// of the CustomOp are quantizable. 'is_weight_only' is used specify whether the
// custom op is quantized only for storage and dequantized at runtime.
// 'no_side_effect' is used to determine whether the op can be pruned if
// considered as trivially dead.
struct CustomOpInfo {
  std::vector<std::int32_t> quantizable_input_indices;
  bool is_weight_only = false;
  bool no_side_effect = true;
};

  for (const auto& entry : custom_op_map) {
    quant_specs.custom_map[entry.first].quantizable_input_indices =
        entry.second.quantizable_input_indices;
    quant_specs.custom_map[entry.first].is_weight_only =
        entry.second.is_weight_only;
    quant_specs.custom_map[entry.first].no_side_effect =
        entry.second.no_side_effect;
  }

  if (quant_specs.inference_type == tensorflow::DT_INT8)

          ops_blocklist.find(custom_op_name) != ops_blocklist.end();

      bool weight_only_custom_op = custom_map_iter != custom_op_map.end()
                                       ? custom_map_iter->second.is_weight_only
                                       : false;

      return is_blocklisted || weight_only_custom_op ||
             weight_only_quantization;
    } else {
      auto dynamic_range_op =

  // be quantized.
  CustomOpMap custom_op_map;
  custom_op_map["CustomTestOp"] = {
      {1},   // quantizable_input_indices
      true,  // is_weight_only
  };

  flatbuffers::FlatBufferBuilder builder;
  auto status = QuantizeWeights(&builder, model_, 0, custom_op_map);
  ASSERT_EQ(status, kTfLiteOk);

  const uint8_t* buffer = builder.GetBufferPointer();