Merge branch 'dev' into dev_1209

Author: pekopoke

dingo/model/llm/rag/llm_rag_context_precision.py

@@ -254,7 +254,7 @@ def process_response(cls, responses: List[str]) -> ModelRes:
         result = ModelRes()
         result.score = score
 
-        # 根据分数判断是否通过（默认阈值5，满分10分）
+        # 根据分数判断是否通过，默认阈值为5
         threshold = 5
         if hasattr(cls, 'dynamic_config') and cls.dynamic_config.parameters:
             threshold = cls.dynamic_config.parameters.get('threshold', 5)

dingo/model/llm/rag/llm_rag_context_recall.py

@@ -49,6 +49,7 @@ class LLMRAGContextRecall(BaseOpenAI):
 
     prompt = """上下文召回评估提示词，用于分类陈述归因"""
 
+    @staticmethod
     def context_recall_prompt(question: str, context: str, answer: str) -> str:
         """
         生成上下文召回评估的提示词
@@ -200,7 +201,7 @@ def process_response(cls, response: str) -> ModelRes:
         result = ModelRes()
         result.score = score
 
-        # 根据分数判断是否通过（默认阈值5，满分10分）
+        # 根据分数判断是否通过，默认阈值为5
         threshold = 5
         if hasattr(cls, 'dynamic_config') and cls.dynamic_config.parameters:
             threshold = cls.dynamic_config.parameters.get('threshold', 5)

dingo/model/llm/rag/llm_rag_context_relevancy.py

@@ -45,6 +45,7 @@ class LLMRAGContextRelevancy(BaseOpenAI):
         "source_frameworks": "Ragas + DeepEval + TruLens"
     }
 
+    @staticmethod
     def context_relevance_judge1_prompt(query: str, context: str) -> str:
         """
         First judge template for context relevance evaluation (Chinese version).
@@ -80,6 +81,7 @@ def context_relevance_judge1_prompt(query: str, context: str) -> str:
 请不要尝试解释。
 分析上下文和问题后，相关性分数为 """
 
+    @staticmethod
     def context_relevance_judge2_prompt(query: str, context: str) -> str:
         """
         Second judge template for context relevance evaluation (Chinese version).
@@ -200,7 +202,7 @@ def process_response(cls, response: str) -> ModelRes:
         result = ModelRes()
         result.score = score
 
-        # 根据分数判断是否通过（默认阈值5，满分10分）
+        # 根据分数判断是否通过，默认阈值为5
         threshold = 5
         if hasattr(cls, 'dynamic_config') and cls.dynamic_config.parameters:
             threshold = cls.dynamic_config.parameters.get('threshold', 5)

dingo/model/llm/rag/llm_rag_faithfulness.py

@@ -43,6 +43,7 @@ class LLMRAGFaithfulness(BaseOpenAI):
         "source_frameworks": "Ragas + DeepEval"
     }
 
+    @staticmethod
     def statement_generator_prompt(question: str, answer: str) -> str:
         """
         Prompt to generate statements from answer (Chinese version).
@@ -67,18 +68,19 @@ def statement_generator_prompt(question: str, answer: str) -> str:
 
 请以JSON格式返回结果，格式如下：
 ```json
-{
+{{
     "statements": [
         "陈述1",
         "陈述2",
         "陈述3"
     ]
-}
+}}
 ```
 
 请不要输出其他内容，只返回JSON格式的结果。
 """
 
+    @staticmethod
     def faithfulness_judge_prompt(context: str, statements: List[str]) -> str:
         """
         Prompt to judge faithfulness of statements (Chinese version).
@@ -103,15 +105,15 @@ def faithfulness_judge_prompt(context: str, statements: List[str]) -> str:
 
 请以JSON格式返回结果，格式如下：
 ```json
-{
+{{
     "statements": [
-        {
+        {{
             "statement": "原始陈述，一字不差",
             "reason": "判断理由",
             "verdict": 0或1
-        }
+        }}
     ]
-}
+}}
 ```
 
 请不要输出其他内容，只返回JSON格式的结果。
@@ -284,7 +286,7 @@ def process_response(cls, response: str) -> ModelRes:
         result = ModelRes()
         result.score = score
 
-        # 根据分数判断是否通过（默认阈值5，满分10分）
+        # 根据分数判断是否通过，默认阈值为5
         threshold = 5
         if hasattr(cls, 'dynamic_config') and cls.dynamic_config.parameters:
             threshold = cls.dynamic_config.parameters.get('threshold', 5)