const auto func_type =
        builder.getFunctionType(/*inputs=*/{}, /*results=*/{});
    auto summary_func = builder.create<func::FuncOp>(
        builder.getUnknownLoc(), /*sym_name=*/"summary", func_type);
    summary_func.setPrivate();
    summary_func->setAttr("quantization_summary",
                          builder.getStringAttr(log_message));
  }

 private:

          rewriter.getUnknownLoc(), rewriter.getNoneType(),
          rewriter.getUnitAttr());
      op.getInputGateBiasMutable().assign(none_value);
      return success();
    }

    if (EqualsZero(op.getProjectionBias())) {
      auto none_value = rewriter.create<TFL::NoValueOp>(
          rewriter.getUnknownLoc(), rewriter.getNoneType(),
          rewriter.getUnitAttr());

        min_maxs);
    // TODO(fengliuai): this quantization dimension isn't correct.
    axis = b.getI64IntegerAttr(tensor.quantization->quantized_dimension);
  }
  return b.create<mlir::quantfork::StatisticsOp>(b.getUnknownLoc(), res,
                                                 layer_stats, axis_stats, axis);
}

// Returns true if this is a basic LSTM op.
bool IsBasicLSTMOp(tflite::BuiltinOptionsUnion op_union) {

        // uninitialized variables, use the type information to construct a
        // dummy uninitialized variable operation.
        auto op = builder.create<mlir::tf_saved_model::GlobalTensorOp>(
            builder.getUnknownLoc(),
            builder.getStringAttr(object_names.GetSymbolTableName(node_id)),
            mlir::ElementsAttr(),
            /*type=*/mlir::TypeAttr::get(type),
            /*is_mutable=*/builder.getUnitAttr());