benchmark工具支持受限解码场景指定response_format (#4718)

benchmarks/backend_request_func.py

@@ -51,6 +51,7 @@ class RequestFuncInput:
     ignore_eos: bool = False
     language: Optional[str] = None
     debug: bool = False
+    response_format: Optional[dict] = None
 
 
 @dataclass
@@ -93,8 +94,11 @@ async def async_request_eb_openai_chat_completions(
             "stream_options": {
                 "include_usage": True,
                 "continuous_usage_stats": True,
-            },
+            }
         }
+        if request_func_input.response_format:
+            payload["response_format"] =request_func_input.response_format
+
         # 超参由yaml传入
         payload.update(request_func_input.hyper_parameters)