program(1.0)
[buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "3510.2.1"}, {"coremlc-version", "3500.32.1"}, {"coremltools-component-torch", "2.7.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "9.0"}})]
{
    func main<ios16>(tensor<int32, [1, ?]> input_ids) [FlexibleShapeInformation = tuple<tuple<tensor<string, []>, dict<tensor<string, []>, tensor<int32, [?]>>>, tuple<tensor<string, []>, dict<tensor<string, []>, list<tensor<int32, [2]>, ?>>>>((("DefaultShapes", {{"input_ids", [1, 1]}}), ("RangeDims", {{"input_ids", [[1, 1], [1, 64]]}})))] {
            tensor<int32, []> var_7 = const()[name = tensor<string, []>("op_7"), val = tensor<int32, []>(128)];
            tensor<int32, []> var_14 = const()[name = tensor<string, []>("op_14"), val = tensor<int32, []>(0)];
            tensor<int32, []> var_16 = const()[name = tensor<string, []>("op_16"), val = tensor<int32, []>(-1)];
            tensor<int32, []> var_17 = const()[name = tensor<string, []>("op_17"), val = tensor<int32, []>(1)];
            tensor<int32, [2]> var_23_shape = shape(x = input_ids)[name = tensor<string, []>("op_23_shape")];
            tensor<int32, []> gather_0_indices_0 = const()[name = tensor<string, []>("gather_0_indices_0"), val = tensor<int32, []>(1)];
            tensor<int32, []> gather_0_axis_0 = const()[name = tensor<string, []>("gather_0_axis_0"), val = tensor<int32, []>(0)];
            tensor<int32, []> gather_0_batch_dims_0 = const()[name = tensor<string, []>("gather_0_batch_dims_0"), val = tensor<int32, []>(0)];
            tensor<int32, []> gather_0 = gather(axis = gather_0_axis_0, batch_dims = gather_0_batch_dims_0, indices = gather_0_indices_0, x = var_23_shape)[name = tensor<string, []>("gather_0")];
            tensor<int32, []> var_27_axis_0 = const()[name = tensor<string, []>("op_27_axis_0"), val = tensor<int32, []>(0)];
            tensor<int32, []> var_27_batch_dims_0 = const()[name = tensor<string, []>("op_27_batch_dims_0"), val = tensor<int32, []>(0)];
            tensor<fp16, [63, 128]> encoder_embed_tokens_weight_to_fp16 = const()[name = tensor<string, []>("encoder_embed_tokens_weight_to_fp16"), val = tensor<fp16, [63, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))];
            tensor<fp16, [1, ?, 128]> var_27_cast_fp16 = gather(axis = var_27_axis_0, batch_dims = var_27_batch_dims_0, indices = input_ids, x = encoder_embed_tokens_weight_to_fp16)[name = tensor<string, []>("op_27_cast_fp16")];
            tensor<int32, []> const_0 = const()[name = tensor<string, []>("const_0"), val = tensor<int32, []>(1)];
            tensor<int32, [?]> var_37 = range_1d(end = gather_0, start = var_14, step = const_0)[name = tensor<string, []>("op_37")];
            tensor<int32, [2]> concat_1 = const()[name = tensor<string, []>("concat_1"), val = tensor<int32, [2]>([1, -1])];
            tensor<int32, [1]> expand_dims_0_axes_0 = const()[name = tensor<string, []>("expand_dims_0_axes_0"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1, ?]> expand_dims_0 = expand_dims(axes = expand_dims_0_axes_0, x = var_37)[name = tensor<string, []>("expand_dims_0")];
            tensor<int32, [2]> shape_0 = shape(x = expand_dims_0)[name = tensor<string, []>("shape_0")];
            tensor<bool, [2]> equal_0 = const()[name = tensor<string, []>("equal_0"), val = tensor<bool, [2]>([false, true])];
            tensor<int32, [2]> select_0 = select(a = shape_0, b = concat_1, cond = equal_0)[name = tensor<string, []>("select_0")];
            tensor<int32, [2]> real_div_0 = real_div(x = select_0, y = shape_0)[name = tensor<string, []>("real_div_0")];
            tensor<int32, [?, ?]> positions = tile(reps = real_div_0, x = expand_dims_0)[name = tensor<string, []>("positions")];
            tensor<int32, []> var_40 = const()[name = tensor<string, []>("op_40"), val = tensor<int32, []>(2)];
            tensor<int32, [?, ?]> input_3 = add(x = positions, y = var_40)[name = tensor<string, []>("input_3")];
            tensor<int32, []> embed_pos_1_axis_0 = const()[name = tensor<string, []>("embed_pos_1_axis_0"), val = tensor<int32, []>(0)];
            tensor<int32, []> embed_pos_1_batch_dims_0 = const()[name = tensor<string, []>("embed_pos_1_batch_dims_0"), val = tensor<int32, []>(0)];
            tensor<fp16, [66, 128]> encoder_embed_positions_weight_to_fp16 = const()[name = tensor<string, []>("encoder_embed_positions_weight_to_fp16"), val = tensor<fp16, [66, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16256)))];
            tensor<fp16, [?, ?, 128]> embed_pos_1_cast_fp16 = gather(axis = embed_pos_1_axis_0, batch_dims = embed_pos_1_batch_dims_0, indices = input_3, x = encoder_embed_positions_weight_to_fp16)[name = tensor<string, []>("embed_pos_1_cast_fp16")];
            tensor<fp16, [?, ?, 128]> input_5_cast_fp16 = add(x = var_27_cast_fp16, y = embed_pos_1_cast_fp16)[name = tensor<string, []>("input_5_cast_fp16")];
            tensor<int32, [1]> input_7_axes_0 = const()[name = tensor<string, []>("input_7_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [128]> encoder_layernorm_embedding_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layernorm_embedding_weight_to_fp16"), val = tensor<fp16, [128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(33216)))];
            tensor<fp16, [128]> encoder_layernorm_embedding_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layernorm_embedding_bias_to_fp16"), val = tensor<fp16, [128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(33536)))];
            tensor<fp16, []> var_8_to_fp16 = const()[name = tensor<string, []>("op_8_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [?, ?, 128]> input_7_cast_fp16 = layer_norm(axes = input_7_axes_0, beta = encoder_layernorm_embedding_bias_to_fp16, epsilon = var_8_to_fp16, gamma = encoder_layernorm_embedding_weight_to_fp16, x = input_5_cast_fp16)[name = tensor<string, []>("input_7_cast_fp16")];
            tensor<int32, [3]> var_59_shape_cast_fp16 = shape(x = input_7_cast_fp16)[name = tensor<string, []>("op_59_shape_cast_fp16")];
            tensor<int32, []> gather_3_indices_0 = const()[name = tensor<string, []>("gather_3_indices_0"), val = tensor<int32, []>(0)];
            tensor<int32, []> gather_3_axis_0 = const()[name = tensor<string, []>("gather_3_axis_0"), val = tensor<int32, []>(0)];
            tensor<int32, []> gather_3_batch_dims_0 = const()[name = tensor<string, []>("gather_3_batch_dims_0"), val = tensor<int32, []>(0)];
            tensor<int32, []> gather_3 = gather(axis = gather_3_axis_0, batch_dims = gather_3_batch_dims_0, indices = gather_3_indices_0, x = var_59_shape_cast_fp16)[name = tensor<string, []>("gather_3")];
            tensor<int32, []> gather_4_indices_0 = const()[name = tensor<string, []>("gather_4_indices_0"), val = tensor<int32, []>(1)];
            tensor<int32, []> gather_4_axis_0 = const()[name = tensor<string, []>("gather_4_axis_0"), val = tensor<int32, []>(0)];
            tensor<int32, []> gather_4_batch_dims_0 = const()[name = tensor<string, []>("gather_4_batch_dims_0"), val = tensor<int32, []>(0)];
            tensor<int32, []> gather_4 = gather(axis = gather_4_axis_0, batch_dims = gather_4_batch_dims_0, indices = gather_4_indices_0, x = var_59_shape_cast_fp16)[name = tensor<string, []>("gather_4")];
            tensor<fp16, [128, 128]> encoder_layers_0_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [128, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(33856)))];
            tensor<fp16, [128]> encoder_layers_0_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66688)))];
            tensor<fp16, [?, ?, 128]> linear_0_cast_fp16 = linear(bias = encoder_layers_0_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_0_self_attn_q_proj_weight_to_fp16, x = input_7_cast_fp16)[name = tensor<string, []>("linear_0_cast_fp16")];
            tensor<fp16, [128, 128]> encoder_layers_0_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [128, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67008)))];
            tensor<fp16, [128]> encoder_layers_0_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99840)))];
            tensor<fp16, [?, ?, 128]> linear_1_cast_fp16 = linear(bias = encoder_layers_0_self_attn_k_proj_bias_to_fp16, weight = encoder_layers_0_self_attn_k_proj_weight_to_fp16, x = input_7_cast_fp16)[name = tensor<string, []>("linear_1_cast_fp16")];
            tensor<int32, []> concat_2_axis_0 = const()[name = tensor<string, []>("concat_2_axis_0"), val = tensor<int32, []>(0)];
            tensor<bool, []> concat_2_interleave_0 = const()[name = tensor<string, []>("concat_2_interleave_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> concat_2 = concat(axis = concat_2_axis_0, interleave = concat_2_interleave_0, values = (gather_3, var_16, var_17, var_7))[name = tensor<string, []>("concat_2")];
            tensor<fp16, [?, ?, 1, 128]> var_68_cast_fp16 = reshape(shape = concat_2, x = linear_1_cast_fp16)[name = tensor<string, []>("op_68_cast_fp16")];
            tensor<fp16, [128, 128]> encoder_layers_0_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [128, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(100160)))];
            tensor<fp16, [128]> encoder_layers_0_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132992)))];
            tensor<fp16, [?, ?, 128]> linear_2_cast_fp16 = linear(bias = encoder_layers_0_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_0_self_attn_v_proj_weight_to_fp16, x = input_7_cast_fp16)[name = tensor<string, []>("linear_2_cast_fp16")];
            tensor<fp16, [?, ?, 1, 128]> var_75_cast_fp16 = reshape(shape = concat_2, x = linear_2_cast_fp16)[name = tensor<string, []>("op_75_cast_fp16")];
            tensor<int32, [4]> var_76_perm_0 = const()[name = tensor<string, []>("op_76_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, []> concat_4_axis_0 = const()[name = tensor<string, []>("concat_4_axis_0"), val = tensor<int32, []>(0)];
            tensor<bool, []> concat_4_interleave_0 = const()[name = tensor<string, []>("concat_4_interleave_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> concat_4 = concat(axis = concat_4_axis_0, interleave = concat_4_interleave_0, values = (gather_3, gather_4, var_17, var_7))[name = tensor<string, []>("concat_4")];
            tensor<fp16, [?, ?, 1, 128]> var_79_cast_fp16 = reshape(shape = concat_4, x = linear_0_cast_fp16)[name = tensor<string, []>("op_79_cast_fp16")];
            tensor<fp16, []> mul_0_y_0_to_fp16 = const()[name = tensor<string, []>("mul_0_y_0_to_fp16"), val = tensor<fp16, []>(0x1.6ap-4)];
            tensor<fp16, [?, ?, 1, 128]> mul_0_cast_fp16 = mul(x = var_79_cast_fp16, y = mul_0_y_0_to_fp16)[name = tensor<string, []>("mul_0_cast_fp16")];
            tensor<bool, []> matmul_0_transpose_y_0 = const()[name = tensor<string, []>("matmul_0_transpose_y_0"), val = tensor<bool, []>(true)];
            tensor<bool, []> matmul_0_transpose_x_0 = const()[name = tensor<string, []>("matmul_0_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_4_perm_0 = const()[name = tensor<string, []>("transpose_4_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_5_perm_0 = const()[name = tensor<string, []>("transpose_5_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [?, 1, ?, 128]> transpose_5 = transpose(perm = transpose_5_perm_0, x = var_68_cast_fp16)[name = tensor<string, []>("transpose_7")];
            tensor<fp16, [?, 1, ?, 128]> transpose_4 = transpose(perm = transpose_4_perm_0, x = mul_0_cast_fp16)[name = tensor<string, []>("transpose_8")];
            tensor<fp16, [?, 1, ?, ?]> matmul_0_cast_fp16 = matmul(transpose_x = matmul_0_transpose_x_0, transpose_y = matmul_0_transpose_y_0, x = transpose_4, y = transpose_5)[name = tensor<string, []>("matmul_0_cast_fp16")];
            tensor<int32, []> softmax_0_axis_0 = const()[name = tensor<string, []>("softmax_0_axis_0"), val = tensor<int32, []>(-1)];
            tensor<fp16, [?, 1, ?, ?]> softmax_0_cast_fp16 = softmax(axis = softmax_0_axis_0, x = matmul_0_cast_fp16)[name = tensor<string, []>("softmax_0_cast_fp16")];
            tensor<bool, []> attn_output_1_transpose_x_0 = const()[name = tensor<string, []>("attn_output_1_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_1_transpose_y_0 = const()[name = tensor<string, []>("attn_output_1_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [?, 1, ?, 128]> var_76_cast_fp16 = transpose(perm = var_76_perm_0, x = var_75_cast_fp16)[name = tensor<string, []>("transpose_9")];
            tensor<fp16, [?, 1, ?, 128]> attn_output_1_cast_fp16 = matmul(transpose_x = attn_output_1_transpose_x_0, transpose_y = attn_output_1_transpose_y_0, x = softmax_0_cast_fp16, y = var_76_cast_fp16)[name = tensor<string, []>("attn_output_1_cast_fp16")];
            tensor<int32, [4]> attn_output_perm_0 = const()[name = tensor<string, []>("attn_output_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, []> concat_5_axis_0 = const()[name = tensor<string, []>("concat_5_axis_0"), val = tensor<int32, []>(0)];
            tensor<bool, []> concat_5_interleave_0 = const()[name = tensor<string, []>("concat_5_interleave_0"), val = tensor<bool, []>(false)];
            tensor<int32, [3]> concat_5 = concat(axis = concat_5_axis_0, interleave = concat_5_interleave_0, values = (gather_3, gather_4, var_7))[name = tensor<string, []>("concat_5")];
            tensor<fp16, [?, ?, 1, 128]> attn_output_cast_fp16 = transpose(perm = attn_output_perm_0, x = attn_output_1_cast_fp16)[name = tensor<string, []>("transpose_6")];
            tensor<fp16, [?, ?, 128]> input_9_cast_fp16 = reshape(shape = concat_5, x = attn_output_cast_fp16)[name = tensor<string, []>("input_9_cast_fp16")];
            tensor<fp16, [128, 128]> encoder_layers_0_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [128, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(133312)))];
            tensor<fp16, [128]> encoder_layers_0_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(166144)))];
            tensor<fp16, [?, ?, 128]> linear_3_cast_fp16 = linear(bias = encoder_layers_0_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_0_self_attn_out_proj_weight_to_fp16, x = input_9_cast_fp16)[name = tensor<string, []>("linear_3_cast_fp16")];
            tensor<fp16, [?, ?, 128]> input_13_cast_fp16 = add(x = input_7_cast_fp16, y = linear_3_cast_fp16)[name = tensor<string, []>("input_13_cast_fp16")];
            tensor<int32, [1]> input_15_axes_0 = const()[name = tensor<string, []>("input_15_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [128]> encoder_layers_0_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(166464)))];
            tensor<fp16, [128]> encoder_layers_0_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(166784)))];
            tensor<fp16, [?, ?, 128]> input_15_cast_fp16 = layer_norm(axes = input_15_axes_0, beta = encoder_layers_0_self_attn_layer_norm_bias_to_fp16, epsilon = var_8_to_fp16, gamma = encoder_layers_0_self_attn_layer_norm_weight_to_fp16, x = input_13_cast_fp16)[name = tensor<string, []>("input_15_cast_fp16")];
            tensor<fp16, [1024, 128]> encoder_layers_0_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_fc1_weight_to_fp16"), val = tensor<fp16, [1024, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(167104)))];
            tensor<fp16, [1024]> encoder_layers_0_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_fc1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(429312)))];
            tensor<fp16, [?, ?, 1024]> linear_4_cast_fp16 = linear(bias = encoder_layers_0_fc1_bias_to_fp16, weight = encoder_layers_0_fc1_weight_to_fp16, x = input_15_cast_fp16)[name = tensor<string, []>("linear_4_cast_fp16")];
            tensor<string, []> input_19_mode_0 = const()[name = tensor<string, []>("input_19_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [?, ?, 1024]> input_19_cast_fp16 = gelu(mode = input_19_mode_0, x = linear_4_cast_fp16)[name = tensor<string, []>("input_19_cast_fp16")];
            tensor<fp16, [128, 1024]> encoder_layers_0_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_fc2_weight_to_fp16"), val = tensor<fp16, [128, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(431424)))];
            tensor<fp16, [128]> encoder_layers_0_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_fc2_bias_to_fp16"), val = tensor<fp16, [128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(693632)))];
            tensor<fp16, [?, ?, 128]> linear_5_cast_fp16 = linear(bias = encoder_layers_0_fc2_bias_to_fp16, weight = encoder_layers_0_fc2_weight_to_fp16, x = input_19_cast_fp16)[name = tensor<string, []>("linear_5_cast_fp16")];
            tensor<fp16, [?, ?, 128]> input_cast_fp16 = add(x = input_15_cast_fp16, y = linear_5_cast_fp16)[name = tensor<string, []>("input_cast_fp16")];
            tensor<int32, [1]> var_108_axes_0 = const()[name = tensor<string, []>("op_108_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [128]> encoder_layers_0_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(693952)))];
            tensor<fp16, [128]> encoder_layers_0_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(694272)))];
            tensor<fp16, [?, ?, 128]> encoder_hidden_states = layer_norm(axes = var_108_axes_0, beta = encoder_layers_0_final_layer_norm_bias_to_fp16, epsilon = var_8_to_fp16, gamma = encoder_layers_0_final_layer_norm_weight_to_fp16, x = input_cast_fp16)[name = tensor<string, []>("op_108_cast_fp16")];
        } -> (encoder_hidden_states);
}