func.func @serving_default(%arg0: tensor<1x4xf32> {tf_saved_model.index_path = ["x"]}) -> (tensor<1x3xf32> {tf_saved_model.index_path = ["output"]}) attributes {tf.entry_function = {control_outputs = "", inputs = "serving_default_x:0", outputs = "PartitionedCall:0"}, tf_saved_model.exported_names = ["serving_default"]} {

    return %2 : tensor<1x3x2x2xf32>
  }

  func.func @serving_default(%arg0: tensor<1x3x4x3xf32>) -> (tensor<1x3x2x2xf32>) {
    %0 = "tf.PartitionedCall"(%arg0) {config = "", config_proto = "", executor_type = "", f = @conv2d_with_inliner}
        : (tensor<1x3x4x3xf32>) -> tensor<1x3x2x2xf32>
    return %0 : tensor<1x3x2x2xf32>
  }

// CHECK-LABEL: func @serving_default
// CHECK: "tf.PartitionedCall"

# ==============================================================================

# LINT.IfChange(savedmodel_to_stablehlo)
def savedmodel_to_stablehlo(
    input_path: str,
    exported_model_signatures: list[str] = ["serving_default"],
    tag_names: list[str] = ["serve"],
    input_arg_shapes_str: str = "",
) -> bytes: ...
# LINT.ThenChange()

# LINT.IfChange(tensorflow_module_to_stablehlo)
def tensorflow_module_to_stablehlo(

// CHECK-NEXT:      tensor_index: 5
// CHECK-NEXT:    }, {
// CHECK-NEXT:      name: "start_logits",
// CHECK-NEXT:      tensor_index: 6
// CHECK-NEXT:    } ],
// CHECK-NEXT:    signature_key: "serving_default"
// CHECK-NEXT:  } ]
// CHECK-NEXT:}
module attributes {tf.versions = {bad_consumers = [], min_consumer = 12 : i32, producer = 554 : i32}, tf_saved_model.semantics} {

        exported_model_signatures=["serving_default"],
        tag_names=["serve"],
        input_arg_shapes_str="1,28,28,3::32"
)

```

#### Arguments:

* `input_path` (required): Path to your SavedModel directory.
* `exported_model_signatures` (optional): List of signature names to convert.
                                          Defaults to ["serving_default"].

#include "mlir/IR/BuiltinOps.h"  // from @llvm-project

namespace mlir::quant {

// Returns a public `func::FuncOp` in `module_op` whose name matches either
// `main` or `serving_default`. If `func::FuncOps` with both names exist, the
// function with name "main" takes precedence. Returns null if no such a
// function exists.
func::FuncOp FindMainFuncOp(ModuleOp module_op);

}  // namespace mlir::quant

    tf.quantization.experimental.TfRecordRepresentativeDatasetSaver(
          path_map={'serving_default': '/tmp/representative_dataset_path'}
      ).save({'serving_default': representative_dataset})
  )

  # Using in QuantizationOptions.
  quantization_options = tf.quantization.experimental.QuantizationOptions(
      signature_keys=['serving_default'],
      representative_datasets=dataset_file_map,
  )

  TF_ASSERT_OK(MapFunctionSignaturesFromTFSavedModelMLIR(
      module.get(), [&](const TFRTSavedModelSignatureInfo& sig_info) {
        // Only check the signature of "serving_default".
        if (sig_info.func_name != "serving_default") return;

        transform(sig_info.input_names, std::back_inserter(inputs),
                  [](llvm::StringRef x) { return x.str(); });
        in_specs.assign(sig_info.input_specs.begin(),

            preset_method=quant_opts_pb2.QuantizationMethod.PresetMethod.METHOD_STATIC_RANGE_INT8
        ),
        tags={tag_constants.SERVING},
        signature_keys=['serving_default'],
    )

    model = quantize_model.quantize(
        temp_path,
        quantization_options=quantization_options,
        representative_dataset=data_gen(),
    )
    return model

      self, output_saved_model_path: str
  ) -> str:
    """Extracts the first XlaCallModule op from output saved model to string."""
    root = load.load(output_saved_model_path)
    tf_graph_def = root.signatures['serving_default'].graph.as_graph_def()
    for function in tf_graph_def.library.function:
      for node_def in function.node_def:
        if node_def.op == 'XlaCallModule':
          with ir.Context() as context: