#include "tensorflow/compiler/mlir/lite/schema/schema_generated.h"

namespace mlir {
namespace lite {

// Sparsify the `input_model` and write the result to a flatbuffer `builder`.
absl::Status SparsifyModel(const tflite::ModelT& input_model,
                           flatbuffers::FlatBufferBuilder* builder);
}  // namespace lite
}  // namespace mlir

  metadata_t->buffer = input_model.buffers.size() - 1;
  input_model.metadata.push_back(std::move(metadata_t));

  // Sparsify and create output model
  flatbuffers::FlatBufferBuilder output_builder;
  ASSERT_TRUE(SparsifyModel(input_model, &output_builder).ok());
  auto output_fbm = tflite::FlatBufferModel::BuildFromBuffer(
      reinterpret_cast<const char*>(output_builder.GetCurrentBufferPointer()),
      output_builder.GetSize());

#include "tensorflow/core/framework/types.pb.h"
#include "tensorflow/lite/tools/optimize/reduced_precision_support.h"

namespace mlir {
namespace lite {

absl::Status SparsifyModel(const tflite::ModelT& input_model,
                           flatbuffers::FlatBufferBuilder* builder) {
  MLIRContext context;
  StatusScopedDiagnosticHandler statusHandler(&context,

    return nullptr;
  }
  auto tflite_model = std::make_unique<tflite::ModelT>();
  model->GetModel()->UnPackTo(tflite_model.get(), nullptr);

  flatbuffers::FlatBufferBuilder builder;
  auto status = mlir::lite::SparsifyModel(*tflite_model, &builder);

  if (!status.ok()) {
    error_reporter->exception();
    return nullptr;
  }
  return tflite::python_utils::ConvertToPyString(