diff --git a/InsightEngine/llms/base.py b/InsightEngine/llms/base.py
index 48a5157..090c10c 100644
--- a/InsightEngine/llms/base.py
+++ b/InsightEngine/llms/base.py
@@ -83,7 +83,6 @@ class LLMClient:
             return self.validate_response(response.choices[0].message.content)
         return ""
 
-    @with_retry(LLM_RETRY_CONFIG)
     def stream_invoke(self, system_prompt: str, user_prompt: str, **kwargs) -> Generator[str, None, None]:
         """
         流式调用LLM，逐步返回响应内容
@@ -131,6 +130,7 @@ class LLMClient:
             logger.error(f"流式请求失败: {str(e)}")
             raise e
     
+    @with_retry(LLM_RETRY_CONFIG)
     def stream_invoke_to_string(self, system_prompt: str, user_prompt: str, **kwargs) -> str:
         """
         流式调用LLM并安全地拼接为完整字符串（避免UTF-8多字节字符截断）
diff --git a/MediaEngine/llms/base.py b/MediaEngine/llms/base.py
index 4cbb3ca..888b0a5 100644
--- a/MediaEngine/llms/base.py
+++ b/MediaEngine/llms/base.py
@@ -86,7 +86,6 @@ class LLMClient:
             return self.validate_response(response.choices[0].message.content)
         return ""
 
-    @with_retry(LLM_RETRY_CONFIG)
     def stream_invoke(self, system_prompt: str, user_prompt: str, **kwargs) -> Generator[str, None, None]:
         """
         流式调用LLM，逐步返回响应内容
@@ -134,6 +133,7 @@ class LLMClient:
             logger.error(f"流式请求失败: {str(e)}")
             raise e
     
+    @with_retry(LLM_RETRY_CONFIG)
     def stream_invoke_to_string(self, system_prompt: str, user_prompt: str, **kwargs) -> str:
         """
         流式调用LLM并安全地拼接为完整字符串（避免UTF-8多字节字符截断）
diff --git a/QueryEngine/llms/base.py b/QueryEngine/llms/base.py
index 399ab65..8acc6cf 100644
--- a/QueryEngine/llms/base.py
+++ b/QueryEngine/llms/base.py
@@ -83,7 +83,6 @@ class LLMClient:
             return self.validate_response(response.choices[0].message.content)
         return ""
 
-    @with_retry(LLM_RETRY_CONFIG)
     def stream_invoke(self, system_prompt: str, user_prompt: str, **kwargs) -> Generator[str, None, None]:
         """
         流式调用LLM，逐步返回响应内容
@@ -131,6 +130,7 @@ class LLMClient:
             logger.error(f"流式请求失败: {str(e)}")
             raise e
     
+    @with_retry(LLM_RETRY_CONFIG)
     def stream_invoke_to_string(self, system_prompt: str, user_prompt: str, **kwargs) -> str:
         """
         流式调用LLM并安全地拼接为完整字符串（避免UTF-8多字节字符截断）
diff --git a/ReportEngine/llms/base.py b/ReportEngine/llms/base.py
index 64e34ac..29723c5 100644
--- a/ReportEngine/llms/base.py
+++ b/ReportEngine/llms/base.py
@@ -76,7 +76,6 @@ class LLMClient:
             return self.validate_response(response.choices[0].message.content)
         return ""
 
-    @with_retry(LLM_RETRY_CONFIG)
     def stream_invoke(self, system_prompt: str, user_prompt: str, **kwargs) -> Generator[str, None, None]:
         """
         流式调用LLM，逐步返回响应内容
@@ -118,6 +117,7 @@ class LLMClient:
             logger.error(f"流式请求失败: {str(e)}")
             raise e
     
+    @with_retry(LLM_RETRY_CONFIG)
     def stream_invoke_to_string(self, system_prompt: str, user_prompt: str, **kwargs) -> str:
         """
         流式调用LLM并安全地拼接为完整字符串（避免UTF-8多字节字符截断）