Method _evaluate_answer

src/server/tasks/os_interaction/task.py:671–698

评估答案

(
            self, answer, config: JudgeConfig, container: Container, session: Session
    )

        ))

    async def _evaluate_answer(
            self, answer, config: JudgeConfig, container: Container, session: Session
    ) -> dict:
        """评估答案"""
        result = {"success": False}

        # 处理答案格式
        if isinstance(answer, str) and config.match and config.match["strip"]:
            answer = answer.strip()
        logging.info(f"Final answer: {answer}")

        # 使用匹配标准评估
        if config.match:
            result["success"] = self._evaluate_by_match(answer, config)
        # 使用检查脚本评估
        elif config.check:
            result["success"] = await self._evaluate_by_check_scripts(answer, config, container)
        # 无评估方法
        else:
            logging.error("No evaluation method specified")
            final_rewardhistory = RewardHistoryItem(reward=0, score=0)
            session.inject(final_rewardhistory)
            result["error"] = True
            result["result"] = TaskSampleExecutionResult(
                status=SampleStatus.TASK_ERROR, result={"result": False}
            )

        return result

    def _evaluate_by_match(self, answer: str, config: JudgeConfig) -> bool:
        """使用匹配标准评估答案"""