fix: remove <|im_end|> returned by llama-server api.

Author: jason-ni

mineru_vl_utils/vlm_client/http_client.py

@@ -275,6 +275,8 @@ def get_response_content(self, response_data: dict) -> str:
         content = message["content"]
         if not (content is None or isinstance(content, str)):
             raise ServerError(f"Unexpected content type: {type(content)}.")
+        if isinstance(content, str) and content.endswith('<|im_end|>'):
+            content = content[:-10]
         return content or ""
 
     def predict(