_951_cast_fp16 = einsum(equation = var_951_equation_0, values = (var_893_cast_fp16, var_935_cast_fp16))[name = string("op_951_cast_fp16")];
            bool input_53_interleave_0 = const()[name = string("input_53_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 640, 1, 1024]> input_53_cast_fp16 = concat(axis = var_676, interleave = input_53_interleave_0, values = (var_937_cast_fp16, var_939_cast_fp16, var_941_cast_fp16, var_943_cast_fp16, var_945_cast_fp16, var_947_cast_fp16, var_949_cast_fp16, var_951_cast_fp16))[name = string("input_53_cast_fp16")];
            string var_961_pad_type_0 = const()[name = string("op_961_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_961_strides_0 = const()[name = string("op_961_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_961_pad_0 = const()[name = string("op_961_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_961_dilations_0 = const()[name = string("op_961_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_961_groups_0 = const()[name = string("op_961_groups_0"), val = int32(1)];
            tensor<fp16, [640, 640, 1, 1]> down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = string("down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32008064)))];
            tensor<fp16, [640]> down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = string("down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32827328)))];
            tensor<fp16, [1, 640, 1, 1024]> var_961_cast_fp16 = conv(bias = down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_961_dilations_0, groups = var_961_groups_0, pad = var_961_pad_0, pad_type = var_961_pad_type_0, strides = var_961_strides_0, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_53_cast_fp16)[name = string("op_961_cast_fp16")];
            tensor<fp16, [1, 640, 1, 1024]> inputs_9_cast_fp16 = add(x = var_961_cast_fp16, y = inputs_7_cast_fp16)[name = string("inputs_9_cast_fp16")];
            tensor<int32, [1]> hidden_states_29_axes_0 = const()[name = string("hidden_states_29_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [640]> hidden_states_29_gamma_0_to_fp16 = const()[name = string("hidden_states_29_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32828672)))];
            tensor<fp16, [640]> hidden_states_29_beta_0_to_fp16 = const()[name = string("hidden_states_29_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32830016)))];
            fp16 var_971_to_fp16 = const()[name = string("op_971_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 640, 1, 1024]> hidden_states_29_cast_fp16 = layer_norm(axes = hidden_states_29_axes_0, beta = hidden_states_29_beta_0_to_fp16, epsilon = var_971_to_fp16, gamma = hidden_states_29_gamma_0_to_fp16, x = inputs_9_cast_fp16)[name = string("hidden_states_29_cast_fp16")];
            string q_7_pad_type_0 = const()[name = string("q_7_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> q_7_strides_0 = const()[name = string("q_7_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_7_pad_0 = const()[name = string("q_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_7_dilations_0 = const()[name = string("q_7_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 q_7_groups_0 = const()[name = string("q_7_groups_0"), val = int32(1)];
            tensor<fp16, [640, 640, 1, 1]> down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = string("down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32831360)))];
            tensor<fp16, [1, 640, 1, 1024]> q_7_cast_fp16 = conv(dilations = q_7_dilations_0, groups = q_7_groups_0, pad = q_7_pad_0, pad_type = q_7_pad_type_0, strides = q_7_strides_0, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_29_cast_fp16)[name = string("q_7_cast_fp16")];
            string k_13_pad_type_0 = const()[name = string("k_13_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> k_13_strides_0 = const()[name = string("k_13_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_13_pad_0 = const()[name = string("k_13_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_13_dilations_0 = const()[name = string("k_13_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 k_13_groups_0 = const()[name = string("k_13_groups_0"), val = int32(1)];
            tensor<fp16, [640, 768, 1, 1]> down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = string("down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [640, 768, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33650624)))];
            tensor<fp16, [1, 640, 1, 77]> k_13_cast_fp16 = conv(dilations = k_13_dilations_0, groups = k_13_groups_0, pad = k_13_pad_0, pad_type = k_13_pad_type_0, strides = k_13_strides_0, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = string("k_13_cast_fp16")];
            string v_7_pad_type_0 = const()[name = string("v_7_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> v_7_strides_0 = const()[name = string("v_7_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_7_pad_0 = const()[name = string("v_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_7_dilations_0 = const()[name = string("v_7_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 v_7_groups_0 = const()[name = string("v_7_groups_0"), val = int32(1)];
            tensor<fp16, [640, 768, 1, 1]> down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = string("down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [640, 768, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34633728)))];
            tensor<fp16, [1, 640, 1, 77]> v_7_cast_fp16 = conv(dilations = v_7_dilations_0, groups = v_7_groups_0, pad = v_7_pad_0, pad_type = v_7_pad_type_0, strides = v_7_strides_0, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = string("v_7_cast_fp16")];
            tensor<int32, [4]> var_1004_begin_0 = const()[name = string("op_1004_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1004_end_0 = const()[name = string("op_1004_end_0"), val = tensor<int32, [4]>([1, 80, 1, 1024])];
            tensor<bool, [4]> var_1004_end_mask_0 = const()[name = string("op_1004_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 80, 1, 1024]> var_1004_cast_fp16 = slice_by_index(begin = var_1004_begin_0, end = var_1004_end_0, end_mask = var_1004_end_mask_0, x = q_7_cast_fp16)[name = string("op_1004_cast_fp16")];
            tensor<int32, [4]> var_1008_begin_0 = const()[name = string("op_1008_begin_0"), val = tensor<int32, [4]>([0, 80, 0, 0])];
            tensor<int32, [4]> var_1008_end_0 = const()[name = string("op_1008_end_0"), val = tensor<int32, [4]>([1, 160, 1, 1024])];
            tensor<bool, [4]> var_1008_end_mask_0 = const()[name = string("op_1008_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 80, 1, 1024]> var_1008_cast_fp16 = slice_by_index(begin = var_1008_begin_0, end = var_1008_end_0, end_mask = var_1008_end_mask_0, x = q_7_cast_fp16)[name = string("op_1008_cast_fp16")];
            tensor<int32, [4]> var_1012_begin_0 = const()[name = string("op_1012_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_1012_end_0 = const()[name = string("op_1012_end_0"), val = tensor<int32, [4]>([1, 240, 1, 1024])];
            tensor<bool, [4]> var_1012_end_mask_0 = const()[name = string("op_1012_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 80, 1, 1024]> var_1012_cast_fp16 = slice_by_index(begin = var_1012_begin_0, end = var_1012_end_0, end_mask = var_1012_end_mask_0, x = q_7_cast_fp16)[name = string("op_1012_cast_fp16")];
            tensor<int32, [4]> var_1016_begin_0 = const()[name = string("op_1016_begin_0"), val = tensor<int32, [4]>([0, 240, 0, 0])];
            tensor<int32, [4]> var_1016_end_0 = const()[name = string("op_1016_end_0"), val = tensor<int32, [4]>([1, 320, 1, 1024])];
            tensor<bool, [4]> var_1016_end_mask_0 = const()[name = string("op_1016_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 80, 1, 1024]> var_1016_cast_fp16 = slice_by_index(begin = var_1016_begin_0, end = var_1016_end_0, end_mask = var_1016_end_mask_0, x = q_7_cast_fp16)[name = string("op_1016_cast_fp16")];
            tensor<int32, [4]> var_1020_begin_0 = const()[name = string("op_1020_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_1020_end_0 = const()[name = string("op_1020_end_0"), val = tensor<int32, [4]>([1, 400, 1, 1024])];
            tensor<bool, [4]> var_1020_end_mask_0 = const()[name = string("op_1020_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 80, 1, 1024]> var_1020_cast_fp16 = slice_by_index(begin = var_1020_begin_0, end = var_1020_end_0, end_mask = var_1020_end_mask_0, x = q_7_cast_fp16)[name = string("op_1020_cast_fp16")];
            tensor<int32, [4]> var_1024_begin_0 = const()[name = string("op_1024_begin_0"), val = tensor<int32, [4]>([0, 400, 0, 0])];
            tensor<int32, [4]> var_1024_end_0 = const()[name = string("op_1024_end_0"), val = tensor<int32, [4]>([1, 480, 1, 1024])];
            tensor<bool, [4]> var_1024_end_mask_0 = const()[name = string("op_1024_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 80, 1, 1024]> var_1024_cast_fp16 = slice_by_index(begin = var_1024_begin_0, end = var_1024_end_0, end_mask = var_1024_end_mask_0, x = q_7_cast_fp16)[name = string("op_1024_cast_fp16")];
            tensor<int32, [4]> var_1028_begin_0 = const()[name = string("op_1028_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_1028_end_0 = const()[name = string("op_1028_end_0"), val = tensor<int32, [4]>([1, 560, 1, 1024])];
            tensor<bool, [4]> var_1028_end_mask_0 = const()[name = string("op_1028_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 80, 1, 1024]> var_1028_cast_fp16 = slice_by_index(begin = var_1028_begin_0, end = var_1028_end_0, end_mask = var_1028_end_mask_0, x = q_7_cast_fp16)[name = string("op_1028_cast_fp16")];
            tensor<int32, [4]> var_1032_begin_0 = const()[name = string("op_1032_begin_0"), val = tensor<int32, [4]>([0, 560, 0, 0])];
            tensor<int32, [4]> var_1032_end_0 = const()[name = string("op_1032_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1024])];
            tensor<bool, [4]> var_1032_end_mask_0 = const()[name = string("op_1032_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 80, 1, 1024]> var_1032_cast_fp16 = slice_by_index(begin = var_1032_begin_0, end = var_1032_end_0, end_mask = var_1032_end_mask_0, x = q_7_cast_fp16)[name = string("op_1032_cast_fp16")];
            tensor<int32, [4]> k_15_perm_0 = const()[name = string("k_15_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_1039_begin_0 = const()[name = string("op_1039_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1039_end_0 = const()[name = string("op_1039_end_0"), val = tensor<int32, [4]>([1, 77, 1, 80])];
            tensor<bool, [4]> var_1039_end_mask_0 = const()[name = string("op_1039_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 77, 1, 640]> k_15_cast_fp16 = transpose(perm = k_15_perm_0, x = k_13_cast_fp16)[name = string("transpose_4")];
            tensor<fp16, [1, 77, 1, 80]> var_1039_cast_fp16 = slice_by_index(begin = var_1039_begin_0, end = var_1039_end_0, end_mask = var_1039_end_mask_0, x = k_15_cast_fp16)[name = string("op_1039_cast_fp16")];
            tensor<int32, [4]> var_1043_begin_0 = const()[name = string("op_1043_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 80])];
            tensor<int32, [4]> var_1043_end_0 = const()[name = string("op_1043_end_0"), val = tensor<int32, [4]>([1, 77, 1, 160])];
            tensor<bool, [4]> var_1043_end_mask_0 = const()[name = string("op_1043_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 77, 1, 80]> var_1043_cast_fp16 = slice_by_index(begin = var_1043_begin_0, end = var_1043_end_0, end_mask = var_1043_end_mask_0, x = k_15_cast_fp16)[name = string("op_1043_cast_fp16")];
            tensor<int32, [4]> var_1047_begin_0 = const()[name = string("op_1047_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 160])];
            tensor<int32, [4]> var_1047_end_0 = const()[name = string("op_1047_end_0"), val = tensor<int32, [4]>([1, 77, 1, 240])];
            tensor<bool, [4]> var_1047_end_mask_0 = const()[name = string("op_1047_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 77, 1, 80]> var_1047_cast_fp16 = slice_by_index(begin = var_1047_begin_0, end = var_1047_end_0, end_mask = var_1047_end_mask_0, x = k_15_cast_fp16)[name = string("op_1047_cast_fp16")];
            tensor<int32, [4]> var_1051_begin_0 = const()[name = string("op_1051_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 240])];
            tensor<int32, [4]> var_1051_end_0 = const()[name = string("op_1051_end_0"), val = tensor<int32, [4]>([1, 77, 1, 320])];
            tensor<bool, [4]> var_1051_end_mask_0 = const()[name = string("op_1051_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 77, 1, 80]> var_1051_cast_fp16 = slice_by_index(begin = var_1051_begin_0, end = var_1051_end_0, end_mask = var_1051_end_mask_0, x = k_15_cast_fp16)[name = string("op_1051_cast_fp16")];
            tensor<int32, [4]> var_1055_begin_0 = const()[name = string("op_1055_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_1055_end_0 = const()[name = string("op_1055_end_0"), val = tensor<int32, [4]>([1, 77, 1, 400])];
            tensor<bool, [4]> var_1055_end_mask_0 = const()[name = string("op_1055_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 77, 1, 80]> var_1055_cast_fp16 = slice_by_index(begin = var_1055_begin_0, end = var_1055_end_0, end_mask = var_1055_end_mask_0, x = k_15_cast_fp16)[name = string("op_1055_cast_fp16")];
            tensor<int32, [4]> var_1059_begin_0 = const()[name = string("op_1059_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 400])];
            tensor<int32, [4]> var_1059_end_0 = const()[name = string("op_1059_end_0"), val = tensor<int32, [4]>([1, 77, 1, 480])];
            tensor<bool, [4]> var_1059_end_mask_0 = const()[name = string("op_1059_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 77, 1, 80]> var_1059_cast_fp16 = slice_by_index(begin = var_1059_begin_0, end = var_1059_end_0, end_mask = var_1059_end_mask_0, x = k_15_cast_fp16)[name = string("op_1059_cast_fp16")];
            tensor<int32, [4]> var_1063_begin_0 = const()[name = string("op_1063_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 480])];
            tensor<int32, [4]> var_1063_end_0 = const()[name = string("op_1063_end_0"), val = tensor<int32, [4]>([1, 77, 1, 560])];
            tensor<bool, [4]> var_1063_end_mask_0 = const()[name = string("op_1063_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 77, 1, 80]> var_1063_cast_fp16 = slice_by_index(begin = var_1063_begin_0, end = var_1063_end_0, end_mask = var_1063_end_mask_0, x = k_15_cast_fp16)[name = string("op_1063_cast_fp16")];
            tensor<int32, [4]> var_1067_begin_0 = const()[name = string("op_1067_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 560])];
            tensor<int32, [4]> var_1067_end_0 = const()[name = string("op_1067_end_0"), val = tensor<int32, [4]>([1, 77, 1, 1])];
            tensor<bool, [4]> var_1067_end_mask_0 = const()[name = string("op_1067_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 77, 1, 80]> var_1067_cast_fp16 = slice_by_index(begin = var_1067_begin_0, end = var_1067_end_0, end_mask = var_1067_end_mask_0, x = k_15_cast_fp16)[name = string("op_1067_cast_fp16")];
            tensor<int32, [4]> var_1069_begin_0 = const()[name = string("op_1069_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1069_end_0 = const()[name = string("op_1069_end_0"), val = tensor<int32, [4]>([1, 80, 1, 77])];
            tensor<bool, [4]> var_1069_end_mask_0 = const()[name = string("op_1069_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 80, 1, 77]> var_1069_cast_fp16 = slice_by_index(begin = var_1069_begin_0, end = var_1069_end_0, end_mask = var_1069_end_mask_0, x = v_7_cast_fp16)[name = string("op_1069_cast_fp16")];
            tensor<int32, [4]> var_1073_begin_0 = const()[name = string("op_1073_begin_0"), val = tensor<int32, [4]>([0, 80, 0, 0])];
            tensor<int32, [4]> var_1073_end_0 = const()[name = string("op_1073_end_0"), val = tensor<int32, [4]>([1, 160, 1, 77])];
            tensor<bool, [4]> var_1073_end_mask_0 = const()[name = string("op_1073_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 80, 1, 77]> var_1073_cast_fp16 = slice_by_index(begin = var_1073_begin_0, end = var_1073_end_0, end_mask = var_1073_end_mask_0, x = v_7_cast_fp16)[name = string("op_1073_cast_fp16")];
            tensor<int32, [4]> var_1077_begin_0 = const()[name = string("op_1077_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_1077_end_0 = const()[name = string("op_1077_end_0"), val = tensor<int32, [4]>([1, 240, 1, 77])];
            tensor<bool, [4]> var_1077_end_mask_0 = const()[name = string("op_1077_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 80, 1, 77]> var_1077_cast_fp16 = slice_by_index(begin = var_1077_begin_0, end = var_1077_end_0, end_mask = var_1077_end_mask_0, x = v_7_cast_fp16)[name = string("op_1077_cast_fp16")];
            tensor<int32, [4]> var_1081_begin_0 = const()[name = string("op_1081_begin_0"), val = tensor<int32, [4]>([0, 240, 0, 0])];
            tensor<int32, [4]> var_1081_end_0 = const()[name = string("op_1081_end_0"), val = tensor<int32, [4]>([1, 320, 1, 77])];
            tensor<bool, [4]> var_1081_end_mask_0 = const()[name = string("op_1081_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 80, 1, 77]> var_1081_cast_fp16 = slice_by_index(begin = var_1081_begin_0, end = var_1081_end_0, end_mask = var_1081_end_mask_0, x = v_7_cast_fp16)[name = string("op_1081_cast_fp16")];
            tensor<int32, [4]> var_1085_begin_0 = const()[name = string("op_1085_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_1085_end_0 = const()[name = string("op_1085_end_0"), val = tensor<int32, [4]>([1, 400, 1, 77])];
            tensor<bool, [4]> var_1085_end_mask_0 = const()[name = string("op_1085_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 80, 1, 77]> var_1085_cast_fp16 = slice_by_index(begin = var_1085_begin_0, end = var_1085_end_0, end_mask = var_1085_end_mask_0, x = v_7_cast_fp16)[name = string("op_1085_cast_fp16")];
            tensor<int32, [4]> var_1089_begin_0 = const()[name = string("op_1089_begin_0"), val = tensor<int32, [4]>([0, 400, 0, 0])];
            tensor<int32, [4]> var_1089_end_0 = const()[name = string("op_1089_end_0"), val = tensor<int32, [4]>([1, 480, 1, 77])];
            tensor<bool, [4]> var_1089_end_mask_0 = const()[name = string("op_1089_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 80, 1, 77]> var_1089_cast_fp16 = slice_by_index(begin = var_1089_begin_0, end = var_1089_end_0, end_mask = var_1089_end_mask_0, x = v_7_cast_fp16)[name = string("op_1089_cast_fp16")];
            tensor<int32, [4]> var_1093_begin_0 = const()[name = string("op_1093_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_1093_end_0 = const()[name = string("op_1093_end_0"), val = tensor<int32, [4]>([1, 560, 1, 77])];
            tensor<bool, [4]> var_1093_end_mask_0 = const()[name = string("op_1093_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 80, 1, 77]> var_1093_cast_fp16 = slice_by_index(begin = var_1093_begin_0, end = var_1093_end_0, end_mask = var_1093_end_mask_0, x = v_7_cast_fp16)[name = string("op_1093_cast_fp16")];
            tensor<int32, [4]> var_1097_begin_0 = const()[name = string("op_1097_begin_0"), val = tensor<int32, [4]>([0, 560, 0, 0])];
            tensor<int32, [4]> var_1097_end_0 = const()[name = string("op_1097_end_0"), val = tensor<int32, [4]>([1, 1, 1, 77])];
            tensor<bool, [4]> var_1097_end_mask_0 = const()[name = string("op_1097_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 80, 1, 77]> var_1097_cast_fp16 = slice_by_index(begin = var_1097_begin_0, end = var_1097_end_0, end_mask = var_1097_end_mask_0, x = v_7_cast_fp16)[name = string("op_1097_cast_fp16")];
            string var_1101_equation_0 = const()[name = string("op_1101_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 77, 1, 1024]> var_1101_cast_fp16 = einsum(equation = var_1101_equation_0, values = (var_1039_cast_fp16, var_1004_cast_fp16))[name = string("op_1101_cast_fp16")];
            fp16 var_1102_to_fp16 = const()[name = string("op_1102_to_fp16"), val = fp16(0x1.cap-4)];
            tensor<fp16, [1, 77, 1, 1024]> aw_49_cast_fp16 = mul(x = var_1101_cast_fp16, y = var_1102_to_fp16)[name = string("aw_49_cast_fp16")];
            string var_1105_equation_0 = const()[name = string("op_1105_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 77, 1, 1024]> var_1105_cast_fp16 = einsum(equation = var_1105_equation_0, values = (var_1043_cast_fp16, var_1008_cast_fp16))[name = string("op_1105_cast_fp16")];
            fp16 var_1106_to_fp16 = const()[name = string("op_1106_to_fp16"), val = fp16(0x1.cap-4)];
            tensor<fp16, [1, 77, 1, 1024]> aw_51_cast_fp16 = mul(x = var_1105_cast_fp16, y = var_1106_to_fp16)[name = string("aw_51_cast_fp16")];
            string var_1109_equation_0 = const()[name = string("op_1109_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 77, 1, 1024]> var_1109_cast_fp16 = einsum(equation = var_1109_equation_0, values = (var_1047_cast_fp16, var_1012_cast_fp16))[name = string("op_1109_cast_fp16")];
            fp16 var_1110_to_fp16 = const()[name = string("op_1110_to_fp16"), val = fp16(0x1.cap-4)];
            tensor<fp16, [1, 77, 1, 1024]> aw_53_cast_fp16 = mul(x = var_1109_cast_fp16, y = var_1110_to_fp16)[name = string("aw_53_cast_fp16")];
            string var_1113_equation_0 = const()[name = string("op_1113_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 77, 1, 1024]> var_1113_cast_fp16 = einsum(equation = var_1113_equation_0, values = (var_1051_cast_fp16, var_1016_cast_fp16))[name = string("op_1113_cast_fp16")];
            fp16 var_1114_to_fp16 = const()[name = string("op_1114_to_fp16"), val = fp16(0x1.cap-4)];
            tensor<fp16, [1, 77, 1, 1024]> aw_55_cast_fp16 = mul(x = var_1113_cast_fp16, y = var_1114_to_fp16)[name = string("aw_55_cast_fp16")];
            string var_1117_equation_0 = const()[name = string("op_1117_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 77, 1, 1024]> var_1117_cast_fp16 = einsum(equation = var_1117_equation_0, values = (var_1055_cast_fp16, var_1020_cast_fp16))[name = string("op_1117_cast_fp16")];
            fp16 var_1118_to_fp16 = const()[name = string("op_1118_to_fp16"), val = fp16(0x1.cap-4)];
            tensor<fp16, [1, 77, 1, 1024]> aw_57_cast_fp16 = mul(x = var_1117_cast_fp16, y = var_1118_to_fp16)[name = string("aw_57_cast_fp16")];
            string var_1121_equation_0 = const()[name = string("op_1121_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 77, 1, 1024]> var_1121_cast_fp16 = einsum(equation = var_1121_equation_0, values = (var_1059_cast_fp16, var_1024_cast_fp16))[name = string("op_1121_cast_fp16")];
            fp16 var_1122_to_fp16 = const()[name = string("op_1122_to_fp16"), val = fp16(0x1.cap-4)];
            tensor<fp16, [1, 77, 1, 1024]> aw_59_cast_fp16 = mul(x = var_1121_cast_fp16, y = var_1122_to_fp16)[name = string("aw_59_cast_fp16")];
            string var_1125_equation_0 = const()[name = string("op_1125_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 77, 1, 1024]> var_1125_cast_fp16 = einsum(equation = var_1125_equation_0, values = (var_1063_cast_fp16, var_1028_cast_fp16))[name = string("op_1125_cast_fp16")];
            fp16 var_1126_to_fp16 = const()[name = string("op_1126_to_fp16"), val = fp16(0x1.cap-4)];
            tensor<fp16, [1, 77, 1, 1024]> aw_61_cast_fp16 = mul(x = var_1125_cast_fp16, y = var_1126_to_fp16)[name = string("aw_61_cast_fp16")];
            string var_1129_equation_0 = const()[name = string("op_1129_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 77, 1, 1024]> var_1129_cast_fp16 = einsum(equation = var_1129_equation_0, values = (var_1067_cast_fp16, var_1032_cast_fp16))[name = string("op_1129_cast_fp16")];
            fp16 var_1130_to_fp16 = const()[name = string("op_1130_to_fp16"), val = fp16(0x1.cap-4)];
            tensor<fp16, [1, 77, 1, 1024]> aw_63_cast_fp16 = mul(x = var_1129_cast_fp16, y = var_1130_to_fp16)[name = string("aw_63_cast_fp16")];
            tensor<fp16, [1, 77, 1, 1024]> var_1132_cast_fp16 = softmax(axis = var_676, x = aw_49_cast_fp16)[name = string("op_1132_cast_fp16")];
            tensor<fp16, [1, 77, 1, 1024]> var_1133_cast_fp16 = softmax(axis = var_676, x = aw_51_cast_fp16)[name = string("op_1133_cast_fp16")];
            tensor<fp16, [1, 77, 1, 1024]> var_1134_cast_fp16 = softmax(axis = var_676, x = aw_53_cast_fp16)[name = string("op_1134_cast_fp16")];
            tensor<fp16, [1, 77, 1, 1024]> var_1135_cast_fp16 = softmax(axis = var_676, x = aw_55_cast_fp16)[name = string("op_1135_cast_fp16")];
            tensor<fp16, [1, 77, 1, 1024]> var_1136_cast_fp16 = softmax(axis = var_676, x = aw_57_cast_fp16)[name = string("op_1136_cast_fp16")];
            tensor<fp16, [1, 77, 1, 1024]> var_1137_cast_fp16 = softmax(axis = var_676, x = aw_59_cast_fp16)[name = string("op_1137_cast_fp16")];
            tensor<fp16, [1, 77, 1, 1024]> var_1138_cast_fp16 = softmax(axis = var_676, x = aw_61_cast_fp16)[name = string("op_1138_cast_fp16")];
            tensor<fp16, [1, 77, 1, 1024]> var_1139_cast_fp16 = softmax(axis = var_676, x = aw_63_cast_fp16)[name = string("op_1139_cast_fp16")];
            string var_1141_equation_0 = const()[name = string("op_1141_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 80, 1, 1024]> var_1141_cast_fp16 = einsum(equation = var_1141_equation_0, values = (var_1069_cast_fp16, var_1132_cast_fp16))[name = string("op_1141_cast_fp16")];
            string var_1143_equation_0 = const()[name = string("op_1143_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 80, 1, 1024]> var_1143_cast_fp16 = einsum(equation = var_1143_equation_0, values = (var_1073_cast_fp16, var_1133_cast_fp16))[name = string("op_1143_cast_fp16")];
            string var_1145_equation_0 = const()[name = string("op_1145_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 80, 1, 1024]> var_1145_cast_fp16 = einsum(equation = var_1145_equation_0, values = (var_1077_cast_fp16, var_1134_cast_fp16))[name = string("op_1145_cast_fp16")];
            string var_1147_equation_0 = const()[name = string("op_1147_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 80, 1, 1024]> var_1147_cast_fp16 = einsum(equation = var_1147_equation_0, values = (var_1081_cast_fp16, var_1135_cast_fp16))[name = string("op_1147_cast_fp16")];
            string var_1149_equation_0 = const()[name = string("op_1149_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 80, 1, 1024]> var_1149_cast_fp16 = einsum(equation = var_1149_equation_0, values = (var_1085_cast_fp16, var_1136_cast_fp16))[name = string("op_1149_cast_fp16")];
            string var_1151_equation_0 = const()[name = string("op_1151_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 80, 1, 1024]> var_1151_cast_fp16 = einsum(equation = var_1151_equation_0, values = (var_1089_cast_fp16, var_1137_cast_fp16))[name = string("op_1151_cast_fp16")];
            string var_1153_equation_0 = const()[name = string("op_1153_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 80, 1, 1024]> var_1153_cast_fp16 = einsum(equation = var_1153_equation_0, values = (var_1093_cast_fp16, var_1138_cast_fp16))[name = string("op_1153_cast_fp16")];
            string var_1155_equation_0 = const()[name = string("op_1155_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 80, 1, 1024]> var_1155_cast_fp16 = einsum(equation = var_1155_equation_0, values = (var_1097_cast_fp16, var_1139_cast_fp16))[name = string("op_1155_cast_fp16")];
            bool input_55_interleave_0 = const()[name = string("input_55_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 640, 1, 1024]> input_55_cast_fp16 = concat(axis = var_676, interleave = input_55_interleave_0, values = (var_1141_cast_fp16, var_1143_cast_fp16, var_1145_cast_fp16, var_1147_cast_fp16, var_1149_cast_fp16, var_1151_cast_fp16, var_1153_cast_fp16, var_1155_cast_fp16))[name = string("input_55_cast_fp16")];
            string var_1165_pad_type_0 = const()[name = string("op_1165_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_1165_strides_0 = const()[name = string("op_1165_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_1165_pad_0 = const()[name = string("op_1165_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_1165_dilations_0 = const()[name = string("op_1165_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_1165_groups_0 = const()[name = string("op_1165_groups_0"), val = int32(1)];
            tensor<fp16, [640, 640, 1, 1]> down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = string("down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35616832)))];
            tensor<fp16, [640]> down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = string("down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36436096)))];
            tensor<fp16, [1, 640, 1, 1024]> var_1165_cast_fp16 = conv(bias = down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_1165_dilations_0, groups = var_1165_groups_0, pad = var_1165_pad_0, pad_type = var_1165_pad_type_0, strides = var_1165_strides_0, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_55_cast_fp16)[name = string("op_1165_cast_fp16")];
            tensor<fp16, [1, 640, 1, 1024]> inputs_11_cast_fp16 = add(x = var_1165_cast_fp16, y = inputs_9_cast_fp16)[name = string("inputs_11_cast_fp16")];
            tensor<int32, [1]> input_57_axes_0 = const()[name = string("input_57_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [640]> input_57_gamma_0_to_fp16 = const()[name = string("input_57_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36437440)))];
            tensor<fp16, [640]> input_57_beta_0_to_fp16 = const()[name = string("input_57_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36438784)))];
            fp16 var_1175_to_fp16 = const()[name = string("op_1175_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 640, 1, 1024]> input_57_cast_fp16 = layer_norm(axes = input_57_axes_0, beta = input_57_beta_0_to_fp16, epsilon = var_1175_to_fp16, gamma = input_57_gamma_0_to_fp16, x = inputs_11_cast_fp16)[name = string("input_57_cast_fp16")];
            string var_1195_pad_type_0 = const()[name = string("op_1195_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_1195_strides_0 = const()[name = string("op_1195_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_1195_pad_0 = const()[name = string("op_1195_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_1195_dilations_0 = const()[name = string("op_1195_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_1195_groups_0 = const()[name = string("op_1195_groups_0"), val = int32(1)];
            tensor<fp16, [5120, 640, 1, 1]> down_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = string("down_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [5120, 640, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36440128)))];
            tensor<fp16, [5120]> down_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = string("down_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42993792)))];
            tensor<fp16, [1, 5120, 1, 1024]> var_1195_cast_fp16 = conv(bias = down_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_1195_dilations_0, groups = var_1195_groups_0, pad = var_1195_pad_0, pad_type = var_1195_pad_type_0, strides = var_1195_strides_0, weight = down_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_57_cast_fp16)[name = string("op_1195_cast_fp16")];
            tensor<int32, [2]> var_1196_split_sizes_0 = const()[name = string("op_1196_split_sizes_0"), val = tensor<int32, [2]>([2560, 2560])];
            int32 var_1196_axis_0 = const()[name = string("op_1196_axis_0"), val = int32(1)];
            tensor<fp16, [1, 2560, 1, 1024]> var_1196_cast_fp16_0, tensor<fp16, [1, 2560, 1, 1024]> var_1196_cast_fp16_1 = split(axis = var_1196_axis_0, split_sizes = var_1196_split_sizes_0, x = var_1195_cast_fp16)[name = string("op_1196_cast_fp16")];
            string var_1198_mode_0 = const()[name = string("op_1198_mode_0"), val = string("EXACT")];
            tensor<fp16, [1, 2560, 1, 1024]> var_1198_cast_fp16 = gelu(mode = var_1198_mode_0, x = var_1196_cast_fp16_1)[name = string("op_1198_cast_fp16")];
            tensor<fp16, [1, 2560, 1, 1024]> input_59_cast_fp16 = mul(x = var_1196_cast_fp16_0, y = var_1198_cast_fp16)[name = string("input_59_cast_fp16")];
            string var_1206_pad_type_0 = const()[name = string("op_1206_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_1206_strides_0 = const()[name = string("op_1206_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_1206_pad_0 = const()[name = string("op_1206_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_1206_dilations_0 = const()[name = string("op_1206_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_1206_groups_0 = const()[name = string("op_1206_groups_0"), val = int32(1)];
            tensor<fp16, [640, 2560, 1, 1]> down_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = string("down_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor<fp16, [640, 2560, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43004096)))];
            tensor<fp16, [640]> down_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = string("down_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46280960)))];
            tensor<fp16, [1, 640, 1, 1024]> var_1206_cast_fp16 = conv(bias = down_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_1206_dilations_0, groups = var_1206_groups_0, pad = var_1206_pad_0, pad_type = var_1206_pad_type_0, strides = var_1206_strides_0, weight = down_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_59_cast_fp16)[name = string("op_1206_cast_fp16")];
            tensor<fp16, [1, 640, 1, 1024]> hidden_states_33_cast_fp16 = add(x = var_1206_cast_fp16, y = inputs_11_cast_fp16)[name = string("hidden_states_33_cast_fp16")];
            tensor<int32, [4]> var_1208 = const()[name = string("op_1208"), val = tensor<int32, [4]>([1, 640, 32, 32])];
            tensor<fp16, [1, 640, 32, 32]> input_61_cast_fp16 = reshape(shape = var_1208, x = hidden_states_33_cast_fp16)[name = string("input_61_cast_fp16")];
            string hidden_states_35_pad_type_0 = const()[name = string("hidden_states_35_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> hidden_states_35_strides_0 = const()[name = string("hidden_states_35_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_35_pad_0 = const()[name = string("hidden_states_35_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_35_dilations_0 = const()[name = string("hidden_states_35_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_35_groups_0 = const()[name = string("hidden_states_35_groups_0"), val = int32(1)];
            tensor<fp16, [640, 640, 1, 1]> down_blocks_1_attentions_0_proj_out_weight_to_fp16 = const()[name = string("down_blocks_1_attentions_0_proj_out_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46282304)))];
            tensor<fp16, [640]> down_blocks_1_attentions_0_proj_out_bias_to_fp16 = const()[name = string("down_blocks_1_attentions_0_proj_out_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47101568)))];
            tensor<fp16, [1, 640, 32, 32]> hidden_states_35_cast_fp16 = conv(bias = down_blocks_1_attentions_0_proj_out_bias_to_fp16, dilations = hidden_states_35_dilations_0, groups = hidden_states_35_groups_0, pad = hidden_states_35_pad_0, pad_type = hidden_states_35_pad_type_0, strides = hidden_states_35_strides_0, weight = down_blocks_1_attentions_0_proj_out_weight_to_fp16, x = input_61_cast_fp16)[name = string("hidden_states_35_cast_fp16")];
            tensor<fp16, [1, 640, 32, 32]> input_63_cast_fp16_1 = add(x = hidden_states_35_cast_fp16, y = hidden_states_23_cast_fp16)[name = string("input_63_cast_fp16")];
            string input_65_pad_type_0 = const()[name = string("input_65_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> input_65_pad_0 = const()[name = string("input_65_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> input_65_strides_0 = const()[name = string("input_65_strides_0"), val = tensor<int32, [2]>([2, 2])];
            tensor<int32, [2]> input_65_dilations_0 = const()[name = string("input_65_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_65_groups_0 = const()[name = string("input_65_groups_0"), val = int32(1)];
            tensor<fp16, [640, 640, 3, 3]> down_blocks_1_downsamplers_0_conv_weight_to_fp16 = const()[name = string("down_blocks_1_downsamplers_0_conv_weight_to_fp16"), val = tensor<fp16, [640, 640, 3, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47102912)))];
            tensor<fp16, [640]> down_blocks_1_downsamplers_0_conv_bias_to_fp16 = const()[name = string("down_blocks_1_downsamplers_0_conv_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54475776)))];
            tensor<fp16, [1, 640, 16, 16]> input_65_cast_fp16_1 = conv(bias = down_blocks_1_downsamplers_0_conv_bias_to_fp16, dilations = input_65_dilations_0, groups = input_65_groups_0, pad = input_65_pad_0, pad_type = input_65_pad_type_0, strides = input_65_strides_0, weight = down_blocks_1_downsamplers_0_conv_weight_to_fp16, x = input_63_cast_fp16_1)[name = string("input_65_cast_fp16")];
            int32 var_1250 = const()[name = string("op_1250"), val = int32(1)];
            tensor<int32, [5]> reshape_24_shape_0 = const()[name = string("reshape_24_shape_0"), val = tensor<int32, [5]>([1, 32, 20, 16, 16])];
            tensor<fp16, [1, 32, 20, 16, 16]> reshape_24_cast_fp16 = reshape(shape = reshape_24_shape_0, x = input_65_cast_fp16_1)[name = string("reshape_24_cast_fp16")];
            tensor<int32, [3]> reduce_mean_18_axes_0 = const()[name = string("reduce_mean_18_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            bool reduce_mean_18_keep_dims_0 = const()[name = string("reduce_mean_18_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 1, 1, 1]> reduce_mean_18_cast_fp16 = reduce_mean(axes = reduce_mean_18_axes_0, keep_dims = reduce_mean_18_keep_dims_0, x = reshape_24_cast_fp16)[name = string("reduce_mean_18_cast_fp16")];
            tensor<fp16, [1, 32, 20, 16, 16]> sub_12_cast_fp16 = sub(x = reshape_24_cast_fp16, y = reduce_mean_18_cast_fp16)[name = string("sub_12_cast_fp16")];
            tensor<fp16, [1, 32, 20, 16, 16]> square_6_cast_fp16 = square(x = sub_12_cast_fp16)[name = string("square_6_cast_fp16")];
            tensor<int32, [3]> reduce_mean_20_axes_0 = const()[name = string("reduce_mean_20_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            bool reduce_mean_20_keep_dims_0 = const()[name = string("reduce_mean_20_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 1, 1, 1]> reduce_mean_20_cast_fp16 = reduce_mean(axes = reduce_mean_20_axes_0, keep_dims = reduce_mean_20_keep_dims_0, x = square_6_cast_fp16)[name = string("reduce_mean_20_cast_fp16")];
            fp16 add_12_y_0_to_fp16 = const()[name = string("add_12_y_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 32, 1, 1, 1]> add_12_cast_fp16 = add(x = reduce_mean_20_cast_fp16, y = add_12_y_0_to_fp16)[name = string("add_12_cast_fp16")];
            tensor<fp16, [1, 32, 1, 1, 1]> sqrt_6_cast_fp16 = sqrt(x = add_12_cast_fp16)[name = string("sqrt_6_cast_fp16")];
            tensor<fp16, [1, 32, 20, 16, 16]> real_div_6_cast_fp16 = real_div(x = sub_12_cast_fp16, y = sqrt_6_cast_fp16)[name = string("real_div_6_cast_fp16")];
            tensor<int32, [4]> reshape_25_shape_0 = const()[name = string("reshape_25_shape_0"), val = tensor<int32, [4]>([1, 640, 16, 16])];
            tensor<fp16, [1, 640, 16, 16]> reshape_25_cast_fp16 = reshape(shape = reshape_25_shape_0, x = real_div_6_cast_fp16)[name = string("reshape_25_cast_fp16")];
            tensor<fp16, [640]> add_13_gamma_0_to_fp16 = const()[name = string("add_13_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54477120)))];
            tensor<fp16, [640]> add_13_beta_0_to_fp16 = const()[name = string("add_13_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54478464)))];
            fp16 add_13_epsilon_0_to_fp16 = const()[name = string("add_13_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 640, 16, 16]> add_13_cast_fp16 = batch_norm(beta = add_13_beta_0_to_fp16, epsilon = add_13_epsilon_0_to_fp16, gamma = add_13_gamma_0_to_fp16, mean = add_9_mean_0_to_fp16, variance = add_9_variance_0_to_fp16, x = reshape_25_cast_fp16)[name = string("add_13_cast_fp16")];
            tensor<fp16, [1, 640, 16, 16]> input_69_cast_fp16 = silu(x = add_13_cast_fp16)[name = string("input_69_cast_fp16")];
            string hidden_states_37_pad_type_0 = const()[name = string("hidden_states_37_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> hidden_states_37_pad_0 = const()[name = string("hidden_states_37_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> hidden_states_37_strides_0 = const()[name = string("hidden_states_37_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> hidden_states_37_dilations_0 = const()[name = string("hidden_states_37_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_37_groups_0 = const()[name = string("hidden_states_37_groups_0"), val = int32(1)];
            tensor<fp16, [1280, 640, 3, 3]> down_blocks_2_resnets_0_conv1_weight_to_fp16 = const()[name = string("down_blocks_2_resnets_0_conv1_weight_to_fp16"), val = tensor<fp16, [1280, 640, 3, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54479808)))];
            tensor<fp16, [1280]> down_blocks_2_resnets_0_conv1_bias_to_fp16 = const()[name = string("down_blocks_2_resnets_0_conv1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(69225472)))];
            tensor<fp16, [1, 1280, 16, 16]> hidden_states_37_cast_fp16 = conv(bias = down_blocks_2_resnets_0_conv1_bias_to_fp16, dilations = hidden_states_37_dilations_0, groups = hidden_states_37_groups_0, pad = hidden_states_37_pad_0, pad_type = hidden_states_37_pad_type_0, strides = hidden_states_37_strides_0, weight = down_blocks_2_resnets_0_conv1_weight_to_fp16, x = input_69_cast_fp16)[name = string("hidden_states_37_cast_fp16")];
            string temb_5_pad_type_0 = const()[name = string("temb_5_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> temb_5_strides_0 = const()[name = string("temb_5_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> temb_5_pad_0 = const()[name = string("temb_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> temb_5_dilations_0 = const()[name = string("temb_5_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 temb_5_groups_0 = const()[name = string("temb_5_groups_0"), val = int32(1)];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_resnets_0_time_emb_proj_weight_to_fp16 = const()[name = string("down_blocks_2_resnets_0_time_emb_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(69228096)))];
            tensor<fp16, [1280]> down_blocks_2_resnets_0_time_emb_proj_bias_to_fp16 = const()[name = string("down_blocks_2_resnets_0_time_emb_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72504960)))];
            tensor<fp16, [1, 1280, 1, 1]> temb_5_cast_fp16 = conv(bias = down_blocks_2_resnets_0_time_emb_proj_bias_to_fp16, dilations = temb_5_dilations_0, groups = temb_5_groups_0, pad = temb_5_pad_0, pad_type = temb_5_pad_type_0, strides = temb_5_strides_0, weight = down_blocks_2_resnets_0_time_emb_proj_weight_to_fp16, x = input_15_cast_fp16_1)[name = string("temb_5_cast_fp16")];
            tensor<fp16, [1, 1280, 16, 16]> input_73_cast_fp16 = add(x = hidden_states_37_cast_fp16, y = temb_5_cast_fp16)[name = string("input_73_cast_fp16")];
            tensor<int32, [5]> reshape_28_shape_0 = const()[name = string("reshape_28_shape_0"), val = tensor<int32, [5]>([1, 32, 40, 16, 16])];
            tensor<fp16, [1, 32, 40, 16, 16]> reshape_28_cast_fp16 = reshape(shape = reshape_28_shape_0, x = input_73_cast_fp16)[name = string("reshape_28_cast_fp16")];
            tensor<int32, [3]> reduce_mean_21_axes_0 = const()[name = string("reduce_mean_21_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            bool reduce_mean_21_keep_dims_0 = const()[name = string("reduce_mean_21_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 1, 1, 1]> reduce_mean_21_cast_fp16 = reduce_mean(axes = reduce_mean_21_axes_0, keep_dims = reduce_mean_21_keep_dims_0, x = reshape_28_cast_fp16)[name = string("reduce_mean_21_cast_fp16")];
            tensor<fp16, [1, 32, 40, 16, 16]> sub_14_cast_fp16 = sub(x = reshape_28_cast_fp16, y = reduce_mean_21_cast_fp16)[name = string("sub_14_cast_fp16")];
            tensor<fp16, [1, 32, 40, 16, 16]> square_7_cast_fp16 = square(x = sub_14_cast_fp16)[name = string("square_7_cast_fp16")];
            tensor<int32, [3]> reduce_mean_23_axes_0 = const()[name = string("reduce_mean_23_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            bool reduce_mean_23_keep_dims_0 = const()[name = string("reduce_mean_23_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 1, 1, 1]> reduce_mean_23_cast_fp16 = reduce_mean(axes = reduce_mean_23_axes_0, keep_dims = reduce_mean_23_keep_dims_0, x = square_7_cast_fp16)[name = string("reduce_mean_23_cast_fp16")];
            fp16 add_14_y_0_to_fp16 = const()[name = string("add_14_y_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 32, 1, 1, 1]> add_14_cast_fp16 = add(x = reduce_mean_23_cast_fp16, y = add_14_y_0_to_fp16)[name = string("add_14_cast_fp16")];
            tensor<fp16, [1, 32, 1, 1, 1]> sqrt_7_cast_fp16 = sqrt(x = add_14_cast_fp16)[name = string("sqrt_7_cast_fp16")];
            tensor<fp16, [1, 32, 40, 16, 16]> real_div_7_cast_fp16 = real_div(x = sub_14_cast_fp16, y = sqrt_7_cast_fp16)[name = string("real_div_7_cast_fp16")];
            tensor<int32, [4]> reshape_29_shape_0 = const()[name = string("reshape_29_shape_0"), val = tensor<int32, [4]>([1, 1280, 16, 16])];
            tensor<fp16, [1, 1280, 16, 16]> reshape_29_cast_fp16 = reshape(shape = reshape_29_shape_0, x = real_div_7_cast_fp16)[name = string("reshape_29_cast_fp16")];
            tensor<fp16, [1280]> add_15_mean_0_to_fp16 = const()[name = string("add_15_mean_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72507584)))];
            tensor<fp16, [1280]> add_15_variance_0_to_fp16 = const()[name = string("add_15_variance_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72510208)))];
            tensor<fp16, [1280]> add_15_gamma_0_to_fp16 = const()[name = string("add_15_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72512832)))];
            tensor<fp16, [1280]> add_15_beta_0_to_fp16 = const()[name = string("add_15_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72515456)))];
            fp16 add_15_epsilon_0_to_fp16 = const()[name = string("add_15_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1280, 16, 16]> add_15_cast_fp16 = batch_norm(beta = add_15_beta_0_to_fp16, epsilon = add_15_epsilon_0_to_fp16, gamma = add_15_gamma_0_to_fp16, mean = add_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_29_cast_fp16)[name = string("add_15_cast_fp16")];
            tensor<fp16, [1, 1280, 16, 16]> input_77_cast_fp16 = silu(x = add_15_cast_fp16)[name = string("input_77_cast_fp16")];
            string hidden_states_39_pad_type_0 = const()[name = string("hidden_states_39_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> hidden_states_39_pad_0 = const()[name = string("hidden_states_39_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> hidden_states_39_strides_0 = const()[name = string("hidden_states_39_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> hidden_states_39_dilations_0 = const()[name = string("hidden_states_39_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_39_groups_0 = const()[name = string("hidden_states_39_groups_0"), val = int32(1)];
            tensor<fp16, [1280, 1280, 3, 3]> down_blocks_2_resnets_0_conv2_weight_to_fp16 = const()[name = string("down_blocks_2_resnets_0_conv2_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 3, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72518080)))];
            tensor<fp16, [1280]> down_blocks_2_resnets_0_conv2_bias_to_fp16 = const()[name = string("down_blocks_2_resnets_0_conv2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102009344)))];
            tensor<fp16, [1, 1280, 16, 16]> hidden_states_39_cast_fp16 = conv(bias = down_blocks_2_resnets_0_conv2_bias_to_fp16, dilations = hidden_states_39_dilations_0, groups = hidden_states_39_groups_0, pad = hidden_states_39_pad_0, pad_type = hidden_states_39_pad_type_0, strides = hidden_states_39_strides_0, weight = down_blocks_2_resnets_0_conv2_weight_to_fp16, x = input_77_cast_fp16)[name = string("hidden_states_39_cast_fp16")];
            string x_3_pad_type_0 = const()[name = string("x_3_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> x_3_strides_0 = const()[name = string("x_3_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> x_3_pad_0 = const()[name = string("x_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> x_3_dilations_0 = const()[name = string("x_3_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 x_3_groups_0 = const()[name = string("x_3_groups_0"), val = int32(1)];
            tensor<fp16, [1280, 640, 1, 1]> down_blocks_2_resnets_0_conv_shortcut_weight_to_fp16 = const()[name = string("down_blocks_2_resnets_0_conv_shortcut_weight_to_fp16"), val = tensor<fp16, [1280, 640, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102011968)))];
            tensor<fp16, [1280]> down_blocks_2_resnets_0_conv_shortcut_bias_to_fp16 = const()[name = string("down_blocks_2_resnets_0_conv_shortcut_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103650432)))];
            tensor<fp16, [1, 1280, 16, 16]> x_3_cast_fp16 = conv(bias = down_blocks_2_resnets_0_conv_shortcut_bias_to_fp16, dilations = x_3_dilations_0, groups = x_3_groups_0, pad = x_3_pad_0, pad_type = x_3_pad_type_0, strides = x_3_strides_0, weight = down_blocks_2_resnets_0_conv_shortcut_weight_to_fp16, x = input_65_cast_fp16_1)[name = string("x_3_cast_fp16")];
            tensor<fp16, [1, 1280, 16, 16]> hidden_states_41_cast_fp16 = add(x = x_3_cast_fp16, y = hidden_states_39_cast_fp16)[name = string("hidden_states_41_cast_fp16")];
            tensor<int32, [5]> reshape_32_shape_0 = const()[name = string("reshape_32_shape_0"), val = tensor<int32, [5]>([1, 32, 40, 16, 16])];
            tensor<fp16, [1, 32, 40, 16, 16]> reshape_32_cast_fp16 = reshape(shape = reshape_32_shape_0, x = hidden_states_41_cast_fp16)[name = string("reshape_32_cast_fp16")];
            tensor<int32, [3]> reduce_mean_24_axes_0 = const()[name = string("reduce_mean_24_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            bool reduce_mean_24_keep_dims_0 = const()[name = string("reduce_mean_24_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 1, 1, 1]> reduce_mean_24_cast_fp16 = reduce_mean(axes = reduce_mean_24_axes_0, keep_dims = reduce_mean_24_keep_dims_0, x = reshape_32_cast_fp16)[name = string("reduce_mean_24_cast_fp16")];
            tensor<fp16, [1, 32, 40, 16, 16]> sub_16_cast_fp16 = sub(x = reshape_32_cast_fp16, y = reduce_mean_24_cast_fp16)[name = string("sub_16_cast_fp16")];
            tensor<fp16, [1, 32, 40, 16, 16]> square_8_cast_fp16 = square(x = sub_16_cast_fp16)[name = string("square_8_cast_fp16")];
            tensor<int32, [3]> reduce_mean_26_axes_0 = const()[name = string("reduce_mean_26_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            bool reduce_mean_26_keep_dims_0 = const()[name = string("reduce_mean_26_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 1, 1, 1]> reduce_mean_26_cast_fp16 = reduce_mean(axes = reduce_mean_26_axes_0, keep_dims = reduce_mean_26_keep_dims_0, x = square_8_cast_fp16)[name = string("reduce_mean_26_cast_fp16")];
            fp16 add_16_y_0_to_fp16 = const()[name = string("add_16_y_0_to_fp16"), val = fp16(0x1.1p-20)];
            tensor<fp16, [1, 32, 1, 1, 1]> add_16_cast_fp16 = add(x = reduce_mean_26_cast_fp16, y = add_16_y_0_to_fp16)[name = string("add_16_cast_fp16")];
            tensor<fp16, [1, 32, 1, 1, 1]> sqrt_8_cast_fp16 = sqrt(x = add_16_cast_fp16)[name = string("sqrt_8_cast_fp16")];
            tensor<fp16, [1, 32, 40, 16, 16]> real_div_8_cast_fp16 = real_div(x = sub_16_cast_fp16, y = sqrt_8_cast_fp16)[name = string("real_div_8_cast_fp16")];
            tensor<int32, [4]> reshape_33_shape_0 = const()[name = string("reshape_33_shape_0"), val = tensor<int32, [4]>([1, 1280, 16, 16])];
            tensor<fp16, [1, 1280, 16, 16]> reshape_33_cast_fp16 = reshape(shape = reshape_33_shape_0, x = real_div_8_cast_fp16)[name = string("reshape_33_cast_fp16")];
            tensor<fp16, [1280]> add_17_gamma_0_to_fp16 = const()[name = string("add_17_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103653056)))];
            tensor<fp16, [1280]> add_17_beta_0_to_fp16 = const()[name = string("add_17_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103655680)))];
            fp16 add_17_epsilon_0_to_fp16 = const()[name = string("add_17_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1280, 16, 16]> add_17_cast_fp16 = batch_norm(beta = add_17_beta_0_to_fp16, epsilon = add_17_epsilon_0_to_fp16, gamma = add_17_gamma_0_to_fp16, mean = add_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_33_cast_fp16)[name = string("add_17_cast_fp16")];
            string hidden_states_43_pad_type_0 = const()[name = string("hidden_states_43_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> hidden_states_43_strides_0 = const()[name = string("hidden_states_43_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_43_pad_0 = const()[name = string("hidden_states_43_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_43_dilations_0 = const()[name = string("hidden_states_43_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_43_groups_0 = const()[name = string("hidden_states_43_groups_0"), val = int32(1)];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_proj_in_weight_to_fp16 = const()[name = string("down_blocks_2_attentions_0_proj_in_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103658304)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_0_proj_in_bias_to_fp16 = const()[name = string("down_blocks_2_attentions_0_proj_in_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106935168)))];
            tensor<fp16, [1, 1280, 16, 16]> hidden_states_43_cast_fp16 = conv(bias = down_blocks_2_attentions_0_proj_in_bias_to_fp16, dilations = hidden_states_43_dilations_0, groups = hidden_states_43_groups_0, pad = hidden_states_43_pad_0, pad_type = hidden_states_43_pad_type_0, strides = hidden_states_43_strides_0, weight = down_blocks_2_attentions_0_proj_in_weight_to_fp16, x = add_17_cast_fp16)[name = string("hidden_states_43_cast_fp16")];
            tensor<int32, [4]> var_1323 = const()[name = string("op_1323"), val = tensor<int32, [4]>([1, 1280, 1, 256])];
            tensor<fp16, [1, 1280, 1, 256]> inputs_13_cast_fp16 = reshape(shape = var_1323, x = hidden_states_43_cast_fp16)[name = string("inputs_13_cast_fp16")];
            tensor<int32, [1]> hidden_states_45_axes_0 = const()[name = string("hidden_states_45_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_45_gamma_0_to_fp16 = const()[name = string("hidden_states_45_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106937792)))];
            tensor<fp16, [1280]> hidden_states_45_beta_0_to_fp16 = const()[name = string("hidden_states_45_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106940416)))];
            fp16 var_1339_to_fp16 = const()[name = string("op_1339_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1280, 1, 256]> hidden_states_45_cast_fp16 = layer_norm(axes = hidden_states_45_axes_0, beta = hidden_states_45_beta_0_to_fp16, epsilon = var_1339_to_fp16, gamma = hidden_states_45_gamma_0_to_fp16, x = inputs_13_cast_fp16)[name = string("hidden_states_45_cast_fp16")];
            string q_9_pad_type_0 = const()[name = string("q_9_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> q_9_strides_0 = const()[name = string("q_9_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_9_pad_0 = const()[name = string("q_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_9_dilations_0 = const()[name = string("q_9_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 q_9_groups_0 = const()[name = string("q_9_groups_0"), val = int32(1)];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = string("down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106943040)))];
            tensor<fp16, [1, 1280, 1, 256]> q_9_cast_fp16 = conv(dilations = q_9_dilations_0, groups = q_9_groups_0, pad = q_9_pad_0, pad_type = q_9_pad_type_0, strides = q_9_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_45_cast_fp16)[name = string("q_9_cast_fp16")];
            string k_17_pad_type_0 = const()[name = string("k_17_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> k_17_strides_0 = const()[name = string("k_17_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_17_pad_0 = const()[name = string("k_17_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_17_dilations_0 = const()[name = string("k_17_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 k_17_groups_0 = const()[name = string("k_17_groups_0"), val = int32(1)];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = string("down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110219904)))];
            tensor<fp16, [1, 1280, 1, 256]> k_17_cast_fp16 = conv(dilations = k_17_dilations_0, groups = k_17_groups_0, pad = k_17_pad_0, pad_type = k_17_pad_type_0, strides = k_17_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_45_cast_fp16)[name = string("k_17_cast_fp16")];
            string v_9_pad_type_0 = const()[name = string("v_9_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> v_9_strides_0 = const()[name = string("v_9_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_9_pad_0 = const()[name = string("v_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_9_dilations_0 = const()[name = string("v_9_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 v_9_groups_0 = const()[name = string("v_9_groups_0"), val = int32(1)];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = string("down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113496768)))];
            tensor<fp16, [1, 1280, 1, 256]> v_9_cast_fp16 = conv(dilations = v_9_dilations_0, groups = v_9_groups_0, pad = v_9_pad_0, pad_type = v_9_pad_type_0, strides = v_9_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_45_cast_fp16)[name = string("v_9_cast_fp16")];
            tensor<int32, [4]> var_1372_begin_0 = const()[name = string("op_1372_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1372_end_0 = const()[name = string("op_1372_end_0"), val = tensor<int32, [4]>([1, 160, 1, 256])];
            tensor<bool, [4]> var_1372_end_mask_0 = const()[name = string("op_1372_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 256]> var_1372_cast_fp16 = slice_by_index(begin = var_1372_begin_0, end = var_1372_end_0, end_mask = var_1372_end_mask_0, x = q_9_cast_fp16)[name = string("op_1372_cast_fp16")];
            tensor<int32, [4]> var_1376_begin_0 = const()[name = string("op_1376_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_1376_end_0 = const()[name = string("op_1376_end_0"), val = tensor<int32, [4]>([1, 320, 1, 256])];
            tensor<bool, [4]> var_1376_end_mask_0 = const()[name = string("op_1376_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 256]> var_1376_cast_fp16 = slice_by_index(begin = var_1376_begin_0, end = var_1376_end_0, end_mask = var_1376_end_mask_0, x = q_9_cast_fp16)[name = string("op_1376_cast_fp16")];
            tensor<int32, [4]> var_1380_begin_0 = const()[name = string("op_1380_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_1380_end_0 = const()[name = string("op_1380_end_0"), val = tensor<int32, [4]>([1, 480, 1, 256])];
            tensor<bool, [4]> var_1380_end_mask_0 = const()[name = string("op_1380_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 256]> var_1380_cast_fp16 = slice_by_index(begin = var_1380_begin_0, end = var_1380_end_0, end_mask = var_1380_end_mask_0, x = q_9_cast_fp16)[name = string("op_1380_cast_fp16")];
            tensor<int32, [4]> var_1384_begin_0 = const()[name = string("op_1384_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_1384_end_0 = const()[name = string("op_1384_end_0"), val = tensor<int32, [4]>([1, 640, 1, 256])];
            tensor<bool, [4]> var_1384_end_mask_0 = const()[name = string("op_1384_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 256]> var_1384_cast_fp16 = slice_by_index(begin = var_1384_begin_0, end = var_1384_end_0, end_mask = var_1384_end_mask_0, x = q_9_cast_fp16)[name = string("op_1384_cast_fp16")];
            tensor<int32, [4]> var_1388_begin_0 = const()[name = string("op_1388_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_1388_end_0 = const()[name = string("op_1388_end_0"), val = tensor<int32, [4]>([1, 800, 1, 256])];
            tensor<bool, [4]> var_1388_end_mask_0 = const()[name = string("op_1388_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 256]> var_1388_cast_fp16 = slice_by_index(begin = var_1388_begin_0, end = var_1388_end_0, end_mask = var_1388_end_mask_0, x = q_9_cast_fp16)[name = string("op_1388_cast_fp16")];
            tensor<int32, [4]> var_1392_begin_0 = const()[name = string("op_1392_begin_0"), val = tensor<int32, [4]>([0, 800, 0, 0])];
            tensor<int32, [4]> var_1392_end_0 = const()[name = string("op_1392_end_0"), val = tensor<int32, [4]>([1, 960, 1, 256])];
            tensor<bool, [4]> var_1392_end_mask_0 = const()[name = string("op_1392_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 256]> var_1392_cast_fp16 = slice_by_index(begin = var_1392_begin_0, end = var_1392_end_0, end_mask = var_1392_end_mask_0, x = q_9_cast_fp16)[name = string("op_1392_cast_fp16")];
            tensor<int32, [4]> var_1396_begin_0 = const()[name = string("op_1396_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_1396_end_0 = const()[name = string("op_1396_end_0"), val = tensor<int32, [4]>([1, 1120, 1, 256])];
            tensor<bool, [4]> var_1396_end_mask_0 = const()[name = string("op_1396_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 256]> var_1396_cast_fp16 = slice_by_index(begin = var_1396_begin_0, end = var_1396_end_0, end_mask = var_1396_end_mask_0, x = q_9_cast_fp16)[name = string("op_1396_cast_fp16")];
            tensor<int32, [4]> var_1400_begin_0 = const()[name = string("op_1400_begin_0"), val = tensor<int32, [4]>([0, 1120, 0, 0])];
            tensor<int32, [4]> var_1400_end_0 = const()[name = string("op_1400_end_0"), val = tensor<int32, [4]>([1, 1, 1, 256])];
            tensor<bool, [4]> var_1400_end_mask_0 = const()[name = string("op_1400_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 160, 1, 256]> var_1400_cast_fp16 = slice_by_index(begin = var_1400_begin_0, end = var_1400_end_0, end_mask = var_1400_end_mask_0, x = q_9_cast_fp16)[name = string("op_1400_cast_fp16")];
            tensor<int32, [4]> k_19_perm_0 = const()[name = string("k_19_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_1407_begin_0 = const()[name = string("op_1407_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1407_end_0 = const()[name = string("op_1407_end_0"), val = tensor<int32, [4]>([1, 256, 1, 160])];
            tensor<bool, [4]> var_1407_end_mask_0 = const()[name = string("op_1407_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 256, 1, 1280]> k_19_cast_fp16 = transpose(perm = k_19_perm_0, x = k_17_cast_fp16)[name = string("transpose_3")];
            tensor<fp16, [1, 256, 1, 160]> var_1407_cast_fp16 = slice_by_index(begin = var_1407_begin_0, end = var_1407_end_0, end_mask = var_1407_end_mask_0, x = k_19_cast_fp16)[name = string("op_1407_cast_fp16")];
            tensor<int32, [4]> var_1411_begin_0 = const()[name = string("op_1411_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 160])];
            tensor<int32, [4]> var_1411_end_0 = const()[name = string("op_1411_end_0"), val = tensor<int32, [4]>([1, 256, 1, 320])];
            tensor<bool, [4]> var_1411_end_mask_0 = const()[name = string("op_1411_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 256, 1, 160]> var_1411_cast_fp16 = slice_by_index(begin = var_1411_begin_0, end = var_1411_end_0, end_mask = var_1411_end_mask_0, x = k_19_cast_fp16)[name = string("op_1411_cast_fp16")];
            tensor<int32, [4]> var_1415_begin_0 = const()[name = string("op_1415_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_1415_end_0 = const()[name = string("op_1415_end_0"), val = tensor<int32, [4]>([1, 256, 1, 480])];
            tensor<bool, [4]> var_1415_end_mask_0 = const()[name = string("op_1415_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 256, 1, 160]> var_1415_cast_fp16 = slice_by_index(begin = var_1415_begin_0, end = var_1415_end_0, end_mask = var_1415_end_mask_0, x = k_19_cast_fp16)[name = string("op_1415_cast_fp16")];
            tensor<int32, [4]> var_1419_begin_0 = const()[name = string("op_1419_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 480])];
            tensor<int32, [4]> var_1419_end_0 = const()[name = string("op_1419_end_0"), val = tensor<int32, [4]>([1, 256, 1, 640])];
            tensor<bool, [4]> var_1419_end_mask_0 = const()[name = string("op_1419_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 256, 1, 160]> var_1419_cast_fp16 = slice_by_index(begin = var_1419_begin_0, end = var_1419_end_0, end_mask = var_1419_end_mask_0, x = k_19_cast_fp16)[name = string("op_1419_cast_fp16")];
            tensor<int32, [4]> var_1423_begin_0 = const()[name = string("op_1423_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_1423_end_0 = const()[name = string("op_1423_end_0"), val = tensor<int32, [4]>([1, 256, 1, 800])];
            tensor<bool, [4]> var_1423_end_mask_0 = const()[name = string("op_1423_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 256, 1, 160]> var_1423_cast_fp16 = slice_by_index(begin = var_1423_begin_0, end = var_1423_end_0, end_mask = var_1423_end_mask_0, x = k_19_cast_fp16)[name = string("op_1423_cast_fp16")];
            tensor<int32, [4]> var_1427_begin_0 = const()[name = string("op_1427_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 800])];
            tensor<int32, [4]> var_1427_end_0 = const()[name = string("op_1427_end_0"), val = tensor<int32, [4]>([1, 256, 1, 960])];
            tensor<bool, [4]> var_1427_end_mask_0 = const()[name = string("op_1427_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 256, 1, 160]> var_1427_cast_fp16 = slice_by_index(begin = var_1427_begin_0, end = var_1427_end_0, end_mask = var_1427_end_mask_0, x = k_19_cast_fp16)[name = string("op_1427_cast_fp16")];
            tensor<int32, [4]> var_1431_begin_0 = const()[name = string("op_1431_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_1431_end_0 = const()[name = string("op_1431_end_0"), val = tensor<int32, [4]>([1, 256, 1, 1120])];
            tensor<bool, [4]> var_1431_end_mask_0 = const()[name = string("op_1431_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 256, 1, 160]> var_1431_cast_fp16 = slice_by_index(begin = var_1431_begin_0, end = var_1431_end_0, end_mask = var_1431_end_mask_0, x = k_19_cast_fp16)[name = string("op_1431_cast_fp16")];
            tensor<int32, [4]> var_1435_begin_0 = const()[name = string("op_1435_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1120])];
            tensor<int32, [4]> var_1435_end_0 = const()[name = string("op_1435_end_0"), val = tensor<int32, [4]>([1, 256, 1, 1])];
            tensor<bool, [4]> var_1435_end_mask_0 = const()[name = string("op_1435_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 256, 1, 160]> var_1435_cast_fp16 = slice_by_index(begin = var_1435_begin_0, end = var_1435_end_0, end_mask = var_1435_end_mask_0, x = k_19_cast_fp16)[name = string("op_1435_cast_fp16")];
            tensor<int32, [4]> var_1437_begin_0 = const()[name = string("op_1437_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1437_end_0 = const()[name = string("op_1437_end_0"), val = tensor<int32, [4]>([1, 160, 1, 256])];
            tensor<bool, [4]> var_1437_end_mask_0 = const()[name = string("op_1437_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 256]> var_1437_cast_fp16 = slice_by_index(begin = var_1437_begin_0, end = var_1437_end_0, end_mask = var_1437_end_mask_0, x = v_9_cast_fp16)[name = string("op_1437_cast_fp16")];
            tensor<int32, [4]> var_1441_begin_0 = const()[name = string("op_1441_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_1441_end_0 = const()[name = string("op_1441_end_0"), val = tensor<int32, [4]>([1, 320, 1, 256])];
            tensor<bool, [4]> var_1441_end_mask_0 = const()[name = string("op_1441_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 256]> var_1441_cast_fp16 = slice_by_index(begin = var_1441_begin_0, end = var_1441_end_0, end_mask = var_1441_end_mask_0, x = v_9_cast_fp16)[name = string("op_1441_cast_fp16")];
            tensor<int32, [4]> var_1445_begin_0 = const()[name = string("op_1445_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_1445_end_0 = const()[name = string("op_1445_end_0"), val = tensor<int32, [4]>([1, 480, 1, 256])];
            tensor<bool, [4]> var_1445_end_mask_0 = const()[name = string("op_1445_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 256]> var_1445_cast_fp16 = slice_by_index(begin = var_1445_begin_0, end = var_1445_end_0, end_mask = var_1445_end_mask_0, x = v_9_cast_fp16)[name = string("op_1445_cast_fp16")];
            tensor<int32, [4]> var_1449_begin_0 = const()[name = string("op_1449_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_1449_end_0 = const()[name = string("op_1449_end_0"), val = tensor<int32, [4]>([1, 640, 1, 256])];
            tensor<bool, [4]> var_1449_end_mask_0 = const()[name = string("op_1449_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 256]> var_1449_cast_fp16 = slice_by_index(begin = var_1449_begin_0, end = var_1449_end_0, end_mask = var_1449_end_mask_0, x = v_9_cast_fp16)[name = string("op_1449_cast_fp16")];
            tensor<int32, [4]> var_1453_begin_0 = const()[name = string("op_1453_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_1453_end_0 = const()[name = string("op_1453_end_0"), val = tensor<int32, [4]>([1, 800, 1, 256])];
            tensor<bool, [4]> var_1453_end_mask_0 = const()[name = string("op_1453_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 256]> var_1453_cast_fp16 = slice_by_index(begin = var_1453_begin_0, end = var_1453_end_0, end_mask = var_1453_end_mask_0, x = v_9_cast_fp16)[name = string("op_1453_cast_fp16")];
            tensor<int32, [4]> var_1457_begin_0 = const()[name = string("op_1457_begin_0"), val = tensor<int32, [4]>([0, 800, 0, 0])];
            tensor<int32, [4]> var_1457_end_0 = const()[name = string("op_1457_end_0"), val = tensor<int32, [4]>([1, 960, 1, 256])];
            tensor<bool, [4]> var_1457_end_mask_0 = const()[name = string("op_1457_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 256]> var_1457_cast_fp16 = slice_by_index(begin = var_1457_begin_0, end = var_1457_end_0, end_mask = var_1457_end_mask_0, x = v_9_cast_fp16)[name = string("op_1457_cast_fp16")];
            tensor<int32, [4]> var_1461_begin_0 = const()[name = string("op_1461_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_1461_end_0 = const()[name = string("op_1461_end_0"), val = tensor<int32, [4]>([1, 1120, 1, 256])];
            tensor<bool, [4]> var_1461_end_mask_0 = const()[name = string("op_1461_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 256]> var_1461_cast_fp16 = slice_by_index(begin = var_1461_begin_0, end = var_1461_end_0, end_mask = var_1461_end_mask_0, x = v_9_cast_fp16)[name = string("op_1461_cast_fp16")];
            tensor<int32, [4]> var_1465_begin_0 = const()[name = string("op_1465_begin_0"), val = tensor<int32, [4]>([0, 1120, 0, 0])];
            tensor<int32, [4]> var_1465_end_0 = const()[name = string("op_1465_end_0"), val = tensor<int32, [4]>([1, 1, 1, 256])];
            tensor<bool, [4]> var_1465_end_mask_0 = const()[name = string("op_1465_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 160, 1, 256]> var_1465_cast_fp16 = slice_by_index(begin = var_1465_begin_0, end = var_1465_end_0, end_mask = var_1465_end_mask_0, x = v_9_cast_fp16)[name = string("op_1465_cast_fp16")];
            string var_1469_equation_0 = const()[name = string("op_1469_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 256, 1, 256]> var_1469_cast_fp16 = einsum(equation = var_1469_equation_0, values = (var_1407_cast_fp16, var_1372_cast_fp16))[name = string("op_1469_cast_fp16")];
            fp16 var_1470_to_fp16 = const()[name = string("op_1470_to_fp16"), val = fp16(0x1.43cp-4)];
            tensor<fp16, [1, 256, 1, 256]> aw_65_cast_fp16 = mul(x = var_1469_cast_fp16, y = var_1470_to_fp16)[name = string("aw_65_cast_fp16")];
            string var_1473_equation_0 = const()[name = string("op_1473_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 256, 1, 256]> var_1473_cast_fp16 = einsum(equation = var_1473_equation_0, values = (var_1411_cast_fp16, var_1376_cast_fp16))[name = string("op_1473_cast_fp16")];
            fp16 var_1474_to_fp16 = const()[name = string("op_1474_to_fp16"), val = fp16(0x1.43cp-4)];
            tensor<fp16, [1, 256, 1, 256]> aw_67_cast_fp16 = mul(x = var_1473_cast_fp16, y = var_1474_to_fp16)[name = string("aw_67_cast_fp16")];
            string var_1477_equation_0 = const()[name = string("op_1477_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 256, 1, 256]> var_1477_cast_fp16 = einsum(equation = var_1477_equation_0, values = (var_1415_cast_fp16, var_1380_cast_fp16))[name = string("op_1477_cast_fp16")];
            fp16 var_1478_to_fp16 = const()[name = string("op_1478_to_fp16"), val = fp16(0x1.43cp-4)];
            tensor<fp16, [1, 256, 1, 256]> aw_69_cast_fp16 = mul(x = var_1477_cast_fp16, y = var_1478_to_fp16)[name = string("aw_69_cast_fp16")];
            string var_1481_equation_0 = const()[name = string("op_1481_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 256, 1, 256]> var_1481_cast_fp16 = einsum(equation = var_1481_equation_0, values = (var_1419_cast_fp16, var_1384_cast_fp16))[name = string("op_1481_cast_fp16")];
            fp16 var_1482_to_fp16 = const()[name = string("op_1482_to_fp16"), val = fp16(0x1.43cp-4)];
            tensor<fp16, [1, 256, 1, 256]> aw_71_cast_fp16 = mul(x = var_1481_cast_fp16, y = var_1482_to_fp16)[name = string("aw_71_cast_fp16")];
            string var_1485_equation_0 = const()[name = string("op_1485_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 256, 1, 256]> var_1485_cast_fp16 = einsum(equation = var_1485_equation_0, values = (var_1423_cast_fp16, var_1388_cast_fp16))[name = string("op_1485_cast_fp16")];
            fp16 var_1486_to_fp16 = const()[name = string("op_1486_to_fp16"), val = fp16(0x1.43cp-4)];
            tensor<fp16, [1, 256, 1, 256]> aw_73_cast_fp16 = mul(x = var_1485_cast_fp16, y = var_1486_to_fp16)[name = string("aw_73_cast_fp16")];
            string var_1489_equation_0 = const()[name = string("op_1489_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 256, 1, 256]> var_1489_cast_fp16 = einsum(equation = var_1489_equation_0, values = (var_1427_cast_fp16, var_1392_cast_fp16))[name = string("op_1489_cast_fp16")];
            fp16 var_1490_to_fp16 = const()[name = string("op_1490_to_fp16"), val = fp16(0x1.43cp-4)];
            tensor<fp16, [1, 256, 1, 256]> aw_75_cast_fp16 = mul(x = var_1489_cast_fp16, y = var_1490_to_fp16)[name = string("aw_75_cast_fp16")];
            string var_1493_equation_0 = const()[name = string("op_1493_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 256, 1, 256]> var_1493_cast_fp16 = einsum(equation = var_1493_equation_0, values = (var_1431_cast_fp16, var_1396_cast_fp16))[name = string("op_1493_cast_fp16")];
            fp16 var_1494_to_fp16 = const()[name = string("op_1494_to_fp16"), val = fp16(0x1.43cp-4)];
            tensor<fp16, [1, 256, 1, 256]> aw_77_cast_fp16 = mul(x = var_1493_cast_fp16, y = var_1494_to_fp16)[name = string("aw_77_cast_fp16")];
            string var_1497_equation_0 = const()[name = string("op_1497_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 256, 1, 256]> var_1497_cast_fp16 = einsum(equation = var_1497_equation_0, values = (var_1435_cast_fp16, var_1400_cast_fp16))[name = string("op_1497_cast_fp16")];
            fp16 var_1498_to_fp16 = const()[name = string("op_1498_to_fp16"), val = fp16(0x1.43cp-4)];
            tensor<fp16, [1, 256, 1, 256]> aw_79_cast_fp16 = mul(x = var_1497_cast_fp16, y = var_1498_to_fp16)[name = string("aw_79_cast_fp16")];
            tensor<fp16, [1, 256, 1, 256]> var_1500_cast_fp16 = softmax(axis = var_1250, x = aw_65_cast_fp16)[name = string("op_1500_cast_fp16")];
            tensor<fp16, [1, 256, 1, 256]> var_1501_cast_fp16 = softmax(axis = var_1250, x = aw_67_cast_fp16)[name = string("op_1501_cast_fp16")];
            tensor<fp16, [1, 256, 1, 256]> var_1502_cast_fp16 = softmax(axis = var_1250, x = aw_69_cast_fp16)[name = string("op_1502_cast_fp16")];
            tensor<fp16, [1, 256, 1, 256]> var_1503_cast_fp16 = softmax(axis = var_1250, x = aw_71_cast_fp16)[name = string("op_1503_cast_fp16")];
            tensor<fp16, [1, 256, 1, 256]> var_1504_cast_fp16 = softmax(axis = var_1250, x = aw_73_cast_fp16)[name = string("op_1504_cast_fp16")];
            tensor<fp16, [1, 256, 1, 256]> var_1505_cast_fp16 = softmax(axis = var_1250, x = aw_75_cast_fp16)[name = string("op_1505_cast_fp16")];
            tensor<fp16, [1, 256, 1, 256]> var_1506_cast_fp16 = softmax(axis = var_1250, x = aw_77_cast_fp16)[name = string("op_1506_cast_fp16")];
            tensor<fp16, [1, 256, 1, 256]> var_1507_cast_fp16 = softmax(axis = var_1250, x = aw_79_cast_fp16)[name = string("op_1507_cast_fp16")];
            string var_1509_equation_0 = const()[name = string("op_1509_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 160, 1, 256]> var_1509_cast_fp16 = einsum(equation = var_1509_equation_0, values = (var_1437_cast_fp16, var_1500_cast_fp16))[name = string("op_1509_cast_fp16")];
            string var_1511_equation_0 = const()[name = string("op_1511_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 160, 1, 256]> var_1511_cast_fp16 = einsum(equation = var_1511_equation_0, values = (var_1441_cast_fp16, var_1501_cast_fp16))[name = string("op_1511_cast_fp16")];
            string var_1513_equation_0 = const()[name = string("op_1513_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 160, 1, 256]> var_1513_cast_fp16 = einsum(equation = var_1513_equation_0, values = (var_1445_cast_fp16, var_1502_cast_fp16))[name = string("op_1513_cast_fp16")];
            string var_1515_equation_0 = const()[name = string("op_1515_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 160, 1, 256]> var_1515_cast_fp16 = einsum(equation = var_1515_equation_0, values = (var_1449_cast_fp16, var_1503_cast_fp16))[name = string("op_1515_cast_fp16")];
            string var_1517_equation_0 = const()[name = string("op_1517_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 160, 1, 256]> var_1517_cast_fp16 = einsum(equation = var_1517_equation_0, values = (var_1453_cast_fp16, var_1504_cast_fp16))[name = string("op_1517_cast_fp16")];
            string var_1519_equation_0 = const()[name = string("op_1519_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 160, 1, 256]> var_1519_cast_fp16 = einsum(equation = var_1519_equation_0, values = (var_1457_cast_fp16, var_1505_cast_fp16))[name = string("op_1519_cast_fp16")];
            string var_1521_equation_0 = const()[name = string("op_1521_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 160, 1, 256]> var_1521_cast_fp16 = einsum(equation = var_1521_equation_0, values = (var_1461_cast_fp16, var_1506_cast_fp16))[name = string("op_1521_cast_fp16")];
            string var_1523_equation_0 = const()[name = string("op_1523_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 160, 1, 256]> var_1523_cast_fp16 = einsum(equation = var_1523_equation_0, values = (var_1465_cast_fp16, var_1507_cast_fp16))[name = string("op_1523_cast_fp16")];
            bool input_81_interleave_0 = const()[name = string("input_81_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 1280, 1, 256]> input_81_cast_fp16 = concat(axis = var_1250, interleave = input_81_interleave_0, values = (var_1509_cast_fp16, var_1511_cast_fp16, var_1513_cast_fp16, var_1515_cast_fp16, var_1517_cast_fp16, var_1519_cast_fp16, var_1521_cast_fp16, var_1523_cast_fp16))[name = string("input_81_cast_fp16")];
            string var_1533_pad_type_0 = const()[name = string("op_1533_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_1533_strides_0 = const()[name = string("op_1533_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_1533_pad_0 = const()[name = string("op_1533_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_1533_dilations_0 = const()[name = string("op_1533_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_1533_groups_0 = const()[name = string("op_1533_groups_0"), val = int32(1)];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = string("down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(116773632)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = string("down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120050496)))];
            tensor<fp16, [1, 1280, 1, 256]> var_1533_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_1533_dilations_0, groups = var_1533_groups_0, pad = var_1533_pad_0, pad_type = var_1533_pad_type_0, strides = var_1533_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_81_cast_fp16)[name = string("op_1533_cast_fp16")];
            tensor<fp16, [1, 1280, 1, 256]> inputs_15_cast_fp16 = add(x = var_1533_cast_fp16, y = inputs_13_cast_fp16)[name = string("inputs_15_cast_fp16")];
            tensor<int32, [1]> hidden_states_47_axes_0 = const()[name = string("hidden_states_47_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_47_gamma_0_to_fp16 = const()[name = string("hidden_states_47_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120053120)))];
            tensor<fp16, [1280]> hidden_states_47_beta_0_to_fp16 = const()[name = string("hidden_states_47_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120055744)))];
            fp16 var_1543_to_fp16 = const()[name = string("op_1543_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1280, 1, 256]> hidden_states_47_cast_fp16 = layer_norm(axes = hidden_states_47_axes_0, beta = hidden_states_47_beta_0_to_fp16, epsilon = var_1543_to_fp16, gamma = hidden_states_47_gamma_0_to_fp16, x = inputs_15_cast_fp16)[name = string("hidden_states_47_cast_fp16")];
            string q_11_pad_type_0 = const()[name = string("q_11_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> q_11_strides_0 = const()[name = string("q_11_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_11_pad_0 = const()[name = string("q_11_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_11_dilations_0 = const()[name = string("q_11_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 q_11_groups_0 = const()[name = string("q_11_groups_0"), val = int32(1)];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = string("down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120058368)))];
            tensor<fp16, [1, 1280, 1, 256]> q_11_cast_fp16 = conv(dilations = q_11_dilations_0, groups = q_11_groups_0, pad = q_11_pad_0, pad_type = q_11_pad_type_0, strides = q_11_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_47_cast_fp16)[name = string("q_11_cast_fp16")];
            string k_21_pad_type_0 = const()[name = string("k_21_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> k_21_strides_0 = const()[name = string("k_21_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_21_pad_0 = const()[name = string("k_21_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_21_dilations_0 = const()[name = string("k_21_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 k_21_groups_0 = const()[name = string("k_21_groups_0"), val = int32(1)];
            tensor<fp16, [1280, 768, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = string("down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 768, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123335232)))];
            tensor<fp16, [1, 1280, 1, 77]> k_21_cast_fp16 = conv(dilations = k_21_dilations_0, groups = k_21_groups_0, pad = k_21_pad_0, pad_type = k_21_pad_type_0, strides = k_21_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = string("k_21_cast_fp16")];
            string v_11_pad_type_0 = const()[name = string("v_11_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> v_11_strides_0 = const()[name = string("v_11_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_11_pad_0 = const()[name = string("v_11_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_11_dilations_0 = const()[name = string("v_11_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 v_11_groups_0 = const()[name = string("v_11_groups_0"), val = int32(1)];
            tensor<fp16, [1280, 768, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = string("down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 768, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125301376)))];
            tensor<fp16, [1, 1280, 1, 77]> v_11_cast_fp16 = conv(dilations = v_11_dilations_0, groups = v_11_groups_0, pad = v_11_pad_0, pad_type = v_11_pad_type_0, strides = v_11_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = string("v_11_cast_fp16")];
            tensor<int32, [4]> var_1576_begin_0 = const()[name = string("op_1576_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1576_end_0 = const()[name = string("op_1576_end_0"), val = tensor<int32, [4]>([1, 160, 1, 256])];
            tensor<bool, [4]> var_1576_end_mask_0 = const()[name = string("op_1576_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 256]> var_1576_cast_fp16 = slice_by_index(begin = var_1576_begin_0, end = var_1576_end_0, end_mask = var_1576_end_mask_0, x = q_11_cast_fp16)[name = string("op_1576_cast_fp16")];
            tensor<int32, [4]> var_1580_begin_0 = const()[name = string("op_1580_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_1580_end_0 = const()[name = string("op_1580_end_0"), val = tensor<int32, [4]>([1, 320, 1, 256])];
            tensor<bool, [4]> var_1580_end_mask_0 = const()[name = string("op_1580_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 256]> var_1580_cast_fp16 = slice_by_index(begin = var_1580_begin_0, end = var_1580_end_0, end_mask = var_1580_end_mask_0, x = q_11_cast_fp16)[name = string("op_1580_cast_fp16")];
            tensor<int32, [4]> var_1584_begin_0 = const()[name = string("op_1584_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_1584_end_0 = const()[name = string("op_1584_end_0"), val = tensor<int32, [4]>([1, 480, 1, 256])];
            tensor<bool, [4]> var_1584_end_mask_0 = const()[name = string("op_1584_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 256]> var_1584_cast_fp16 = slice_by_index(begin = var_1584_begin_0, end = var_1584_end_0, end_mask = var_1584_end_mask_0, x = q_11_cast_fp16)[name = string("op_1584_cast_fp16")];
            tensor<int32, [4]> var_1588_begin_0 = const()[name = string("op_1588_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_1588_end_0 = const()[name = string("op_1588_end_0"), val = tensor<int32, [4]>([1, 640, 1, 256])];
            tensor<bool, [4]> var_1588_end_mask_0 = const()[name = string("op_1588_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 256]> var_1588_cast_fp16 = slice_by_index(begin = var_1588_begin_0, end = var_1588_end_0, end_mask = var_1588_end_mask_0, x = q_11_cast_fp16)[name = string("op_1588_cast_fp16")];
            tensor<int32, [4]> var_1592_begin_0 = const()[name = string("op_1592_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_1592_end_0 = const()[name = string("op_1592_end_0"), val = tensor<int32, [4]>([1, 800, 1, 256])];
            tensor<bool, [4]> var_1592_end_mask_0 = const()[name = string("op_1592_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 256]> var_1592_cast_fp16 = slice_by_index(begin = var_1592_begin_0, end = var_1592_end_0, end_mask = var_1592_end_mask_0, x = q_11_cast_fp16)[name = string("op_1592_cast_fp16")];
            tensor<int32, [4]> var_1596_begin_0 = const()[name = string("op_1596_begin_0"), val = tensor<int32, [4]>([0, 800, 0, 0])];
            tensor<int32, [4]> var_1596_end_0 = const()[name = string("op_1596_end_0"), val = tensor<int32, [4]>([1, 960, 1, 256])];
            tensor<bool, [4]> var_1596_end_mask_0 = const()[name = string("op_1596_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 256]> var_1596_cast_fp16 = slice_by_index(begin = var_1596_begin_0, end = var_1596_end_0, end_mask = var_1596_end_mask_0, x = q_11_cast_fp16)[name = string("op_1596_cast_fp16")];
            tensor<int32, [4]> var_1600_begin_0 = const()[name = string("op_1600_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_1600_end_0 = const()[name = string("op_1600_end_0"), val = tensor<int32, [4]>([1, 1120, 1, 256])];
            tensor<bool, [4]> var_1600_end_mask_0 = const()[name = string("op_1600_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 256]> var_1600_cast_fp16 = slice_by_index(begin = var_1600_begin_0, end = var_1600_end_0, end_mask = var_1600_end_mask_0, x = q_11_cast_fp16)[name = string("op_1600_cast_fp16")];
            tensor<int32, [4]> var_1604_begin_0 = const()[name = string("op_1604_begin_0"), val = tensor<int32, [4]>([0, 1120, 0, 0])];
            tensor<int32, [4]> var_1604_end_0 = const()[name = string("op_1604_end_0"), val = tensor<int32, [4]>([1, 1, 1, 256])];
            tensor<bool, [4]> var_1604_end_mask_0 = const()[name = string("op_1604_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 160, 1, 256]> var_1604_cast_fp16 = slice_by_index(begin = var_1604_begin_0, end = var_1604_end_0, end_mask = var_1604_end_mask_0, x = q_11_cast_fp16)[name = string("op_1604_cast_fp16")];
            tensor<int32, [4]> k_23_perm_0 = const()[name = string("k_23_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_1611_begin_0 = const()[name = string("op_1611_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1611_end_0 = const()[name = string("op_1611_end_0"), val = tensor<int32, [4]>([1, 77, 1, 160])];
            tensor<bool, [4]> var_1611_end_mask_0 = const()[name = string("op_1611_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 77, 1, 1280]> k_23_cast_fp16 = transpose(perm = k_23_perm_0, x = k_21_cast_fp16)[name = string("transpose_2")];
            tensor<fp16, [1, 77, 1, 160]> var_1611_cast_fp16 = slice_by_index(begin = var_1611_begin_0, end = var_1611_end_0, end_mask = var_1611_end_mask_0, x = k_23_cast_fp16)[name = string("op_1611_cast_fp16")];
            tensor<int32, [4]> var_1615_begin_0 = const()[name = string("op_1615_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 160])];
            tensor<int32, [4]> var_1615_end_0 = const()[name = string("op_1615_end_0"), val = tensor<int32, [4]>([1, 77, 1, 320])];
            tensor<bool, [4]> var_1615_end_mask_0 = const()[name = string("op_1615_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 77, 1, 160]> var_1615_cast_fp16 = slice_by_index(begin = var_1615_begin_0, end = var_1615_end_0, end_mask = var_1615_end_mask_0, x = k_23_cast_fp16)[name = string("op_1615_cast_fp16")];
            tensor<int32, [4]> var_1619_begin_0 = const()[name = string("op_1619_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_1619_end_0 = const()[name = string("op_1619_end_0"), val = tensor<int32, [4]>([1, 77, 1, 480])];
            tensor<bool, [4]> var_1619_end_mask_0 = const()[name = string("op_1619_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 77, 1, 160]> var_1619_cast_fp16 = slice_by_index(begin = var_1619_begin_0, end = var_1619_end_0, end_mask = var_1619_end_mask_0, x = k_23_cast_fp16)[name = string("op_1619_cast_fp16")];
            tensor<int32, [4]> var_1623_begin_0 = const()[name = string("op_1623_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 480])];
            tensor<int32, [4]> var_1623_end_0 = const()[name = string("op_1623_end_0"), val = tensor<int32, [4]>([1, 77, 1, 640])];
            tensor<bool, [4]> var_1623_end_mask_0 = const()[name = string("op_1623_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 77, 1, 160]> var_1623_cast_fp16 = slice_by_index(begin = var_1623_begin_0, end = var_1623_end_0, end_mask = var_1623_end_mask_0, x = k_23_cast_fp16)[name = string("op_1623_cast_fp16")];
            tensor<int32, [4]> var_1627_begin_0 = const()[name = string("op_1627_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_1627_end_0 = const()[name = string("op_1627_end_0"), val = tensor<int32, [4]>([1, 77, 1, 800])];
            tensor<bool, [4]> var_1627_end_mask_0 = const()[name = string("op_1627_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 77, 1, 160]> var_1627_cast_fp16 = slice_by_index(begin = var_1627_begin_0, end = var_1627_end_0, end_mask = var_1627_end_mask_0, x = k_23_cast_fp16)[name = string("op_1627_cast_fp16")];
            tensor<int32, [4]> var_1631_begin_0 = const()[name = string("op_1631_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 800])];
            tensor<int32, [4]> var_1631_end_0 = const()[name = string("op_1631_end_0"), val = tensor<int32, [4]>([1, 77, 1, 960])];
            tensor<bool, [4]> var_1631_end_mask_0 = const()[name = string("op_1631_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 77, 1, 160]> var_1631_cast_fp16 = slice_by_index(begin = var_1631_begin_0, end = var_1631_end_0, end_mask = var_1631_end_mask_0, x = k_23_cast_fp16)[name = string("op_1631_cast_fp16")];
            tensor<int32, [4]> var_1635_begin_0 = const()[name = string("op_1635_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_1635_end_0 = const()[name = string("op_1635_end_0"), val = tensor<int32, [4]>([1, 77, 1, 1120])];
            tensor<bool, [4]> var_1635_end_mask_0 = const()[name = string("op_1635_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 77, 1, 160]> var_1635_cast_fp16 = slice_by_index(begin = var_1635_begin_0, end = var_1635_end_0, end_mask = var_1635_end_mask_0, x = k_23_cast_fp16)[name = string("op_1635_cast_fp16")];
            tensor<int32, [4]> var_1639_begin_0 = const()[name = string("op_1639_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1120])];
            tensor<int32, [4]> var_1639_end_0 = const()[name = string("op_1639_end_0"), val = tensor<int32, [4]>([1, 77, 1, 1])];
            tensor<bool, [4]> var_1639_end_mask_0 = const()[name = string("op_1639_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 77, 1, 160]> var_1639_cast_fp16 = slice_by_index(begin = var_1639_begin_0, end = var_1639_end_0, end_mask = var_1639_end_mask_0, x = k_23_cast_fp16)[name = string("op_1639_cast_fp16")];
            tensor<int32, [4]> var_1641_begin_0 = const()[name = string("op_1641_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1641_end_0 = const()[name = string("op_1641_end_0"), val = tensor<int32, [4]>([1, 160, 1, 77])];
            tensor<bool, [4]> var_1641_end_mask_0 = const()[name = string("op_1641_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 77]> var_1641_cast_fp16 = slice_by_index(begin = var_1641_begin_0, end = var_1641_end_0, end_mask = var_1641_end_mask_0, x = v_11_cast_fp16)[name = string("op_1641_cast_fp16")];
            tensor<int32, [4]> var_1645_begin_0 = const()[name = string("op_1645_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_1645_end_0 = const()[name = string("op_1645_end_0"), val = tensor<int32, [4]>([1, 320, 1, 77])];
            tensor<bool, [4]> var_1645_end_mask_0 = const()[name = string("op_1645_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 77]> var_1645_cast_fp16 = slice_by_index(begin = var_1645_begin_0, end = var_1645_end_0, end_mask = var_1645_end_mask_0, x = v_11_cast_fp16)[name = string("op_1645_cast_fp16")];
            tensor<int32, [4]> var_1649_begin_0 = const()[name = string("op_1649_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_1649_end_0 = const()[name = string("op_1649_end_0"), val = tensor<int32, [4]>([1, 480, 1, 77])];
            tensor<bool, [4]> var_1649_end_mask_0 = const()[name = string("op_1649_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 77]> var_1649_cast_fp16 = slice_by_index(begin = var_1649_begin_0, end = var_1649_end_0, end_mask = var_1649_end_mask_0, x = v_11_cast_fp16)[name = string("op_1649_cast_fp16")];
            tensor<int32, [4]> var_1653_begin_0 = const()[name = string("op_1653_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_1653_end_0 = const()[name = string("op_1653_end_0"), val = tensor<int32, [4]>([1, 640, 1, 77])];
            tensor<bool, [4]> var_1653_end_mask_0 = const()[name = string("op_1653_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 77]> var_1653_cast_fp16 = slice_by_index(begin = var_1653_begin_0, end = var_1653_end_0, end_mask = var_1653_end_mask_0, x = v_11_cast_fp16)[name = string("op_1653_cast_fp16")];
            tensor<int32, [4]> var_1657_begin_0 = const()[name = string("op_1657_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_1657_end_0 = const()[name = string("op_1657_end_0"), val = tensor<int32, [4]>([1, 800, 1, 77])];
            tensor<bool, [4]> var_1657_end_mask_0 = const()[name = string("op_1657_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 77]> var_1657_cast_fp16 = slice_by_index(begin = var_1657_begin_0, end = var_1657_end_0, end_mask = var_1657_end_mask_0, x = v_11_cast_fp16)[name = string("op_1657_cast_fp16")];
            tensor<int32, [4]> var_1661_begin_0 = const()[name = string("op_1661_begin_0"), val = tensor<int32, [4]>([0, 800, 0, 0])];
            tensor<int32, [4]> var_1661_end_0 = const()[name = string("op_1661_end_0"), val = tensor<int32, [4]>([1, 960, 1, 77])];
            tensor<bool, [4]> var_1661_end_mask_0 = const()[name = string("op_1661_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 77]> var_1661_cast_fp16 = slice_by_index(begin = var_1661_begin_0, end = var_1661_end_0, end_mask = var_1661_end_mask_0, x = v_11_cast_fp16)[name = string("op_1661_cast_fp16")];
            tensor<int32, [4]> var_1665_begin_0 = const()[name = string("op_1665_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_1665_end_0 = const()[name = string("op_1665_end_0"), val = tensor<int32, [4]>([1, 1120, 1, 77])];
            tensor<bool, [4]> var_1665_end_mask_0 = const()[name = string("op_1665_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 77]> var_1665_cast_fp16 = slice_by_index(begin = var_1665_begin_0, end = var_1665_end_0, end_mask = var_1665_end_mask_0, x = v_11_cast_fp16)[name = string("op_1665_cast_fp16")];
            tensor<int32, [4]> var_1669_begin_0 = const()[name = string("op_1669_begin_0"), val = tensor<int32, [4]>([0, 1120, 0, 0])];
            tensor<int32, [4]> var_1669_end_0 = const()[name = string("op_1669_end_0"), val = tensor<int32, [4]>([1, 1, 1, 77])];
            tensor<bool, [4]> var_1669_end_mask_0 = const()[name = string("op_1669_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 160, 1, 77]> var_1669_cast_fp16 = slice_by_index(begin = var_1669_begin_0, end = var_1669_end_0, end_mask = var_1669_end_mask_0, x = v_11_cast_fp16)[name = string("op_1669_cast_fp16")];
            string var_1673_equation_0 = const()[name = string("op_1673_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 77, 1, 256]> var_1673_cast_fp16 = einsum(equation = var_1673_equation_0, values = (var_1611_cast_fp16, var_1576_cast_fp16))[name = string("op_1673_cast_fp16")];
            fp16 var_1674_to_fp16 = const()[name = string("op_1674_to_fp16"), val = fp16(0x1.43cp-4)];
            tensor<fp16, [1, 77, 1, 256]> aw_81_cast_fp16 = mul(x = var_1673_cast_fp16, y = var_1674_to_fp16)[name = string("aw_81_cast_fp16")];
            string var_1677_equation_0 = const()[name = string("op_1677_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 77, 1, 256]> var_1677_cast_fp16 = einsum(equation = var_1677_equation_0, values = (var_1615_cast_fp16, var_1580_cast_fp16))[name = string("op_1677_cast_fp16")];
            fp16 var_1678_to_fp16 = const()[name = string("op_1678_to_fp16"), val = fp16(0x1.43cp-4)];
            tensor<fp16, [1, 77, 1, 256]> aw_83_cast_fp16 = mul(x = var_1677_cast_fp16, y = var_1678_to_fp16)[name = string("aw_83_cast_fp16")];
            string var_1681_equation_0 = const()[name = string("op_1681_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 77, 1, 256]> var_1681_cast_fp16 = einsum(equation = var_1681_equation_0, values = (var_1619_cast_fp16, var_1584_cast_fp16))[name = string("op_1681_cast_fp16")];
            fp16 var_1682_to_fp16 = const()[name = string("op_1682_to_fp16"), val = fp16(0x1.43cp-4)];
            tensor<fp16, [1, 77, 1, 256]> aw_85_cast_fp16 = mul(x = var_1681_cast_fp16, y = var_1682_to_fp16)[name = string("aw_85_cast_fp16")];
            string var_1685_equation_0 = const()[name = string("op_1685_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 77, 1, 256]> var_1685_cast_fp16 = einsum(equation = var_1685_equation_0, values = (var_1623_cast_fp16, var_1588_cast_fp16))[name = string("op_1685_cast_fp16")];
            fp16 var_1686_to_fp16 = const()[name = string("op_1686_to_fp16"), val = fp16(0x1.43cp-4)];
            tensor<fp16, [1, 77, 1, 256]> aw_87_cast_fp16 = mul(x = var_1685_cast_fp16, y = var_1686_to_fp16)[name = string("aw_87_cast_fp16")];
            string var_1689_equation_0 = const()[name = string("op_1689_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 77, 1, 256]> var_1689_cast_fp16 = einsum(equation = var_1689_equation_0, values = (var_1627_cast_fp16, var_1592_cast_fp16))[name = string("op_1689_cast_fp16")];
            fp16 var_1690_to_fp16 = const()[name = string("op_1690_to_fp16"), val = fp16(0x1.43cp-4)];
            tensor<fp16, [1, 77, 1, 256]> aw_89_cast_fp16 = mul(x = var_1689_cast_fp16, y = var_1690_to_fp16)[name = string("aw_89_cast_fp16")];
            string var_1693_equation_0 = const()[name = string("op_1693_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 77, 1, 256]> var_1693_cast_fp16 = einsum(equation = var_1693_equation_0, values = (var_1631_cast_fp16, var_1596_cast_fp16))[name = string("op_1693_cast_fp16")];
            fp16 var_1694_to_fp16 = const()[name = string("op_1694_to_fp16"), val = fp16(0x1.43cp-4)];
            tensor<fp16, [1, 77, 1, 256]> aw_91_cast_fp16 = mul(x = var_1693_cast_fp16, y = var_1694_to_fp16)[name = string("aw_91_cast_fp16")];
            string var_1697_equation_0 = const()[name = string("op_1697_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 77, 1, 256]> var_1697_cast_fp16 = einsum(equation = var_1697_equation_0, values = (var_1635_cast_fp16, var_1600_cast_fp16))[name = string("op_1697_cast_fp16")];
            fp16 var_1698_to_fp16 = const()[name = string("op_1698_to_fp16"), val = fp16(0x1.43cp-4)];
            tensor<fp16, [1, 77, 1, 256]> aw_93_cast_fp16 = mul(x = var_1697_cast_fp16, y = var_1698_to_fp16)[name = string("aw_93_cast_fp16")];
            string var_1701_equation_0 = const()[name = string("op_1701_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 77, 1, 256]> var_1701_cast_fp16 = einsum(equation = var_1701_equation_0, values = (var_1639_cast_fp16, var_1604_cast_fp16))[name = string("op_1701_cast_fp16")];
            fp16 var_1702_to_fp16 = const()[name = string("op_1702_to_fp16"), val = fp16(0x1.43cp-4)];
            tensor<fp16, [1, 77, 1, 256]> aw_95_cast_fp16 = mul(x = var_1701_cast_fp16, y = var_1702_to_fp16)[name = string("aw_95_cast_fp16")];
            tensor<fp16, [1, 77, 1, 256]> var_1704_cast_fp16 = softmax(axis = var_1250, x = aw_81_cast_fp16)[name = string("op_1704_cast_fp16")];
            tensor<fp16, [1, 77, 1, 256]> var_1705_cast_fp16 = softmax(axis = var_1250, x = aw_83_cast_fp16)[name = string("op_1705_cast_fp16")];
            tensor<fp16, [1, 77, 1, 256]> var_1706_cast_fp16 = softmax(axis = var_1250, x = aw_85_cast_fp16)[name = string("op_1706_cast_fp16")];
            tensor<fp16, [1, 77, 1, 256]> var_1707_cast_fp16 = softmax(axis = var_1250, x = aw_87_cast_fp16)[name = string("op_1707_cast_fp16")];
            tensor<fp16, [1, 77, 1, 256]> var_1708_cast_fp16 = softmax(axis = var_1250, x = aw_89_cast_fp16)[name = string("op_1708_cast_fp16")];
            tensor<fp16, [1, 77, 1, 256]> var_1709_cast_fp16 = softmax(axis = var_1250, x = aw_91_cast_fp16)[name = string("op_1709_cast_fp16")];
            tensor<fp16, [1, 77, 1, 256]> var_1710_cast_fp16 = softmax(axis = var_1250, x = aw_93_cast_fp16)[name = string("op_1710_cast_fp16")];
            tensor<fp16, [1, 77, 1, 256]> var_1711_cast_fp16 = softmax(axis = var_1250, x = aw_95_cast_fp16)[name = string("op_1711_cast_fp16")];
            string var_1713_equation_0 = const()[name = string("op_1713_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 160, 1, 256]> var_1713_cast_fp16 = einsum(equation = var_1713_equation_0, values = (var_1641_cast_fp16, var_1704_cast_fp16))[name = string("op_1713_cast_fp16")];
            string var_1715_equation_0 = const()[name = string("op_1715_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 160, 1, 256]> var_1715_cast_fp16 = einsum(equation = var_1715_equation_0, values = (var_1645_cast_fp16, var_1705_cast_fp16))[name = string("op_1715_cast_fp16")];
            string var_1717_equation_0 = const()[name = string("op_1717_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 160, 1, 256]> var_1717_cast_fp16 = einsum(equation = var_1717_equation_0, values = (var_1649_cast_fp16, var_1706_cast_fp16))[name = string("op_1717_cast_fp16")];
            string var_1719_equation_0 = const()[name = string("op_1719_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 160, 1, 256]> var_1719_cast_fp16 = einsum(equation = var_1719_equation_0, values = (var_1653_cast_fp16, var_1707_cast_fp16))[name = string("op_1719_cast_fp16")];
            string var_1721_equation_0 = const()[name = string("op_1721_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 160, 1, 256]> var_1721_cast_fp16 = einsum(equation = var_1721_equation_0, values = (var_1657_cast_fp16, var_1708_cast_fp16))[name = string("op_1721_cast_fp16")];
            string var_1723_equation_0 = const()[name = string("op_1723_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 160, 1, 256]> var_1723_cast_fp16 = einsum(equation = var_1723_equation_0, values = (var_1661_cast_fp16, var_1709_cast_fp16))[name = string("op_1723_cast_fp16")];
            string var_1725_equation_0 = const()[name = string("op_1725_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 160, 1, 256]> var_1725_cast_fp16 = einsum(equation = var_1725_equation_0, values = (var_1665_cast_fp16, var_1710_cast_fp16))[name = string("op_1725_cast_fp16")];
            string var_1727_equation_0 = const()[name = string("op_1727_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 160, 1, 256]> var_1727_cast_fp16 = einsum(equation = var_1727_equation_0, values = (var_1669_cast_fp16, var_1711_cast_fp16))[name = string("op_1727_cast_fp16")];
            bool input_83_interleave_0 = const()[name = string("input_83_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 1280, 1, 256]> input_83_cast_fp16 = concat(axis = var_1250, interleave = input_83_interleave_0, values = (var_1713_cast_fp16, var_1715_cast_fp16, var_1717_cast_fp16, var_1719_cast_fp16, var_1721_cast_fp16, var_1723_cast_fp16, var_1725_cast_fp16, var_1727_cast_fp16))[name = string("input_83_cast_fp16")];
            string var_1737_pad_type_0 = const()[name = string("op_1737_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_1737_strides_0 = const()[name = string("op_1737_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_1737_pad_0 = const()[name = string("op_1737_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_1737_dilations_0 = const()[name = string("op_1737_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_1737_groups_0 = const()[name = string("op_1737_groups_0"), val = int32(1)];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = string("down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(127267520)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = string("down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(130544384)))];
            tensor<fp16, [1, 1280, 1, 256]> var_1737_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_1737_dilations_0, groups = var_1737_groups_0, pad = var_1737_pad_0, pad_type = var_1737_pad_type_0, strides = var_1737_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_83_cast_fp16)[name = string("op_1737_cast_fp16")];
            tensor<fp16, [1, 1280, 1, 256]> inputs_17_cast_fp16 = add(x = var_1737_cast_fp16, y = inputs_15_cast_fp16)[name = string("inputs_17_cast_fp16")];
            tensor<int32, [1]> input_85_axes_0 = const()[name = string("input_85_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> input_85_gamma_0_to_fp16 = const()[name = string("input_85_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(130547008)))];
            tensor<fp16, [1280]> input_85_beta_0_to_fp16 = const()[name = string("input_85_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(130549632)))];
            fp16 var_1747_to_fp16 = const()[name = string("op_1747_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1280, 1, 256]> input_85_cast_fp16 = layer_norm(axes = input_85_axes_0, beta = input_85_beta_0_to_fp16, epsilon = var_1747_to_fp16, gamma = input_85_gamma_0_to_fp16, x = inputs_17_cast_fp16)[name = string("input_85_cast_fp16")];
            string var_1767_pad_type_0 = const()[name = string("op_1767_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_1767_strides_0 = const()[name = string("op_1767_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_1767_pad_0 = const()[name = string("op_1767_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_1767_dilations_0 = const()[name = string("op_1767_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_1767_groups_0 = const()[name = string("op_1767_groups_0"), val = int32(1)];
            tensor<fp16, [10240, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = string("down_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(130552256)))];
            tensor<fp16, [10240]> down_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = string("down_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(156766720)))];
            tensor<fp16, [1, 10240, 1, 256]> var_1767_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_1767_dilations_0, groups = var_1767_groups_0, pad = var_1767_pad_0, pad_type = var_1767_pad_type_0, strides = var_1767_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_85_cast_fp16)[name = string("op_1767_cast_fp16")];
            tensor<int32, [2]> var_1768_split_sizes_0 = const()[name = string("op_1768_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            int32 var_1768_axis_0 = const()[name = string("op_1768_axis_0"), val = int32(1)];
            tensor<fp16, [1, 5120, 1, 256]> var_1768_cast_fp16_0, tensor<fp16, [1, 5120, 1, 256]> var_1768_cast_fp16_1 = split(axis = var_1768_axis_0, split_sizes = var_1768_split_sizes_0, x = var_1767_cast_fp16)[name = string("op_1768_cast_fp16")];
            string var_1770_mode_0 = const()[name = string("op_1770_mode_0"), val = string("EXACT")];
            tensor<fp16, [1, 5120, 1, 256]> var_1770_cast_fp16 = gelu(mode = var_1770_mode_0, x = var_1768_cast_fp16_1)[name = string("op_1770_cast_fp16")];
            tensor<fp16, [1, 5120, 1, 256]> input_87_cast_fp16 = mul(x = var_1768_cast_fp16_0, y = var_1770_cast_fp16)[name = string("input_87_cast_fp16")];
            string var_1778_pad_type_0 = const()[name = string("op_1778_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_1778_strides_0 = const()[name = string("op_1778_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_1778_pad_0 = const()[name = string("op_1778_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_1778_dilations_0 = const()[name = string("op_1778_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_1778_groups_0 = const()[name = string("op_1778_groups_0"), val = int32(1)];
            tensor<fp16, [1280, 5120, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = string("down_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(156787264)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = string("down_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(169894528)))];
            tensor<fp16, [1, 1280, 1, 256]> var_1778_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_1778_dilations_0, groups = var_1778_groups_0, pad = var_1778_pad_0, pad_type = var_1778_pad_type_0, strides = var_1778_strides_0, weight = down_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_87_cast_fp16)[name = string("op_1778_cast_fp16")];
            tensor<fp16, [1, 1280, 1, 256]> hidden_states_51_cast_fp16 = add(x = var_1778_cast_fp16, y = inputs_17_cast_fp16)[name = string("hidden_states_51_cast_fp16")];
            tensor<int32, [4]> var_1780 = const()[name = string("op_1780"), val = tensor<int32, [4]>([1, 1280, 16, 16])];
            tensor<fp16, [1, 1280, 16, 16]> input_89_cast_fp16 = reshape(shape = var_1780, x = hidden_states_51_cast_fp16)[name = string("input_89_cast_fp16")];
            string hidden_states_53_pad_type_0 = const()[name = string("hidden_states_53_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> hidden_states_53_strides_0 = const()[name = string("hidden_states_53_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_53_pad_0 = const()[name = string("hidden_states_53_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_53_dilations_0 = const()[name = string("hidden_states_53_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_53_groups_0 = const()[name = string("hidden_states_53_groups_0"), val = int32(1)];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_proj_out_weight_to_fp16 = const()[name = string("down_blocks_2_attentions_0_proj_out_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(169897152)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_0_proj_out_bias_to_fp16 = const()[name = string("down_blocks_2_attentions_0_proj_out_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(173174016)))];
            tensor<fp16, [1, 1280, 16, 16]> hidden_states_53_cast_fp16 = conv(bias = down_blocks_2_attentions_0_proj_out_bias_to_fp16, dilations = hidden_states_53_dilations_0, groups = hidden_states_53_groups_0, pad = hidden_states_53_pad_0, pad_type = hidden_states_53_pad_type_0, strides = hidden_states_53_strides_0, weight = down_blocks_2_attentions_0_proj_out_weight_to_fp16, x = input_89_cast_fp16)[name = string("hidden_states_53_cast_fp16")];
            tensor<fp16, [1, 1280, 16, 16]> hidden_states_55_cast_fp16 = add(x = hidden_states_53_cast_fp16, y = hidden_states_41_cast_fp16)[name = string("hidden_states_55_cast_fp16")];
            int32 var_1812 = const()[name = string("op_1812"), val = int32(1)];
            bool input_91_interleave_0 = const()[name = string("input_91_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 2560, 16, 16]> input_91_cast_fp16 = concat(axis = var_1812, interleave = input_91_interleave_0, values = (hidden_states_55_cast_fp16, hidden_states_55_cast_fp16))[name = string("input_91_cast_fp16")];
            tensor<int32, [5]> reshape_36_shape_0 = const()[name = string("reshape_36_shape_0"), val = tensor<int32, [5]>([1, 32, 80, 16, 16])];
            tensor<fp16, [1, 32, 80, 16, 16]> reshape_36_cast_fp16 = reshape(shape = reshape_36_shape_0, x = input_91_cast_fp16)[name = string("reshape_36_cast_fp16")];
            tensor<int32, [3]> reduce_mean_27_axes_0 = const()[name = string("reduce_mean_27_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            bool reduce_mean_27_keep_dims_0 = const()[name = string("reduce_mean_27_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 1, 1, 1]> reduce_mean_27_cast_fp16 = reduce_mean(axes = reduce_mean_27_axes_0, keep_dims = reduce_mean_27_keep_dims_0, x = reshape_36_cast_fp16)[name = string("reduce_mean_27_cast_fp16")];
            tensor<fp16, [1, 32, 80, 16, 16]> sub_18_cast_fp16 = sub(x = reshape_36_cast_fp16, y = reduce_mean_27_cast_fp16)[name = string("sub_18_cast_fp16")];
            tensor<fp16, [1, 32, 80, 16, 16]> square_9_cast_fp16 = square(x = sub_18_cast_fp16)[name = string("square_9_cast_fp16")];
            tensor<int32, [3]> reduce_mean_29_axes_0 = const()[name = string("reduce_mean_29_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            bool reduce_mean_29_keep_dims_0 = const()[name = string("reduce_mean_29_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 1, 1, 1]> reduce_mean_29_cast_fp16 = reduce_mean(axes = reduce_mean_29_axes_0, keep_dims = reduce_mean_29_keep_dims_0, x = square_9_cast_fp16)[name = string("reduce_mean_29_cast_fp16")];
            fp16 add_18_y_0_to_fp16 = const()[name = string("add_18_y_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 32, 1, 1, 1]> add_18_cast_fp16 = add(x = reduce_mean_29_cast_fp16, y = add_18_y_0_to_fp16)[name = string("add_18_cast_fp16")];
            tensor<fp16, [1, 32, 1, 1, 1]> sqrt_9_cast_fp16 = sqrt(x = add_18_cast_fp16)[name = string("sqrt_9_cast_fp16")];
            tensor<fp16, [1, 32, 80, 16, 16]> real_div_9_cast_fp16 = real_div(x = sub_18_cast_fp16, y = sqrt_9_cast_fp16)[name = string("real_div_9_cast_fp16")];
            tensor<int32, [4]> reshape_37_shape_0 = const()[name = string("reshape_37_shape_0"), val = tensor<int32, [4]>([1, 2560, 16, 16])];
            tensor<fp16, [1, 2560, 16, 16]> reshape_37_cast_fp16 = reshape(shape = reshape_37_shape_0, x = real_div_9_cast_fp16)[name = string("reshape_37_cast_fp16")];
            tensor<fp16, [2560]> add_19_mean_0_to_fp16 = const()[name = string("add_19_mean_0_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(173176640)))];
            tensor<fp16, [2560]> add_19_variance_0_to_fp16 = const()[name = string("add_19_variance_0_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(173181824)))];
            tensor<fp16, [2560]> add_19_gamma_0_to_fp16 = const()[name = string("add_19_gamma_0_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(173187008)))];
            tensor<fp16, [2560]> add_19_beta_0_to_fp16 = const()[name = string("add_19_beta_0_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(173192192)))];
            fp16 add_19_epsilon_0_to_fp16 = const()[name = string("add_19_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 2560, 16, 16]> add_19_cast_fp16 = batch_norm(beta = add_19_beta_0_to_fp16, epsilon = add_19_epsilon_0_to_fp16, gamma = add_19_gamma_0_to_fp16, mean = add_19_mean_0_to_fp16, variance = add_19_variance_0_to_fp16, x = reshape_37_cast_fp16)[name = string("add_19_cast_fp16")];
            tensor<fp16, [1, 2560, 16, 16]> input_95_cast_fp16 = silu(x = add_19_cast_fp16)[name = string("input_95_cast_fp16")];
            string hidden_states_57_pad_type_0 = const()[name = string("hidden_states_57_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> hidden_states_57_pad_0 = const()[name = string("hidden_states_57_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> hidden_states_57_strides_0 = const()[name = string("hidden_states_57_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> hidden_states_57_dilations_0 = const()[name = string("hidden_states_57_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_57_groups_0 = const()[name = string("hidden_states_57_groups_0"), val = int32(1)];
            tensor<fp16, [1280, 2560, 3, 3]> up_blocks_0_resnets_0_conv1_weight_to_fp16 = const()[name = string("up_blocks_0_resnets_0_conv1_weight_to_fp16"), val = tensor<fp16, [1280, 2560, 3, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(173197376)))];
            tensor<fp16, [1280]> up_blocks_0_resnets_0_conv1_bias_to_fp16 = const()[name = string("up_blocks_0_resnets_0_conv1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(232179840)))];
            tensor<fp16, [1, 1280, 16, 16]> hidden_states_57_cast_fp16 = conv(bias = up_blocks_0_resnets_0_conv1_bias_to_fp16, dilations = hidden_states_57_dilations_0, groups = hidden_states_57_groups_0, pad = hidden_states_57_pad_0, pad_type = hidden_states_57_pad_type_0, strides = hidden_states_57_strides_0, weight = up_blocks_0_resnets_0_conv1_weight_to_fp16, x = input_95_cast_fp16)[name = string("hidden_states_57_cast_fp16")];
            string temb_7_pad_type_0 = const()[name = string("temb_7_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> temb_7_strides_0 = const()[name = string("temb_7_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> temb_7_pad_0 = const()[name = string("temb_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> temb_7_dilations_0 = const()[name = string("temb_7_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 temb_7_groups_0 = const()[name = string("temb_7_groups_0"), val = int32(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_resnets_0_time_emb_proj_weight_to_fp16 = const()[name = string("up_blocks_0_resnets_0_time_emb_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(232182464)))];
            tensor<fp16, [1280]> up_blocks_0_resnets_0_time_emb_proj_bias_to_fp16 = const()[name = string("up_blocks_0_resnets_0_time_emb_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(235459328)))];
            tensor<fp16, [1, 1280, 1, 1]> temb_7_cast_fp16 = conv(bias = up_blocks_0_resnets_0_time_emb_proj_bias_to_fp16, dilations = temb_7_dilations_0, groups = temb_7_groups_0, pad = temb_7_pad_0, pad_type = temb_7_pad_type_0, strides = temb_7_strides_0, weight = up_blocks_0_resnets_0_time_emb_proj_weight_to_fp16, x = input_15_cast_fp16_1)[name = string("temb_7_cast_fp16")];
            tensor<fp16, [1, 1280, 16, 16]> input_99_cast_fp16 = add(x = hidden_states_57_cast_fp16, y = temb_7_cast_fp16)[name = string("input_99_cast_fp16")];
            tensor<int32, [5]> reshape_40_shape_0 = const()[name = string("reshape_40_shape_0"), val = tensor<int32, [5]>([1, 32, 40, 16, 16])];
            tensor<fp16, [1, 32, 40, 16, 16]> reshape_40_cast_fp16 = reshape(shape = reshape_40_shape_0, x = input_99_cast_fp16)[name = string("reshape_40_cast_fp16")];
            tensor<int32, [3]> reduce_mean_30_axes_0 = const()[name = string("reduce_mean_30_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            bool reduce_mean_30_keep_dims_0 = const()[name = string("reduce_mean_30_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 1, 1, 1]> reduce_mean_30_cast_fp16 = reduce_mean(axes = reduce_mean_30_axes_0, keep_dims = reduce_mean_30_keep_dims_0, x = reshape_40_cast_fp16)[name = string("reduce_mean_30_cast_fp16")];
            tensor<fp16, [1, 32, 40, 16, 16]> sub_20_cast_fp16 = sub(x = reshape_40_cast_fp16, y = reduce_mean_30_cast_fp16)[name = string("sub_20_cast_fp16")];
            tensor<fp16, [1, 32, 40, 16, 16]> square_10_cast_fp16 = square(x = sub_20_cast_fp16)[name = string("square_10_cast_fp16")];
            tensor<int32, [3]> reduce_mean_32_axes_0 = const()[name = string("reduce_mean_32_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            bool reduce_mean_32_keep_dims_0 = const()[name = string("reduce_mean_32_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 1, 1, 1]> reduce_mean_32_cast_fp16 = reduce_mean(axes = reduce_mean_32_axes_0, keep_dims = reduce_mean_32_keep_dims_0, x = square_10_cast_fp16)[name = string("reduce_mean_32_cast_fp16")];
            fp16 add_20_y_0_to_fp16 = const()[name = string("add_20_y_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 32, 1, 1, 1]> add_20_cast_fp16 = add(x = reduce_mean_32_cast_fp16, y = add_20_y_0_to_fp16)[name = string("add_20_cast_fp16")];
            tensor<fp16, [1, 32, 1, 1, 1]> sqrt_10_cast_fp16 = sqrt(x = add_20_cast_fp16)[name = string("sqrt_10_cast_fp16")];
            tensor<fp16, [1, 32, 40, 16, 16]> real_div_10_cast_fp16 = real_div(x = sub_20_cast_fp16, y = sqrt_10_cast_fp16)[name = string("real_div_10_cast_fp16")];
            tensor<int32, [4]> reshape_41_shape_0 = const()[name = string("reshape_41_shape_0"), val = tensor<int32, [4]>([1, 1280, 16, 16])];
            tensor<fp16, [1, 1280, 16, 16]> reshape_41_cast_fp16 = reshape(shape = reshape_41_shape_0, x = real_div_10_cast_fp16)[name = string("reshape_41_cast_fp16")];
            tensor<fp16, [1280]> add_21_gamma_0_to_fp16 = const()[name = string("add_21_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(235461952)))];
            tensor<fp16, [1280]> add_21_beta_0_to_fp16 = const()[name = string("add_21_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(235464576)))];
            fp16 add_21_epsilon_0_to_fp16 = const()[name = string("add_21_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1280, 16, 16]> add_21_cast_fp16 = batch_norm(beta = add_21_beta_0_to_fp16, epsilon = add_21_epsilon_0_to_fp16, gamma = add_21_gamma_0_to_fp16, mean = add_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_41_cast_fp16)[name = string("add_21_cast_fp16")];
            tensor<fp16, [1, 1280, 16, 16]> input_103_cast_fp16 = silu(x = add_21_cast_fp16)[name = string("input_103_cast_fp16")];
            string hidden_states_59_pad_type_0 = const()[name = string("hidden_states_59_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> hidden_states_59_pad_0 = const()[name = string("hidden_states_59_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> hidden_states_59_strides_0 = const()[name = string("hidden_states_59_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> hidden_states_59_dilations_0 = const()[name = string("hidden_states_59_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_59_groups_0 = const()[name = string("hidden_states_59_groups_0"), val = int32(1)];
            tensor<fp16, [1280, 1280, 3, 3]> up_blocks_0_resnets_0_conv2_weight_to_fp16 = const()[name = string("up_blocks_0_resnets_0_conv2_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 3, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(235467200)))];
            tensor<fp16, [1280]> up_blocks_0_resnets_0_conv2_bias_to_fp16 = const()[name = string("up_blocks_0_resnets_0_conv2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(264958464)))];
            tensor<fp16, [1, 1280, 16, 16]> hidden_states_59_cast_fp16 = conv(bias = up_blocks_0_resnets_0_conv2_bias_to_fp16, dilations = hidden_states_59_dilations_0, groups = hidden_states_59_groups_0, pad = hidden_states_59_pad_0, pad_type = hidden_states_59_pad_type_0, strides = hidden_states_59_strides_0, weight = up_blocks_0_resnets_0_conv2_weight_to_fp16, x = input_103_cast_fp16)[name = string("hidden_states_59_cast_fp16")];
            string x_5_pad_type_0 = const()[name = string("x_5_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> x_5_strides_0 = const()[name = string("x_5_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> x_5_pad_0 = const()[name = string("x_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> x_5_dilations_0 = const()[name = string("x_5_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 x_5_groups_0 = const()[name = string("x_5_groups_0"), val = int32(1)];
            tensor<fp16, [1280, 2560, 1, 1]> up_blocks_0_resnets_0_conv_shortcut_weight_to_fp16 = const()[name = string("up_blocks_0_resnets_0_conv_shortcut_weight_to_fp16"), val = tensor<fp16, [1280, 2560, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(264961088)))];
            tensor<fp16, [1280]> up_blocks_0_resnets_0_conv_shortcut_bias_to_fp16 = const()[name = string("up_blocks_0_resnets_0_conv_shortcut_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271514752)))];
            tensor<fp16, [1, 1280, 16, 16]> 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, x = input_91_cast_fp16)[name = string("x_5_cast_fp16")];
            tensor<fp16, [1, 1280, 16, 16]> hidden_states_61_cast_fp16_1 = add(x = x_5_cast_fp16, y = hidden_states_59_cast_fp16)[name = string("hidden_states_61_cast_fp16")];
            tensor<int32, [5]> reshape_44_shape_0 = const()[name = string("reshape_44_shape_0"), val = tensor<int32, [5]>([1, 32, 40, 16, 16])];
            tensor<fp16, [1, 32, 40, 16, 16]> reshape_44_cast_fp16 = reshape(shape = reshape_44_shape_0, x = hidden_states_61_cast_fp16_1)[name = string("reshape_44_cast_fp16")];
            tensor<int32, [3]> reduce_mean_33_axes_0 = const()[name = string("reduce_mean_33_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            bool reduce_mean_33_keep_dims_0 = const()[name = string("reduce_mean_33_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 1, 1, 1]> reduce_mean_33_cast_fp16 = reduce_mean(axes = reduce_mean_33_axes_0, keep_dims = reduce_mean_33_keep_dims_0, x = reshape_44_cast_fp16)[name = string("reduce_mean_33_cast_fp16")];
            tensor<fp16, [1, 32, 40, 16, 16]> sub_22_cast_fp16 = sub(x = reshape_44_cast_fp16, y = reduce_mean_33_cast_fp16)[name = string("sub_22_cast_fp16")];
            tensor<fp16, [1, 32, 40, 16, 16]> square_11_cast_fp16 = square(x = sub_22_cast_fp16)[name = string("square_11_cast_fp16")];
            tensor<int32, [3]> reduce_mean_35_axes_0 = const()[name = string("reduce_mean_35_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            bool reduce_mean_35_keep_dims_0 = const()[name = string("reduce_mean_35_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 1, 1, 1]> reduce_mean_35_cast_fp16 = reduce_mean(axes = reduce_mean_35_axes_0, keep_dims = reduce_mean_35_keep_dims_0, x = square_11_cast_fp16)[name = string("reduce_mean_35_cast_fp16")];
            fp16 add_22_y_0_to_fp16 = const()[name = string("add_22_y_0_to_fp16"), val = fp16(0x1.1p-20)];
            tensor<fp16, [1, 32, 1, 1, 1]> add_22_cast_fp16 = add(x = reduce_mean_35_cast_fp16, y = add_22_y_0_to_fp16)[name = string("add_22_cast_fp16")];
            tensor<fp16, [1, 32, 1, 1, 1]> sqrt_11_cast_fp16 = sqrt(x = add_22_cast_fp16)[name = string("sqrt_11_cast_fp16")];
            tensor<fp16, [1, 32, 40, 16, 16]> real_div_11_cast_fp16 = real_div(x = sub_22_cast_fp16, y = sqrt_11_cast_fp16)[name = string("real_div_11_cast_fp16")];
            tensor<int32, [4]> reshape_45_shape_0 = const()[name = string("reshape_45_shape_0"), val = tensor<int32, [4]>([1, 1280, 16, 16])];
            tensor<fp16, [1, 1280, 16, 16]> reshape_45_cast_fp16 = reshape(shape = reshape_45_shape_0, x = real_div_11_cast_fp16)[name = string("reshape_45_cast_fp16")];
            tensor<fp16, [1280]> add_23_gamma_0_to_fp16 = const()[name = string("add_23_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271517376)))];
            tensor<fp16, [1280]> add_23_beta_0_to_fp16 = const()[name = string("add_23_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271520000)))];
            fp16 add_23_epsilon_0_to_fp16 = const()[name = string("add_23_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1280, 16, 16]> add_23_cast_fp16 = batch_norm(beta = add_23_beta_0_to_fp16, epsilon = add_23_epsilon_0_to_fp16, gamma = add_23_gamma_0_to_fp16, mean = add_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_45_cast_fp16)[name = string("add_23_cast_fp16")];
            string hidden_states_63_pad_type_0 = const()[name = string("hidden_states_63_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> hidden_states_63_strides_0 = const()[name = string("hidden_states_63_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_63_pad_0 = const()[name = string("hidden_states_63_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_63_dilations_0 = const()[name = string("hidden_states_63_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_63_groups_0 = const()[name = string("hidden_states_63_groups_0"), val = int32(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_proj_in_weight_to_fp16 = const()[name = string("up_blocks_0_attentions_0_proj_in_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271522624)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_proj_in_bias_to_fp16 = const()[name = string("up_blocks_0_attentions_0_proj_in_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274799488)))];
            tensor<fp16, [1, 1280, 16, 16]> hidden_states_63_cast_fp16 = conv(bias = up_blocks_0_attentions_0_proj_in_bias_to_fp16, dilations = hidden_states_63_dilations_0, groups = hidden_states_63_groups_0, pad = hidden_states_63_pad_0, pad_type = hidden_states_63_pad_type_0, strides = hidden_states_63_strides_0, weight = up_blocks_0_attentions_0_proj_in_weight_to_fp16, x = add_23_cast_fp16)[name = string("hidden_states_63_cast_fp16")];
            tensor<int32, [4]> var_1893 = const()[name = string("op_1893"), val = tensor<int32, [4]>([1, 1280, 1, 256])];
            tensor<fp16, [1, 1280, 1, 256]> inputs_19_cast_fp16 = reshape(shape = var_1893, x = hidden_states_63_cast_fp16)[name = string("inputs_19_cast_fp16")];
            tensor<int32, [1]> hidden_states_65_axes_0 = const()[name = string("hidden_states_65_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_65_gamma_0_to_fp16 = const()[name = string("hidden_states_65_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274802112)))];
            tensor<fp16, [1280]> hidden_states_65_beta_0_to_fp16 = const()[name = string("hidden_states_65_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274804736)))];
            fp16 var_1909_to_fp16 = const()[name = string("op_1909_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1280, 1, 256]> hidden_states_65_cast_fp16 = layer_norm(axes = hidden_states_65_axes_0, beta = hidden_states_65_beta_0_to_fp16, epsilon = var_1909_to_fp16, gamma = hidden_states_65_gamma_0_to_fp16, x = inputs_19_cast_fp16)[name = string("hidden_states_65_cast_fp16")];
            string q_13_pad_type_0 = const()[name = string("q_13_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> q_13_strides_0 = const()[name = string("q_13_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_13_pad_0 = const()[name = string("q_13_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_13_dilations_0 = const()[name = string("q_13_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 q_13_groups_0 = const()[name = string("q_13_groups_0"), val = int32(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = string("up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274807360)))];
            tensor<fp16, [1, 1280, 1, 256]> q_13_cast_fp16 = conv(dilations = q_13_dilations_0, groups = q_13_groups_0, pad = q_13_pad_0, pad_type = q_13_pad_type_0, strides = q_13_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_65_cast_fp16)[name = string("q_13_cast_fp16")];
            string k_25_pad_type_0 = const()[name = string("k_25_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> k_25_strides_0 = const()[name = string("k_25_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_25_pad_0 = const()[name = string("k_25_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_25_dilations_0 = const()[name = string("k_25_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 k_25_groups_0 = const()[name = string("k_25_groups_0"), val = int32(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = string("up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278084224)))];
            tensor<fp16, [1, 1280, 1, 256]> k_25_cast_fp16 = conv(dilations = k_25_dilations_0, groups = k_25_groups_0, pad = k_25_pad_0, pad_type = k_25_pad_type_0, strides = k_25_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_65_cast_fp16)[name = string("k_25_cast_fp16")];
            string v_13_pad_type_0 = const()[name = string("v_13_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> v_13_strides_0 = const()[name = string("v_13_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_13_pad_0 = const()[name = string("v_13_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_13_dilations_0 = const()[name = string("v_13_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 v_13_groups_0 = const()[name = string("v_13_groups_0"), val = int32(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = string("up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(281361088)))];
            tensor<fp16, [1, 1280, 1, 256]> v_13_cast_fp16 = conv(dilations = v_13_dilations_0, groups = v_13_groups_0, pad = v_13_pad_0, pad_type = v_13_pad_type_0, strides = v_13_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_65_cast_fp16)[name = string("v_13_cast_fp16")];
            tensor<int32, [4]> var_1942_begin_0 = const()[name = string("op_1942_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1942_end_0 = const()[name = string("op_1942_end_0"), val = tensor<int32, [4]>([1, 160, 1, 256])];
            tensor<bool, [4]> var_1942_end_mask_0 = const()[name = string("op_1942_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 256]> var_1942_cast_fp16 = slice_by_index(begin = var_1942_begin_0, end = var_1942_end_0, end_mask = var_1942_end_mask_0, x = q_13_cast_fp16)[name = string("op_1942_cast_fp16")];
            tensor<int32, [4]> var_1946_begin_0 = const()[name = string("op_1946_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_1946_end_0 = const()[name = string("op_1946_end_0"), val = tensor<int32, [4]>([1, 320, 1, 256])];
            tensor<bool, [4]> var_1946_end_mask_0 = const()[name = string("op_1946_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 256]> var_1946_cast_fp16 = slice_by_index(begin = var_1946_begin_0, end = var_1946_end_0, end_mask = var_1946_end_mask_0, x = q_13_cast_fp16)[name = string("op_1946_cast_fp16")];
            tensor<int32, [4]> var_1950_begin_0 = const()[name = string("op_1950_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_1950_end_0 = const()[name = string("op_1950_end_0"), val = tensor<int32, [4]>([1, 480, 1, 256])];
            tensor<bool, [4]> var_1950_end_mask_0 = const()[name = string("op_1950_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 256]> var_1950_cast_fp16 = slice_by_index(begin = var_1950_begin_0, end = var_1950_end_0, end_mask = var_1950_end_mask_0, x = q_13_cast_fp16)[name = string("op_1950_cast_fp16")];
            tensor<int32, [4]> var_1954_begin_0 = const()[name = string("op_1954_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_1954_end_0 = const()[name = string("op_1954_end_0"), val = tensor<int32, [4]>([1, 640, 1, 256])];
            tensor<bool, [4]> var_1954_end_mask_0 = const()[name = string("op_1954_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 256]> var_1954_cast_fp16 = slice_by_index(begin = var_1954_begin_0, end = var_1954_end_0, end_mask = var_1954_end_mask_0, x = q_13_cast_fp16)[name = string("op_1954_cast_fp16")];
            tensor<int32, [4]> var_1958_begin_0 = const()[name = string("op_1958_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_1958_end_0 = const()[name = string("op_1958_end_0"), val = tensor<int32, [4]>([1, 800, 1, 256])];
            tensor<bool, [4]> var_1958_end_mask_0 = const()[name = string("op_1958_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 256]> var_1958_cast_fp16 = slice_by_index(begin = var_1958_begin_0, end = var_1958_end_0, end_mask = var_1958_end_mask_0, x = q_13_cast_fp16)[name = string("op_1958_cast_fp16")];
            tensor<int32, [4]> var_1962_begin_0 = const()[name = string("op_1962_begin_0"), val = tensor<int32, [4]>([0, 800, 0, 0])];
            tensor<int32, [4]> var_1962_end_0 = const()[name = string("op_1962_end_0"), val = tensor<int32, [4]>([1, 960, 1, 256])];
            tensor<bool, [4]> var_1962_end_mask_0 = const()[name = string("op_1962_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 256]> var_1962_cast_fp16 = slice_by_index(begin = var_1962_begin_0, end = var_1962_end_0, end_mask = var_1962_end_mask_0, x = q_13_cast_fp16)[name = string("op_1962_cast_fp16")];
            tensor<int32, [4]> var_1966_begin_0 = const()[name = string("op_1966_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_1966_end_0 = const()[name = string("op_1966_end_0"), val = tensor<int32, [4]>([1, 1120, 1, 256])];
            tensor<bool, [4]> var_1966_end_mask_0 = const()[name = string("op_1966_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 256]> var_1966_cast_fp16 = slice_by_index(begin = var_1966_begin_0, end = var_1966_end_0, end_mask = var_1966_end_mask_0, x = q_13_cast_fp16)[name = string("op_1966_cast_fp16")];
            tensor<int32, [4]> var_1970_begin_0 = const()[name = string("op_1970_begin_0"), val = tensor<int32, [4]>([0, 1120, 0, 0])];
            tensor<int32, [4]> var_1970_end_0 = const()[name = string("op_1970_end_0"), val = tensor<int32, [4]>([1, 1, 1, 256])];
            tensor<bool, [4]> var_1970_end_mask_0 = const()[name = string("op_1970_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 160, 1, 256]> var_1970_cast_fp16 = slice_by_index(begin = var_1970_begin_0, end = var_1970_end_0, end_mask = var_1970_end_mask_0, x = q_13_cast_fp16)[name = string("op_1970_cast_fp16")];
            tensor<int32, [4]> k_27_perm_0 = const()[name = string("k_27_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_1977_begin_0 = const()[name = string("op_1977_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1977_end_0 = const()[name = string("op_1977_end_0"), val = tensor<int32, [4]>([1, 256, 1, 160])];
            tensor<bool, [4]> var_1977_end_mask_0 = const()[name = string("op_1977_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 256, 1, 1280]> k_27_cast_fp16 = transpose(perm = k_27_perm_0, x = k_25_cast_fp16)[name = string("transpose_1")];
            tensor<fp16, [1, 256, 1, 160]> var_1977_cast_fp16 = slice_by_index(begin = var_1977_begin_0, end = var_1977_end_0, end_mask = var_1977_end_mask_0, x = k_27_cast_fp16)[name = string("op_1977_cast_fp16")];
            tensor<int32, [4]> var_1981_begin_0 = const()[name = string("op_1981_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 160])];
            tensor<int32, [4]> var_1981_end_0 = const()[name = string("op_1981_end_0"), val = tensor<int32, [4]>([1, 256, 1, 320])];
            tensor<bool, [4]> var_1981_end_mask_0 = const()[name = string("op_1981_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 256, 1, 160]> var_1981_cast_fp16 = slice_by_index(begin = var_1981_begin_0, end = var_1981_end_0, end_mask = var_1981_end_mask_0, x = k_27_cast_fp16)[name = string("op_1981_cast_fp16")];
            tensor<int32, [4]> var_1985_begin_0 = const()[name = string("op_1985_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_1985_end_0 = const()[name = string("op_1985_end_0"), val = tensor<int32, [4]>([1, 256, 1, 480])];
            tensor<bool, [4]> var_1985_end_mask_0 = const()[name = string("op_1985_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 256, 1, 160]> var_1985_cast_fp16 = slice_by_index(begin = var_1985_begin_0, end = var_1985_end_0, end_mask = var_1985_end_mask_0, x = k_27_cast_fp16)[name = string("op_1985_cast_fp16")];
            tensor<int32, [4]> var_1989_begin_0 = const()[name = string("op_1989_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 480])];
            tensor<int32, [4]> var_1989_end_0 = const()[name = string("op_1989_end_0"), val = tensor<int32, [4]>([1, 256, 1, 640])];
            tensor<bool, [4]> var_1989_end_mask_0 = const()[name = string("op_1989_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 256, 1, 160]> var_1989_cast_fp16 = slice_by_index(begin = var_1989_begin_0, end = var_1989_end_0, end_mask = var_1989_end_mask_0, x = k_27_cast_fp16)[name = string("op_1989_cast_fp16")];
            tensor<int32, [4]> var_1993_begin_0 = const()[name = string("op_1993_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_1993_end_0 = const()[name = string("op_1993_end_0"), val = tensor<int32, [4]>([1, 256, 1, 800])];
            tensor<bool, [4]> var_1993_end_mask_0 = const()[name = string("op_1993_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 256, 1, 160]> var_1993_cast_fp16 = slice_by_index(begin = var_1993_begin_0, end = var_1993_end_0, end_mask = var_1993_end_mask_0, x = k_27_cast_fp16)[name = string("op_1993_cast_fp16")];
            tensor<int32, [4]> var_1997_begin_0 = const()[name = string("op_1997_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 800])];
            tensor<int32, [4]> var_1997_end_0 = const()[name = string("op_1997_end_0"), val = tensor<int32, [4]>([1, 256, 1, 960])];
            tensor<bool, [4]> var_1997_end_mask_0 = const()[name = string("op_1997_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 256, 1, 160]> var_1997_cast_fp16 = slice_by_index(begin = var_1997_begin_0, end = var_1997_end_0, end_mask = var_1997_end_mask_0, x = k_27_cast_fp16)[name = string("op_1997_cast_fp16")];
            tensor<int32, [4]> var_2001_begin_0 = const()[name = string("op_2001_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_2001_end_0 = const()[name = string("op_2001_end_0"), val = tensor<int32, [4]>([1, 256, 1, 1120])];
            tensor<bool, [4]> var_2001_end_mask_0 = const()[name = string("op_2001_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 256, 1, 160]> var_2001_cast_fp16 = slice_by_index(begin = var_2001_begin_0, end = var_2001_end_0, end_mask = var_2001_end_mask_0, x = k_27_cast_fp16)[name = string("op_2001_cast_fp16")];
            tensor<int32, [4]> var_2005_begin_0 = const()[name = string("op_2005_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1120])];
            tensor<int32, [4]> var_2005_end_0 = const()[name = string("op_2005_end_0"), val = tensor<int32, [4]>([1, 256, 1, 1])];
            tensor<bool, [4]> var_2005_end_mask_0 = const()[name = string("op_2005_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 256, 1, 160]> var_2005_cast_fp16 = slice_by_index(begin = var_2005_begin_0, end = var_2005_end_0, end_mask = var_2005_end_mask_0, x = k_27_cast_fp16)[name = string("op_2005_cast_fp16")];
            tensor<int32, [4]> var_2007_begin_0 = const()[name = string("op_2007_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2007_end_0 = const()[name = string("op_2007_end_0"), val = tensor<int32, [4]>([1, 160, 1, 256])];
            tensor<bool, [4]> var_2007_end_mask_0 = const()[name = string("op_2007_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 256]> var_2007_cast_fp16 = slice_by_index(begin = var_2007_begin_0, end = var_2007_end_0, end_mask = var_2007_end_mask_0, x = v_13_cast_fp16)[name = string("op_2007_cast_fp16")];
            tensor<int32, [4]> var_2011_begin_0 = const()[name = string("op_2011_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_2011_end_0 = const()[name = string("op_2011_end_0"), val = tensor<int32, [4]>([1, 320, 1, 256])];
            tensor<bool, [4]> var_2011_end_mask_0 = const()[name = string("op_2011_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 256]> var_2011_cast_fp16 = slice_by_index(begin = var_2011_begin_0, end = var_2011_end_0, end_mask = var_2011_end_mask_0, x = v_13_cast_fp16)[name = string("op_2011_cast_fp16")];
            tensor<int32, [4]> var_2015_begin_0 = const()[name = string("op_2015_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_2015_end_0 = const()[name = string("op_2015_end_0"), val = tensor<int32, [4]>([1, 480, 1, 256])];
            tensor<bool, [4]> var_2015_end_mask_0 = const()[name = string("op_2015_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 256]> var_2015_cast_fp16 = slice_by_index(begin = var_2015_begin_0, end = var_2015_end_0, end_mask = var_2015_end_mask_0, x = v_13_cast_fp16)[name = string("op_2015_cast_fp16")];
            tensor<int32, [4]> var_2019_begin_0 = const()[name = string("op_2019_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_2019_end_0 = const()[name = string("op_2019_end_0"), val = tensor<int32, [4]>([1, 640, 1, 256])];
            tensor<bool, [4]> var_2019_end_mask_0 = const()[name = string("op_2019_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 256]> var_2019_cast_fp16 = slice_by_index(begin = var_2019_begin_0, end = var_2019_end_0, end_mask = var_2019_end_mask_0, x = v_13_cast_fp16)[name = string("op_2019_cast_fp16")];
            tensor<int32, [4]> var_2023_begin_0 = const()[name = string("op_2023_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_2023_end_0 = const()[name = string("op_2023_end_0"), val = tensor<int32, [4]>([1, 800, 1, 256])];
            tensor<bool, [4]> var_2023_end_mask_0 = const()[name = string("op_2023_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 256]> var_2023_cast_fp16 = slice_by_index(begin = var_2023_begin_0, end = var_2023_end_0, end_mask = var_2023_end_mask_0, x = v_13_cast_fp16)[name = string("op_2023_cast_fp16")];
            tensor<int32, [4]> var_2027_begin_0 = const()[name = string("op_2027_begin_0"), val = tensor<int32, [4]>([0, 800, 0, 0])];
            tensor<int32, [4]> var_2027_end_0 = const()[name = string("op_2027_end_0"), val = tensor<int32, [4]>([1, 960, 1, 256])];
            tensor<bool, [4]> var_2027_end_mask_0 = const()[name = string("op_2027_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 256]> var_2027_cast_fp16 = slice_by_index(begin = var_2027_begin_0, end = var_2027_end_0, end_mask = var_2027_end_mask_0, x = v_13_cast_fp16)[name = string("op_2027_cast_fp16")];
            tensor<int32, [4]> var_2031_begin_0 = const()[name = string("op_2031_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_2031_end_0 = const()[name = string("op_2031_end_0"), val = tensor<int32, [4]>([1, 1120, 1, 256])];
            tensor<bool, [4]> var_2031_end_mask_0 = const()[name = string("op_2031_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 256]> var_2031_cast_fp16 = slice_by_index(begin = var_2031_begin_0, end = var_2031_end_0, end_mask = var_2031_end_mask_0, x = v_13_cast_fp16)[name = string("op_2031_cast_fp16")];
            tensor<int32, [4]> var_2035_begin_0 = const()[name = string("op_2035_begin_0"), val = tensor<int32, [4]>([0, 1120, 0, 0])];
            tensor<int32, [4]> var_2035_end_0 = const()[name = string("op_2035_end_0"), val = tensor<int32, [4]>([1, 1, 1, 256])];
            tensor<bool, [4]> var_2035_end_mask_0 = const()[name = string("op_2035_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 160, 1, 256]> var_2035_cast_fp16 = slice_by_index(begin = var_2035_begin_0, end = var_2035_end_0, end_mask = var_2035_end_mask_0, x = v_13_cast_fp16)[name = string("op_2035_cast_fp16")];
            string var_2039_equation_0 = const()[name = string("op_2039_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 256, 1, 256]> var_2039_cast_fp16 = einsum(equation = var_2039_equation_0, values = (var_1977_cast_fp16, var_1942_cast_fp16))[name = string("op_2039_cast_fp16")];
            fp16 var_2040_to_fp16 = const()[name = string("op_2040_to_fp16"), val = fp16(0x1.43cp-4)];
            tensor<fp16, [1, 256, 1, 256]> aw_97_cast_fp16 = mul(x = var_2039_cast_fp16, y = var_2040_to_fp16)[name = string("aw_97_cast_fp16")];
            string var_2043_equation_0 = const()[name = string("op_2043_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 256, 1, 256]> var_2043_cast_fp16 = einsum(equation = var_2043_equation_0, values = (var_1981_cast_fp16, var_1946_cast_fp16))[name = string("op_2043_cast_fp16")];
            fp16 var_2044_to_fp16 = const()[name = string("op_2044_to_fp16"), val = fp16(0x1.43cp-4)];
            tensor<fp16, [1, 256, 1, 256]> aw_99_cast_fp16 = mul(x = var_2043_cast_fp16, y = var_2044_to_fp16)[name = string("aw_99_cast_fp16")];
            string var_2047_equation_0 = const()[name = string("op_2047_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 256, 1, 256]> var_2047_cast_fp16 = einsum(equation = var_2047_equation_0, values = (var_1985_cast_fp16, var_1950_cast_fp16))[name = string("op_2047_cast_fp16")];
            fp16 var_2048_to_fp16 = const()[name = string("op_2048_to_fp16"), val = fp16(0x1.43cp-4)];
            tensor<fp16, [1, 256, 1, 256]> aw_101_cast_fp16 = mul(x = var_2047_cast_fp16, y = var_2048_to_fp16)[name = string("aw_101_cast_fp16")];
            string var_2051_equation_0 = const()[name = string("op_2051_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 256, 1, 256]> var_2051_cast_fp16 = einsum(equation = var_2051_equation_0, values = (var_1989_cast_fp16, var_1954_cast_fp16))[name = string("op_2051_cast_fp16")];
            fp16 var_2052_to_fp16 = const()[name = string("op_2052_to_fp16"), val = fp16(0x1.43cp-4)];
            tensor<fp16, [1, 256, 1, 256]> aw_103_cast_fp16 = mul(x = var_2051_cast_fp16, y = var_2052_to_fp16)[name = string("aw_103_cast_fp16")];
            string var_2055_equation_0 = const()[name = string("op_2055_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 256, 1, 256]> var_2055_cast_fp16 = einsum(equation = var_2055_equation_0, values = (var_1993_cast_fp16, var_1958_cast_fp16))[name = string("op_2055_cast_fp16")];
            fp16 var_2056_to_fp16 = const()[name = string("op_2056_to_fp16"), val = fp16(0x1.43cp-4)];
            tensor<fp16, [1, 256, 1, 256]> aw_105_cast_fp16 = mul(x = var_2055_cast_fp16, y = var_2056_to_fp16)[name = string("aw_105_cast_fp16")];
            string var_2059_equation_0 = const()[name = string("op_2059_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 256, 1, 256]> var_2059_cast_fp16 = einsum(equation = var_2059_equation_0, values = (var_1997_cast_fp16, var_1962_cast_fp16))[name = string("op_2059_cast_fp16")];
            fp16 var_2060_to_fp16 = const()[name = string("op_2060_to_fp16"), val = fp16(0x1.43cp-4)];
            tensor<fp16, [1, 256, 1, 256]> aw_107_cast_fp16 = mul(x = var_2059_cast_fp16, y = var_2060_to_fp16)[name = string("aw_107_cast_fp16")];
            string var_2063_equation_0 = const()[name = string("op_2063_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 256, 1, 256]> var_2063_cast_fp16 = einsum(equation = var_2063_equation_0, values = (var_2001_cast_fp16, var_1966_cast_fp16))[name = string("op_2063_cast_fp16")];
            fp16 var_2064_to_fp16 = const()[name = string("op_2064_to_fp16"), val = fp16(0x1.43cp-4)];
            tensor<fp16, [1, 256, 1, 256]> aw_109_cast_fp16 = mul(x = var_2063_cast_fp16, y = var_2064_to_fp16)[name = string("aw_109_cast_fp16")];
            string var_2067_equation_0 = const()[name = string("op_2067_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 256, 1, 256]> var_2067_cast_fp16 = einsum(equation = var_2067_equation_0, values = (var_2005_cast_fp16, var_1970_cast_fp16))[name = string("op_2067_cast_fp16")];
            fp16 var_2068_to_fp16 = const()[name = string("op_2068_to_fp16"), val = fp16(0x1.43cp-4)];
            tensor<fp16, [1, 256, 1, 256]> aw_111_cast_fp16 = mul(x = var_2067_cast_fp16, y = var_2068_to_fp16)[name = string("aw_111_cast_fp16")];
            tensor<fp16, [1, 256, 1, 256]> var_2070_cast_fp16 = softmax(axis = var_1812, x = aw_97_cast_fp16)[name = string("op_2070_cast_fp16")];
            tensor<fp16, [1, 256, 1, 256]> var_2071_cast_fp16 = softmax(axis = var_1812, x = aw_99_cast_fp16)[name = string("op_2071_cast_fp16")];
            tensor<fp16, [1, 256, 1, 256]> var_2072_cast_fp16 = softmax(axis = var_1812, x = aw_101_cast_fp16)[name = string("op_2072_cast_fp16")];
            tensor<fp16, [1, 256, 1, 256]> var_2073_cast_fp16 = softmax(axis = var_1812, x = aw_103_cast_fp16)[name = string("op_2073_cast_fp16")];
            tensor<fp16, [1, 256, 1, 256]> var_2074_cast_fp16 = softmax(axis = var_1812, x = aw_105_cast_fp16)[name = string("op_2074_cast_fp16")];
            tensor<fp16, [1, 256, 1, 256]> var_2075_cast_fp16 = softmax(axis = var_1812, x = aw_107_cast_fp16)[name = string("op_2075_cast_fp16")];
            tensor<fp16, [1, 256, 1, 256]> var_2076_cast_fp16 = softmax(axis = var_1812, x = aw_109_cast_fp16)[name = string("op_2076_cast_fp16")];
            tensor<fp16, [1, 256, 1, 256]> var_2077_cast_fp16 = softmax(axis = var_1812, x = aw_111_cast_fp16)[name = string("op_2077_cast_fp16")];
            string var_2079_equation_0 = const()[name = string("op_2079_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 160, 1, 256]> var_2079_cast_fp16 = einsum(equation = var_2079_equation_0, values = (var_2007_cast_fp16, var_2070_cast_fp16))[name = string("op_2079_cast_fp16")];
            string var_2081_equation_0 = const()[name = string("op_2081_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 160, 1, 256]> var_2081_cast_fp16 = einsum(equation = var_2081_equation_0, values = (var_2011_cast_fp16, var_2071_cast_fp16))[name = string("op_2081_cast_fp16")];
            string var_2083_equation_0 = const()[name = string("op_2083_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 160, 1, 256]> var_2083_cast_fp16 = einsum(equation = var_2083_equation_0, values = (var_2015_cast_fp16, var_2072_cast_fp16))[name = string("op_2083_cast_fp16")];
            string var_2085_equation_0 = const()[name = string("op_2085_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 160, 1, 256]> var_2085_cast_fp16 = einsum(equation = var_2085_equation_0, values = (var_2019_cast_fp16, var_2073_cast_fp16))[name = string("op_2085_cast_fp16")];
            string var_2087_equation_0 = const()[name = string("op_2087_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 160, 1, 256]> var_2087_cast_fp16 = einsum(equation = var_2087_equation_0, values = (var_2023_cast_fp16, var_2074_cast_fp16))[name = string("op_2087_cast_fp16")];
            string var_2089_equation_0 = const()[name = string("op_2089_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 160, 1, 256]> var_2089_cast_fp16 = einsum(equation = var_2089_equation_0, values = (var_2027_cast_fp16, var_2075_cast_fp16))[name = string("op_2089_cast_fp16")];
            string var_2091_equation_0 = const()[name = string("op_2091_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 160, 1, 256]> var_2091_cast_fp16 = einsum(equation = var_2091_equation_0, values = (var_2031_cast_fp16, var_2076_cast_fp16))[name = string("op_2091_cast_fp16")];
            string var_2093_equation_0 = const()[name = string("op_2093_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 160, 1, 256]> var_2093_cast_fp16 = einsum(equation = var_2093_equation_0, values = (var_2035_cast_fp16, var_2077_cast_fp16))[name = string("op_2093_cast_fp16")];
            bool input_107_interleave_0 = const()[name = string("input_107_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 1280, 1, 256]> input_107_cast_fp16 = concat(axis = var_1812, interleave = input_107_interleave_0, values = (var_2079_cast_fp16, var_2081_cast_fp16, var_2083_cast_fp16, var_2085_cast_fp16, var_2087_cast_fp16, var_2089_cast_fp16, var_2091_cast_fp16, var_2093_cast_fp16))[name = string("input_107_cast_fp16")];
            string var_2103_pad_type_0 = const()[name = string("op_2103_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_2103_strides_0 = const()[name = string("op_2103_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_2103_pad_0 = const()[name = string("op_2103_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_2103_dilations_0 = const()[name = string("op_2103_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_2103_groups_0 = const()[name = string("op_2103_groups_0"), val = int32(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = string("up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(284637952)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = string("up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287914816)))];
            tensor<fp16, [1, 1280, 1, 256]> var_2103_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_2103_dilations_0, groups = var_2103_groups_0, pad = var_2103_pad_0, pad_type = var_2103_pad_type_0, strides = var_2103_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_107_cast_fp16)[name = string("op_2103_cast_fp16")];
            tensor<fp16, [1, 1280, 1, 256]> inputs_21_cast_fp16 = add(x = var_2103_cast_fp16, y = inputs_19_cast_fp16)[name = string("inputs_21_cast_fp16")];
            tensor<int32, [1]> hidden_states_67_axes_0 = const()[name = string("hidden_states_67_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> hidden_states_67_gamma_0_to_fp16 = const()[name = string("hidden_states_67_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287917440)))];
            tensor<fp16, [1280]> hidden_states_67_beta_0_to_fp16 = const()[name = string("hidden_states_67_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287920064)))];
            fp16 var_2113_to_fp16 = const()[name = string("op_2113_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1280, 1, 256]> hidden_states_67_cast_fp16 = layer_norm(axes = hidden_states_67_axes_0, beta = hidden_states_67_beta_0_to_fp16, epsilon = var_2113_to_fp16, gamma = hidden_states_67_gamma_0_to_fp16, x = inputs_21_cast_fp16)[name = string("hidden_states_67_cast_fp16")];
            string q_15_pad_type_0 = const()[name = string("q_15_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> q_15_strides_0 = const()[name = string("q_15_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_15_pad_0 = const()[name = string("q_15_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_15_dilations_0 = const()[name = string("q_15_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 q_15_groups_0 = const()[name = string("q_15_groups_0"), val = int32(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = string("up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287922688)))];
            tensor<fp16, [1, 1280, 1, 256]> q_15_cast_fp16 = conv(dilations = q_15_dilations_0, groups = q_15_groups_0, pad = q_15_pad_0, pad_type = q_15_pad_type_0, strides = q_15_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_67_cast_fp16)[name = string("q_15_cast_fp16")];
            string k_29_pad_type_0 = const()[name = string("k_29_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> k_29_strides_0 = const()[name = string("k_29_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_29_pad_0 = const()[name = string("k_29_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_29_dilations_0 = const()[name = string("k_29_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 k_29_groups_0 = const()[name = string("k_29_groups_0"), val = int32(1)];
            tensor<fp16, [1280, 768, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = string("up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 768, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(291199552)))];
            tensor<fp16, [1, 1280, 1, 77]> k_29_cast_fp16 = conv(dilations = k_29_dilations_0, groups = k_29_groups_0, pad = k_29_pad_0, pad_type = k_29_pad_type_0, strides = k_29_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = string("k_29_cast_fp16")];
            string v_15_pad_type_0 = const()[name = string("v_15_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> v_15_strides_0 = const()[name = string("v_15_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_15_pad_0 = const()[name = string("v_15_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_15_dilations_0 = const()[name = string("v_15_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 v_15_groups_0 = const()[name = string("v_15_groups_0"), val = int32(1)];
            tensor<fp16, [1280, 768, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = string("up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 768, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(293165696)))];
            tensor<fp16, [1, 1280, 1, 77]> v_15_cast_fp16 = conv(dilations = v_15_dilations_0, groups = v_15_groups_0, pad = v_15_pad_0, pad_type = v_15_pad_type_0, strides = v_15_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = string("v_15_cast_fp16")];
            tensor<int32, [4]> var_2146_begin_0 = const()[name = string("op_2146_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2146_end_0 = const()[name = string("op_2146_end_0"), val = tensor<int32, [4]>([1, 160, 1, 256])];
            tensor<bool, [4]> var_2146_end_mask_0 = const()[name = string("op_2146_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 256]> var_2146_cast_fp16 = slice_by_index(begin = var_2146_begin_0, end = var_2146_end_0, end_mask = var_2146_end_mask_0, x = q_15_cast_fp16)[name = string("op_2146_cast_fp16")];
            tensor<int32, [4]> var_2150_begin_0 = const()[name = string("op_2150_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_2150_end_0 = const()[name = string("op_2150_end_0"), val = tensor<int32, [4]>([1, 320, 1, 256])];
            tensor<bool, [4]> var_2150_end_mask_0 = const()[name = string("op_2150_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 256]> var_2150_cast_fp16 = slice_by_index(begin = var_2150_begin_0, end = var_2150_end_0, end_mask = var_2150_end_mask_0, x = q_15_cast_fp16)[name = string("op_2150_cast_fp16")];
            tensor<int32, [4]> var_2154_begin_0 = const()[name = string("op_2154_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_2154_end_0 = const()[name = string("op_2154_end_0"), val = tensor<int32, [4]>([1, 480, 1, 256])];
            tensor<bool, [4]> var_2154_end_mask_0 = const()[name = string("op_2154_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 256]> var_2154_cast_fp16 = slice_by_index(begin = var_2154_begin_0, end = var_2154_end_0, end_mask = var_2154_end_mask_0, x = q_15_cast_fp16)[name = string("op_2154_cast_fp16")];
            tensor<int32, [4]> var_2158_begin_0 = const()[name = string("op_2158_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_2158_end_0 = const()[name = string("op_2158_end_0"), val = tensor<int32, [4]>([1, 640, 1, 256])];
            tensor<bool, [4]> var_2158_end_mask_0 = const()[name = string("op_2158_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 256]> var_2158_cast_fp16 = slice_by_index(begin = var_2158_begin_0, end = var_2158_end_0, end_mask = var_2158_end_mask_0, x = q_15_cast_fp16)[name = string("op_2158_cast_fp16")];
            tensor<int32, [4]> var_2162_begin_0 = const()[name = string("op_2162_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_2162_end_0 = const()[name = string("op_2162_end_0"), val = tensor<int32, [4]>([1, 800, 1, 256])];
            tensor<bool, [4]> var_2162_end_mask_0 = const()[name = string("op_2162_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 256]> var_2162_cast_fp16 = slice_by_index(begin = var_2162_begin_0, end = var_2162_end_0, end_mask = var_2162_end_mask_0, x = q_15_cast_fp16)[name = string("op_2162_cast_fp16")];
            tensor<int32, [4]> var_2166_begin_0 = const()[name = string("op_2166_begin_0"), val = tensor<int32, [4]>([0, 800, 0, 0])];
            tensor<int32, [4]> var_2166_end_0 = const()[name = string("op_2166_end_0"), val = tensor<int32, [4]>([1, 960, 1, 256])];
            tensor<bool, [4]> var_2166_end_mask_0 = const()[name = string("op_2166_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 256]> var_2166_cast_fp16 = slice_by_index(begin = var_2166_begin_0, end = var_2166_end_0, end_mask = var_2166_end_mask_0, x = q_15_cast_fp16)[name = string("op_2166_cast_fp16")];
            tensor<int32, [4]> var_2170_begin_0 = const()[name = string("op_2170_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_2170_end_0 = const()[name = string("op_2170_end_0"), val = tensor<int32, [4]>([1, 1120, 1, 256])];
            tensor<bool, [4]> var_2170_end_mask_0 = const()[name = string("op_2170_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 256]> var_2170_cast_fp16 = slice_by_index(begin = var_2170_begin_0, end = var_2170_end_0, end_mask = var_2170_end_mask_0, x = q_15_cast_fp16)[name = string("op_2170_cast_fp16")];
            tensor<int32, [4]> var_2174_begin_0 = const()[name = string("op_2174_begin_0"), val = tensor<int32, [4]>([0, 1120, 0, 0])];
            tensor<int32, [4]> var_2174_end_0 = const()[name = string("op_2174_end_0"), val = tensor<int32, [4]>([1, 1, 1, 256])];
            tensor<bool, [4]> var_2174_end_mask_0 = const()[name = string("op_2174_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 160, 1, 256]> var_2174_cast_fp16 = slice_by_index(begin = var_2174_begin_0, end = var_2174_end_0, end_mask = var_2174_end_mask_0, x = q_15_cast_fp16)[name = string("op_2174_cast_fp16")];
            tensor<int32, [4]> k_31_perm_0 = const()[name = string("k_31_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_2181_begin_0 = const()[name = string("op_2181_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2181_end_0 = const()[name = string("op_2181_end_0"), val = tensor<int32, [4]>([1, 77, 1, 160])];
            tensor<bool, [4]> var_2181_end_mask_0 = const()[name = string("op_2181_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 77, 1, 1280]> k_31_cast_fp16 = transpose(perm = k_31_perm_0, x = k_29_cast_fp16)[name = string("transpose_0")];
            tensor<fp16, [1, 77, 1, 160]> var_2181_cast_fp16 = slice_by_index(begin = var_2181_begin_0, end = var_2181_end_0, end_mask = var_2181_end_mask_0, x = k_31_cast_fp16)[name = string("op_2181_cast_fp16")];
            tensor<int32, [4]> var_2185_begin_0 = const()[name = string("op_2185_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 160])];
            tensor<int32, [4]> var_2185_end_0 = const()[name = string("op_2185_end_0"), val = tensor<int32, [4]>([1, 77, 1, 320])];
            tensor<bool, [4]> var_2185_end_mask_0 = const()[name = string("op_2185_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 77, 1, 160]> var_2185_cast_fp16 = slice_by_index(begin = var_2185_begin_0, end = var_2185_end_0, end_mask = var_2185_end_mask_0, x = k_31_cast_fp16)[name = string("op_2185_cast_fp16")];
            tensor<int32, [4]> var_2189_begin_0 = const()[name = string("op_2189_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_2189_end_0 = const()[name = string("op_2189_end_0"), val = tensor<int32, [4]>([1, 77, 1, 480])];
            tensor<bool, [4]> var_2189_end_mask_0 = const()[name = string("op_2189_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 77, 1, 160]> var_2189_cast_fp16 = slice_by_index(begin = var_2189_begin_0, end = var_2189_end_0, end_mask = var_2189_end_mask_0, x = k_31_cast_fp16)[name = string("op_2189_cast_fp16")];
            tensor<int32, [4]> var_2193_begin_0 = const()[name = string("op_2193_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 480])];
            tensor<int32, [4]> var_2193_end_0 = const()[name = string("op_2193_end_0"), val = tensor<int32, [4]>([1, 77, 1, 640])];
            tensor<bool, [4]> var_2193_end_mask_0 = const()[name = string("op_2193_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 77, 1, 160]> var_2193_cast_fp16 = slice_by_index(begin = var_2193_begin_0, end = var_2193_end_0, end_mask = var_2193_end_mask_0, x = k_31_cast_fp16)[name = string("op_2193_cast_fp16")];
            tensor<int32, [4]> var_2197_begin_0 = const()[name = string("op_2197_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_2197_end_0 = const()[name = string("op_2197_end_0"), val = tensor<int32, [4]>([1, 77, 1, 800])];
            tensor<bool, [4]> var_2197_end_mask_0 = const()[name = string("op_2197_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 77, 1, 160]> var_2197_cast_fp16 = slice_by_index(begin = var_2197_begin_0, end = var_2197_end_0, end_mask = var_2197_end_mask_0, x = k_31_cast_fp16)[name = string("op_2197_cast_fp16")];
            tensor<int32, [4]> var_2201_begin_0 = const()[name = string("op_2201_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 800])];
            tensor<int32, [4]> var_2201_end_0 = const()[name = string("op_2201_end_0"), val = tensor<int32, [4]>([1, 77, 1, 960])];
            tensor<bool, [4]> var_2201_end_mask_0 = const()[name = string("op_2201_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 77, 1, 160]> var_2201_cast_fp16 = slice_by_index(begin = var_2201_begin_0, end = var_2201_end_0, end_mask = var_2201_end_mask_0, x = k_31_cast_fp16)[name = string("op_2201_cast_fp16")];
            tensor<int32, [4]> var_2205_begin_0 = const()[name = string("op_2205_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_2205_end_0 = const()[name = string("op_2205_end_0"), val = tensor<int32, [4]>([1, 77, 1, 1120])];
            tensor<bool, [4]> var_2205_end_mask_0 = const()[name = string("op_2205_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 77, 1, 160]> var_2205_cast_fp16 = slice_by_index(begin = var_2205_begin_0, end = var_2205_end_0, end_mask = var_2205_end_mask_0, x = k_31_cast_fp16)[name = string("op_2205_cast_fp16")];
            tensor<int32, [4]> var_2209_begin_0 = const()[name = string("op_2209_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1120])];
            tensor<int32, [4]> var_2209_end_0 = const()[name = string("op_2209_end_0"), val = tensor<int32, [4]>([1, 77, 1, 1])];
            tensor<bool, [4]> var_2209_end_mask_0 = const()[name = string("op_2209_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 77, 1, 160]> var_2209_cast_fp16 = slice_by_index(begin = var_2209_begin_0, end = var_2209_end_0, end_mask = var_2209_end_mask_0, x = k_31_cast_fp16)[name = string("op_2209_cast_fp16")];
            tensor<int32, [4]> var_2211_begin_0 = const()[name = string("op_2211_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2211_end_0 = const()[name = string("op_2211_end_0"), val = tensor<int32, [4]>([1, 160, 1, 77])];
            tensor<bool, [4]> var_2211_end_mask_0 = const()[name = string("op_2211_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 77]> var_2211_cast_fp16 = slice_by_index(begin = var_2211_begin_0, end = var_2211_end_0, end_mask = var_2211_end_mask_0, x = v_15_cast_fp16)[name = string("op_2211_cast_fp16")];
            tensor<int32, [4]> var_2215_begin_0 = const()[name = string("op_2215_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_2215_end_0 = const()[name = string("op_2215_end_0"), val = tensor<int32, [4]>([1, 320, 1, 77])];
            tensor<bool, [4]> var_2215_end_mask_0 = const()[name = string("op_2215_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 77]> var_2215_cast_fp16 = slice_by_index(begin = var_2215_begin_0, end = var_2215_end_0, end_mask = var_2215_end_mask_0, x = v_15_cast_fp16)[name = string("op_2215_cast_fp16")];
            tensor<int32, [4]> var_2219_begin_0 = const()[name = string("op_2219_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_2219_end_0 = const()[name = string("op_2219_end_0"), val = tensor<int32, [4]>([1, 480, 1, 77])];
            tensor<bool, [4]> var_2219_end_mask_0 = const()[name = string("op_2219_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 77]> var_2219_cast_fp16 = slice_by_index(begin = var_2219_begin_0, end = var_2219_end_0, end_mask = var_2219_end_mask_0, x = v_15_cast_fp16)[name = string("op_2219_cast_fp16")];
            tensor<int32, [4]> var_2223_begin_0 = const()[name = string("op_2223_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_2223_end_0 = const()[name = string("op_2223_end_0"), val = tensor<int32, [4]>([1, 640, 1, 77])];
            tensor<bool, [4]> var_2223_end_mask_0 = const()[name = string("op_2223_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 77]> var_2223_cast_fp16 = slice_by_index(begin = var_2223_begin_0, end = var_2223_end_0, end_mask = var_2223_end_mask_0, x = v_15_cast_fp16)[name = string("op_2223_cast_fp16")];
            tensor<int32, [4]> var_2227_begin_0 = const()[name = string("op_2227_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_2227_end_0 = const()[name = string("op_2227_end_0"), val = tensor<int32, [4]>([1, 800, 1, 77])];
            tensor<bool, [4]> var_2227_end_mask_0 = const()[name = string("op_2227_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 77]> var_2227_cast_fp16 = slice_by_index(begin = var_2227_begin_0, end = var_2227_end_0, end_mask = var_2227_end_mask_0, x = v_15_cast_fp16)[name = string("op_2227_cast_fp16")];
            tensor<int32, [4]> var_2231_begin_0 = const()[name = string("op_2231_begin_0"), val = tensor<int32, [4]>([0, 800, 0, 0])];
            tensor<int32, [4]> var_2231_end_0 = const()[name = string("op_2231_end_0"), val = tensor<int32, [4]>([1, 960, 1, 77])];
            tensor<bool, [4]> var_2231_end_mask_0 = const()[name = string("op_2231_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 77]> var_2231_cast_fp16 = slice_by_index(begin = var_2231_begin_0, end = var_2231_end_0, end_mask = var_2231_end_mask_0, x = v_15_cast_fp16)[name = string("op_2231_cast_fp16")];
            tensor<int32, [4]> var_2235_begin_0 = const()[name = string("op_2235_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_2235_end_0 = const()[name = string("op_2235_end_0"), val = tensor<int32, [4]>([1, 1120, 1, 77])];
            tensor<bool, [4]> var_2235_end_mask_0 = const()[name = string("op_2235_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 160, 1, 77]> var_2235_cast_fp16 = slice_by_index(begin = var_2235_begin_0, end = var_2235_end_0, end_mask = var_2235_end_mask_0, x = v_15_cast_fp16)[name = string("op_2235_cast_fp16")];
            tensor<int32, [4]> var_2239_begin_0 = const()[name = string("op_2239_begin_0"), val = tensor<int32, [4]>([0, 1120, 0, 0])];
            tensor<int32, [4]> var_2239_end_0 = const()[name = string("op_2239_end_0"), val = tensor<int32, [4]>([1, 1, 1, 77])];
            tensor<bool, [4]> var_2239_end_mask_0 = const()[name = string("op_2239_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 160, 1, 77]> var_2239_cast_fp16 = slice_by_index(begin = var_2239_begin_0, end = var_2239_end_0, end_mask = var_2239_end_mask_0, x = v_15_cast_fp16)[name = string("op_2239_cast_fp16")];
            string var_2243_equation_0 = const()[name = string("op_2243_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 77, 1, 256]> var_2243_cast_fp16 = einsum(equation = var_2243_equation_0, values = (var_2181_cast_fp16, var_2146_cast_fp16))[name = string("op_2243_cast_fp16")];
            fp16 var_2244_to_fp16 = const()[name = string("op_2244_to_fp16"), val = fp16(0x1.43cp-4)];
            tensor<fp16, [1, 77, 1, 256]> aw_113_cast_fp16 = mul(x = var_2243_cast_fp16, y = var_2244_to_fp16)[name = string("aw_113_cast_fp16")];
            string var_2247_equation_0 = const()[name = string("op_2247_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 77, 1, 256]> var_2247_cast_fp16 = einsum(equation = var_2247_equation_0, values = (var_2185_cast_fp16, var_2150_cast_fp16))[name = string("op_2247_cast_fp16")];
            fp16 var_2248_to_fp16 = const()[name = string("op_2248_to_fp16"), val = fp16(0x1.43cp-4)];
            tensor<fp16, [1, 77, 1, 256]> aw_115_cast_fp16 = mul(x = var_2247_cast_fp16, y = var_2248_to_fp16)[name = string("aw_115_cast_fp16")];
            string var_2251_equation_0 = const()[name = string("op_2251_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 77, 1, 256]> var_2251_cast_fp16 = einsum(equation = var_2251_equation_0, values = (var_2189_cast_fp16, var_2154_cast_fp16))[name = string("op_2251_cast_fp16")];
            fp16 var_2252_to_fp16 = const()[name = string("op_2252_to_fp16"), val = fp16(0x1.43cp-4)];
            tensor<fp16, [1, 77, 1, 256]> aw_117_cast_fp16 = mul(x = var_2251_cast_fp16, y = var_2252_to_fp16)[name = string("aw_117_cast_fp16")];
            string var_2255_equation_0 = const()[name = string("op_2255_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 77, 1, 256]> var_2255_cast_fp16 = einsum(equation = var_2255_equation_0, values = (var_2193_cast_fp16, var_2158_cast_fp16))[name = string("op_2255_cast_fp16")];
            fp16 var_2256_to_fp16 = const()[name = string("op_2256_to_fp16"), val = fp16(0x1.43cp-4)];
            tensor<fp16, [1, 77, 1, 256]> aw_119_cast_fp16 = mul(x = var_2255_cast_fp16, y = var_2256_to_fp16)[name = string("aw_119_cast_fp16")];
            string var_2259_equation_0 = const()[name = string("op_2259_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 77, 1, 256]> var_2259_cast_fp16 = einsum(equation = var_2259_equation_0, values = (var_2197_cast_fp16, var_2162_cast_fp16))[name = string("op_2259_cast_fp16")];
            fp16 var_2260_to_fp16 = const()[name = string("op_2260_to_fp16"), val = fp16(0x1.43cp-4)];
            tensor<fp16, [1, 77, 1, 256]> aw_121_cast_fp16 = mul(x = var_2259_cast_fp16, y = var_2260_to_fp16)[name = string("aw_121_cast_fp16")];
            string var_2263_equation_0 = const()[name = string("op_2263_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 77, 1, 256]> var_2263_cast_fp16 = einsum(equation = var_2263_equation_0, values = (var_2201_cast_fp16, var_2166_cast_fp16))[name = string("op_2263_cast_fp16")];
            fp16 var_2264_to_fp16 = const()[name = string("op_2264_to_fp16"), val = fp16(0x1.43cp-4)];
            tensor<fp16, [1, 77, 1, 256]> aw_123_cast_fp16 = mul(x = var_2263_cast_fp16, y = var_2264_to_fp16)[name = string("aw_123_cast_fp16")];
            string var_2267_equation_0 = const()[name = string("op_2267_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 77, 1, 256]> var_2267_cast_fp16 = einsum(equation = var_2267_equation_0, values = (var_2205_cast_fp16, var_2170_cast_fp16))[name = string("op_2267_cast_fp16")];
            fp16 var_2268_to_fp16 = const()[name = string("op_2268_to_fp16"), val = fp16(0x1.43cp-4)];
            tensor<fp16, [1, 77, 1, 256]> aw_125_cast_fp16 = mul(x = var_2267_cast_fp16, y = var_2268_to_fp16)[name = string("aw_125_cast_fp16")];
            string var_2271_equation_0 = const()[name = string("op_2271_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 77, 1, 256]> var_2271_cast_fp16 = einsum(equation = var_2271_equation_0, values = (var_2209_cast_fp16, var_2174_cast_fp16))[name = string("op_2271_cast_fp16")];
            fp16 var_2272_to_fp16 = const()[name = string("op_2272_to_fp16"), val = fp16(0x1.43cp-4)];
            tensor<fp16, [1, 77, 1, 256]> aw_127_cast_fp16 = mul(x = var_2271_cast_fp16, y = var_2272_to_fp16)[name = string("aw_127_cast_fp16")];
            tensor<fp16, [1, 77, 1, 256]> var_2274_cast_fp16 = softmax(axis = var_1812, x = aw_113_cast_fp16)[name = string("op_2274_cast_fp16")];
            tensor<fp16, [1, 77, 1, 256]> var_2275_cast_fp16 = softmax(axis = var_1812, x = aw_115_cast_fp16)[name = string("op_2275_cast_fp16")];
            tensor<fp16, [1, 77, 1, 256]> var_2276_cast_fp16 = softmax(axis = var_1812, x = aw_117_cast_fp16)[name = string("op_2276_cast_fp16")];
            tensor<fp16, [1, 77, 1, 256]> var_2277_cast_fp16 = softmax(axis = var_1812, x = aw_119_cast_fp16)[name = string("op_2277_cast_fp16")];
            tensor<fp16, [1, 77, 1, 256]> var_2278_cast_fp16 = softmax(axis = var_1812, x = aw_121_cast_fp16)[name = string("op_2278_cast_fp16")];
            tensor<fp16, [1, 77, 1, 256]> var_2279_cast_fp16 = softmax(axis = var_1812, x = aw_123_cast_fp16)[name = string("op_2279_cast_fp16")];
            tensor<fp16, [1, 77, 1, 256]> var_2280_cast_fp16 = softmax(axis = var_1812, x = aw_125_cast_fp16)[name = string("op_2280_cast_fp16")];
            tensor<fp16, [1, 77, 1, 256]> var_2281_cast_fp16 = softmax(axis = var_1812, x = aw_127_cast_fp16)[name = string("op_2281_cast_fp16")];
            string var_2283_equation_0 = const()[name = string("op_2283_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 160, 1, 256]> var_2283_cast_fp16 = einsum(equation = var_2283_equation_0, values = (var_2211_cast_fp16, var_2274_cast_fp16))[name = string("op_2283_cast_fp16")];
            string var_2285_equation_0 = const()[name = string("op_2285_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 160, 1, 256]> var_2285_cast_fp16 = einsum(equation = var_2285_equation_0, values = (var_2215_cast_fp16, var_2275_cast_fp16))[name = string("op_2285_cast_fp16")];
            string var_2287_equation_0 = const()[name = string("op_2287_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 160, 1, 256]> var_2287_cast_fp16 = einsum(equation = var_2287_equation_0, values = (var_2219_cast_fp16, var_2276_cast_fp16))[name = string("op_2287_cast_fp16")];
            string var_2289_equation_0 = const()[name = string("op_2289_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 160, 1, 256]> var_2289_cast_fp16 = einsum(equation = var_2289_equation_0, values = (var_2223_cast_fp16, var_2277_cast_fp16))[name = string("op_2289_cast_fp16")];
            string var_2291_equation_0 = const()[name = string("op_2291_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 160, 1, 256]> var_2291_cast_fp16 = einsum(equation = var_2291_equation_0, values = (var_2227_cast_fp16, var_2278_cast_fp16))[name = string("op_2291_cast_fp16")];
            string var_2293_equation_0 = const()[name = string("op_2293_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 160, 1, 256]> var_2293_cast_fp16 = einsum(equation = var_2293_equation_0, values = (var_2231_cast_fp16, var_2279_cast_fp16))[name = string("op_2293_cast_fp16")];
            string var_2295_equation_0 = const()[name = string("op_2295_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 160, 1, 256]> var_2295_cast_fp16 = einsum(equation = var_2295_equation_0, values = (var_2235_cast_fp16, var_2280_cast_fp16))[name = string("op_2295_cast_fp16")];
            string var_2297_equation_0 = const()[name = string("op_2297_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 160, 1, 256]> var_2297_cast_fp16 = einsum(equation = var_2297_equation_0, values = (var_2239_cast_fp16, var_2281_cast_fp16))[name = string("op_2297_cast_fp16")];
            bool input_109_interleave_0 = const()[name = string("input_109_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 1280, 1, 256]> input_109_cast_fp16 = concat(axis = var_1812, interleave = input_109_interleave_0, values = (var_2283_cast_fp16, var_2285_cast_fp16, var_2287_cast_fp16, var_2289_cast_fp16, var_2291_cast_fp16, var_2293_cast_fp16, var_2295_cast_fp16, var_2297_cast_fp16))[name = string("input_109_cast_fp16")];
            string var_2307_pad_type_0 = const()[name = string("op_2307_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_2307_strides_0 = const()[name = string("op_2307_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_2307_pad_0 = const()[name = string("op_2307_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_2307_dilations_0 = const()[name = string("op_2307_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_2307_groups_0 = const()[name = string("op_2307_groups_0"), val = int32(1)];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = string("up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(295131840)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = string("up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(298408704)))];
            tensor<fp16, [1, 1280, 1, 256]> var_2307_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_2307_dilations_0, groups = var_2307_groups_0, pad = var_2307_pad_0, pad_type = var_2307_pad_type_0, strides = var_2307_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_109_cast_fp16)[name = string("op_2307_cast_fp16")];
            tensor<fp16, [1, 1280, 1, 256]> inputs_23_cast_fp16_1 = add(x = var_2307_cast_fp16, y = inputs_21_cast_fp16)[name = string("inputs_23_cast_fp16")];
            tensor<int32, [1]> input_111_axes_0 = const()[name = string("input_111_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1280]> input_111_gamma_0_to_fp16 = const()[name = string("input_111_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(298411328)))];
            tensor<fp16, [1280]> input_111_beta_0_to_fp16 = const()[name = string("input_111_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(298413952)))];
            fp16 var_2317_to_fp16 = const()[name = string("op_2317_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1280, 1, 256]> input_111_cast_fp16 = layer_norm(axes = input_111_axes_0, beta = input_111_beta_0_to_fp16, epsilon = var_2317_to_fp16, gamma = input_111_gamma_0_to_fp16, x = inputs_23_cast_fp16_1)[name = string("input_111_cast_fp16")];
            string var_2337_pad_type_0 = const()[name = string("op_2337_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_2337_strides_0 = const()[name = string("op_2337_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_2337_pad_0 = const()[name = string("op_2337_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_2337_dilations_0 = const()[name = string("op_2337_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_2337_groups_0 = const()[name = string("op_2337_groups_0"), val = int32(1)];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = string("up_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(298416576)))];
            tensor<fp16, [10240]> up_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = string("up_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(324631040)))];
            tensor<fp16, [1, 10240, 1, 256]> var_2337_cast_fp16_1 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_2337_dilations_0, groups = var_2337_groups_0, pad = var_2337_pad_0, pad_type = var_2337_pad_type_0, strides = var_2337_strides_0, weight = up_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_111_cast_fp16)[name = string("op_2337_cast_fp16")];
            string input_7_cast_fp16_dtype_0 = const()[name = string("input_7_cast_fp16_dtype_0"), val = string("fp32")];
            string input_63_cast_fp16_dtype_0 = const()[name = string("input_63_cast_fp16_dtype_0"), val = string("fp32")];
            string var_2337_cast_fp16_dtype_0 = const()[name = string("var_2337_cast_fp16_dtype_0"), val = string("fp32")];
            string input_65_cast_fp16_dtype_0 = const()[name = string("input_65_cast_fp16_dtype_0"), val = string("fp32")];
            string input_37_cast_fp16_dtype_0 = const()[name = string("input_37_cast_fp16_dtype_0"), val = string("fp32")];
            string hidden_states_61_cast_fp16_dtype_0 = const()[name = string("hidden_states_61_cast_fp16_dtype_0"), val = string("fp32")];
            string input_35_cast_fp16_dtype_0 = const()[name = string("input_35_cast_fp16_dtype_0"), val = string("fp32")];
            string input_15_cast_fp16_dtype_0 = const()[name = string("input_15_cast_fp16_dtype_0"), val = string("fp32")];
            string inputs_23_cast_fp16_dtype_0 = const()[name = string("inputs_23_cast_fp16_dtype_0"), val = string("fp32")];
            tensor<fp32, [1, 1280, 1, 256]> inputs_23_cast_fp16 = cast(dtype = inputs_23_cast_fp16_dtype_0, x = inputs_23_cast_fp16_1)[name = string("cast_9")];
            tensor<fp32, [1, 1280, 1, 1]> input_15_cast_fp16 = cast(dtype = input_15_cast_fp16_dtype_0, x = input_15_cast_fp16_1)[name = string("cast_10")];
            tensor<fp32, [1, 320, 64, 64]> input_35_cast_fp16 = cast(dtype = input_35_cast_fp16_dtype_0, x = input_35_cast_fp16_1)[name = string("cast_11")];
            tensor<fp32, [1, 1280, 16, 16]> hidden_states_61_cast_fp16 = cast(dtype = hidden_states_61_cast_fp16_dtype_0, x = hidden_states_61_cast_fp16_1)[name = string("cast_12")];
            tensor<fp32, [1, 320, 32, 32]> input_37_cast_fp16 = cast(dtype = input_37_cast_fp16_dtype_0, x = input_37_cast_fp16_1)[name = string("cast_13")];
            tensor<fp32, [1, 640, 16, 16]> input_65_cast_fp16 = cast(dtype = input_65_cast_fp16_dtype_0, x = input_65_cast_fp16_1)[name = string("cast_14")];
            tensor<fp32, [1, 10240, 1, 256]> var_2337_cast_fp16 = cast(dtype = var_2337_cast_fp16_dtype_0, x = var_2337_cast_fp16_1)[name = string("cast_15")];
            tensor<fp32, [1, 640, 32, 32]> input_63_cast_fp16 = cast(dtype = input_63_cast_fp16_dtype_0, x = input_63_cast_fp16_1)[name = string("cast_16")];
            tensor<fp32, [1, 320, 64, 64]> input_7_cast_fp16 = cast(dtype = input_7_cast_fp16_dtype_0, x = input_7_cast_fp16_1)[name = string("cast_17")];
        } -> (input_7_cast_fp16, input_63_cast_fp16, var_2337_cast_fp16, input_65_cast_fp16, input_37_cast_fp16, hidden_states_61_cast_fp16, input_35_cast_fp16, input_15_cast_fp16, inputs_23_cast_fp16);
}