优化DifyCompareTest和ModelTool中的API调用逻辑，增加重试机制以提高稳定性，更新模型名称获取方式为使用环境变量。

rag2_0/dify/DifyCompareTest.py

@@ -46,7 +46,7 @@ class DifyCompareTest:
         self.first_wiki_client = ChatClient(api_key="app-gocvuqduBnJptYNPpnW9V9R6", base_url=os.getenv("DIFY_BSAE_URL"))
         # 词条与工单同时检索
         self.both_wiki_worker_client = ChatClient(api_key="app-CPoOMaGDsLRPAe9TW7Xjhszy", base_url=os.getenv("DIFY_BSAE_URL"))
-        self.llm = OpenAiLLM(base_url=os.getenv("OPENAI_API_BASE"), model="deepseek-ai/DeepSeek-R1")
+        self.llm = OpenAiLLM(base_url=os.getenv("OPENAI_API_BASE"), model=os.getenv("MODEL_NAME"))
 
     def llm_judge_answer(self, old_answer: str, now_answer: str):
         user_prompt = f"""
@@ -76,7 +76,6 @@ class DifyCompareTest:
                 response.content = response.content.strip()
                 clean_output = re.sub(r'<think>.*?</think>', '', response.content, flags=re.DOTALL)
                 result = JsonOutputParser().parse(clean_output)
-                result = json.loads(clean_output)
                 return "回答基本相同" if result.get("is_same", False) else "回答基本不相同"
             except Exception as e:
                 retry_count += 1
@@ -91,17 +90,28 @@ class DifyCompareTest:
         
     def process_workflow(self, workflow_name, client, inputs, query, old_answer):
         """处理单个工作流调用"""
-        try:
+        max_retries = 3
-            response = client.create_chat_message(
+        retry_count = 0
-                inputs=inputs, query=query, user="AutoCodeRun", response_mode="blocking"
+        
-            )
+        while retry_count < max_retries:
-            result = response.json()
+            try:
-            answer = result.get('answer', "")
+                response = client.create_chat_message(
-            judge_result = self.llm_judge_answer(old_answer=old_answer, now_answer=answer)
+                    inputs=inputs, query=query, user="AutoCodeRun", response_mode="blocking"
-            return answer, judge_result
+                )
-        except Exception as e:
+                result = response.json()
-            logging.error(f"{workflow_name}调用失败: {e}")
+                answer = result.get('answer', "")
-            return '', ''
+                if len(answer) == 0:
+                    raise Exception(f"回答为空: {result}")
+                judge_result = self.llm_judge_answer(old_answer=old_answer, now_answer=answer)
+                return answer, judge_result
+            except Exception as e:
+                retry_count += 1
+                if retry_count >= max_retries:
+                    logging.error(f"{workflow_name}调用失败 (尝试 {max_retries} 次后): {e}")
+                    return '', ''
+                else:
+                    import time
+                    time.sleep(1)  # 等待1秒后重试
 
     def process_single_row(self, index, row):
         """处理单行数据的方法，用于多线程执行"""
@@ -247,7 +257,7 @@ if __name__ == "__main__":
         
         # 处理第一个文件
         excel_files = [
-            ("data/excel/5月.xlsx", "data/excel/5月问答对比.xlsx"),
+            # ("data/excel/5月.xlsx", "data/excel/5月问答对比.xlsx"),
             ("data/excel/其他月.xlsx", "data/excel/其他月问答对比.xlsx")
         ]
         

rag2_0/tool/ModelTool.py

@@ -240,7 +240,7 @@ class OpenAiLLM:
 
         self._kwargs = kwargs
 
-    def invoke(self, user_prompt="你是谁？", need_retry=True,**extra_kwargs):
+    def invoke(self, user_prompt="你是谁？", need_retry=True, api_key:str = None, **extra_kwargs):
         # 初始化 OpenAI 客户端
 
         max_retries = 3
@@ -253,10 +253,13 @@ class OpenAiLLM:
             timeout = httpx.Timeout(300.0) 
             self._kwargs["timeout"] = timeout
 
+        if api_key is None:
+            api_key = APIKeyManager.get_api_key()  
+            
         if need_retry:
             while retry_count < max_retries:
                 try:
-                    api_key = APIKeyManager.get_api_key()  
+                    
                     # 使用with语句创建客户端，确保资源会被正确释放
                     with OpenAI(api_key=api_key, base_url=self._url) as client:
                         # 创建 Completion 请求. 超时120s
@@ -276,7 +279,6 @@ class OpenAiLLM:
         else:
             try:
                 # 创建 Completion 请求. 超时120s
-                api_key = APIKeyManager.get_api_key()  
                 # 使用with语句创建客户端，确保资源会被正确释放
                 with OpenAI(api_key=api_key, base_url=self._url) as client:
                     completion = client.chat.completions.create(