fix bug in op reasoninganswerjudgerfilter

dataflow/operators/reasoning/filter/reasoning_answer_model_judge_filter.py

@@ -3,7 +3,7 @@
 from dataflow.core import OperatorABC
 from dataflow.utils.storage import DataFlowStorage
 from dataflow.core import LLMServingABC
-from dataflow.prompts.model_evaluation.general import AnswerJudgePrompt
+from dataflow.prompts.model_evaluation.general import AnswerJudgePromptQuestion, AnswerJudgePrompt
 from dataflow.core.prompt import prompt_restrict, DIYPromptABC
 
 import re
@@ -12,15 +12,16 @@
 from typing import Union
 
 @prompt_restrict(
-    AnswerJudgePrompt
+    AnswerJudgePromptQuestion,
+    AnswerJudgePrompt,
 )
 
 @OPERATOR_REGISTRY.register()
 class ReasoningAnswerModelJudgeFilter(OperatorABC):
     def __init__(self,
                  system_prompt: str = "You are a helpful assistant specialized in evaluating answer correctness.",
                  llm_serving: LLMServingABC = None,
-                 prompt_template: Union[AnswerJudgePrompt, DIYPromptABC] = AnswerJudgePrompt,
+                 prompt_template: Union[AnswerJudgePromptQuestion,AnswerJudgePrompt, DIYPromptABC] = AnswerJudgePromptQuestion,
                  keep_all_samples: bool = False,  # 新增参数，控制是否保留所有样本
                  ):
 

dataflow/prompts/model_evaluation/general.py

@@ -13,7 +13,7 @@ class AnswerJudgePrompt(PromptABC):
     def __init__(self):
         pass
 
-    def build_prompt(self, answer, reference_answer):
+    def build_prompt(self, answer, reference_answer, question=None):
         prompt = f"""
         As an answer evaluation expert, please assess whether the following answer is correct.
 

dataflow/statics/pipelines/api_pipelines/reasoning_diy_pipeline.py

@@ -61,7 +61,7 @@ def __init__(self, llm_serving: LLMServingABC = None):
         )
 
         self.question_filter_step1 = ReasoningQuestionFilter(
-            system_prompt="You are an expert in evaluating mathematical problems. Follow the user's instructions strictly and output your final judgment in the required JSON format.",
+            system_prompt="You are an expert in evaluating problems. Follow the user's instructions strictly and output your final judgment in the required JSON format.",
             llm_serving=self.llm_serving,
             prompt_template=DiyQuestionFilterPrompt(DIY_PROMPT_QUESTION)
         )

dataflow/statics/pipelines/api_pipelines/reasoning_general_pipeline.py

@@ -31,7 +31,7 @@ def __init__(self, llm_serving: LLMServingABC = None):
         )
 
         self.question_filter_step1 = ReasoningQuestionFilter(
-            system_prompt="You are an expert in evaluating mathematical problems. Follow the user's instructions strictly and output your final judgment in the required JSON format.",
+            system_prompt="You are an expert in evaluating problems. Follow the user's instructions strictly and output your final judgment in the required JSON format.",
             llm_serving=self.llm_serving,
             prompt_template=GeneralQuestionFilterPrompt()
         )