源码中调用RolloutManager.preprocess_one方法出现报错

[itlantu]

报错信息

• 报错代码文件： https://github.com/TencentCloudADP/youtu-agent/blob/main/utu/practice/rollout_manager.py

• 报错代码（第112行）：

processed_sample = processer.preprocess_one(sample, recorder)

• 终端提示

  File "C:\Users\IT_la\Desktop\mbpp_youtu\youtu-agent\utu\practice\rollout_manager.py", line 104, in preprocess_batch
    processed_sample = self.preprocess_one(sample, recorder)
  File "C:\Users\IT_la\Desktop\mbpp_youtu\youtu-agent\utu\practice\rollout_manager.py", line 112, in preprocess_one
    processed_sample = processer.preprocess_one(sample, recorder)
TypeError: BaseLLMJudgeProcesser.preprocess_one() takes 2 positional arguments but 3 were given

报错原因分析

`preprocess_one 函数的调用参数与定义的函数签名不匹配，导致调用时触发参数数量错误。

utu/eval/processer/base_processor.py中定义的基类BaseProcesser规定了所有子类必须实现的preprocess_one函数签名

class BaseProcesser:
    # 其他代码

    @abc.abstractmethod
    def preprocess_one(self, sample: EvaluationSample) -> EvaluationSample:
        """Preprocess a single sample."""
        raise NotImplementedError

• preprocess目录： https://github.com/TencentCloudADP/youtu-agent/tree/main/utu/eval/processer
• base_processor.py文件： https://github.com/TencentCloudADP/youtu-agent/blob/main/utu/eval/processer/base_processor.py

在utu/practice/rollout_manager.py中调用preprocess_one方法时，额外传入了recorder参数，导致参数数量超出定义

processed_sample = processer.preprocess_one(sample, recorder)

[YZ-Cai]

您好，Practice模块调用的processer应该为utu/eval/processer/training_free_grpo_processor.py，其中包含如下函数：

def preprocess_one(self, sample: EvaluationSample, recorder=None) -> EvaluationSample:
        """Preprocess a single sample with optional experience recorder.

为了正确调用该processor，请检查您的数据集中source字段是否为training_free_grpo：

{
    "dataset": "YourDataset",       # Name of your dataset
    "source": "training_free_grpo", # Must be "training_free_grpo" for both practice and evaluation data
    "input": "What is 2+2?",        # The question or prompt
    "answer": "4"                   # The expected answer, set to None if not applicable
}

[itlantu]

感谢您的答复，我使用自定义数据，并且确保数据集中的source字段的值都为training_free_grpo，但运行scripts/run_training_free_GRPO.py时依然产生了这个错误，我的单条数据格式如下：

{
    "dataset": "mbpp",
    "source": "training_free_grpo",
    "input": "Write a python function to find the minimum difference between any two elements in a given array. https://www.geeksforgeeks.org/find-minimum-difference-pair/",
    "answer": "def find_min_diff(arr,n): \n    arr = sorted(arr) \n    diff = 10**20 \n    for i in range(n-1): \n        if arr[i+1] - arr[i] < diff: \n            diff = arr[i+1] - arr[i]  \n    return diff "
}

我在产生报错的RolloutManager.preprocess_one函数中添加了一个print(sample)

打印的结果为：

evel=0 
time_cost=None 
confidence=None
created_at=datetime.datetime(2026, 1, 16, 21, 20, 20, 798125) 
augmented_question=''
trajectory=None 
exp_id='mbpp_practice_epoch_0' 
correct_answer='' 
rajectories=None 
updated_at=datetime.datetime(2026, 1, 16, 21, 20, 20, 798228) 
file_name='' 
extracted_final_answer=None 
stage='init' 
dataset='mbpp' 
meta=None 
judged_response=None
dataset_index=0 
trace_id=None 
reasoning=None 
source='' 
trace_url=None
correct=None 
id=366 
raw_question='Write a python function to find the minimum difference between any two elements in a given array. https://www.geeksforgeeks.org/find-minimum-difference-pair/' 
response=None
reward=None

我观察到raw_question与jsonl中的input字段对应，但source为为空字符串而不是'training_free_grpo'
我在手动赋值sample.source = 'training_free_grpo'后代码可以正常执行

[YZ-Cai]

感谢您的反馈，根据您打印的内容来看，是因为scripts/data/upload_dataset.py上传数据时未成功写入source字段。我们已经在 Issue #230 中修复了这个问题，将scripts/data/upload_dataset.py的默认数据格式从llamafactory格式改为了Youtu-Agent格式～