update benchmark scripts (#4497)

benchmarks/backend_request_func.py

@@ -112,6 +112,7 @@ async def async_request_eb_openai_chat_completions(
         output = RequestFuncOutput()
         output.prompt_len = 0
         output.no = request_func_input.no
+        request_id = "None"
 
         ttft = 0.0
         st = time.perf_counter()
@@ -131,6 +132,9 @@ async def async_request_eb_openai_chat_completions(
                             timestamp = time.perf_counter()
                             data = json.loads(chunk)
 
+                            if request_id == "None" and "id" in data:
+                                request_id = data["id"]
+
                             if choices := data.get("choices"):
                                 content = choices[0]["delta"].get("content")
                                 reason_content = choices[0]["delta"].get("reasoning_content")
@@ -175,12 +179,13 @@ async def async_request_eb_openai_chat_completions(
                     )
                     output.error = error_text or ""
                     output.success = False
-                output.request_id = data.get("id", "")
         except Exception:
             output.success = False
             exc_info = sys.exc_info()
             output.error = "".join(traceback.format_exception(*exc_info))
 
+        output.request_id = request_id
+
         # 保存失败请求结果
         if not output.success:
             with open("error_output.txt", "a") as f:

benchmarks/benchmark_serving.py

@@ -177,7 +177,7 @@ def calculate_metrics(
             output_len = outputs[i].output_tokens
 
             if not output_len:
-                print("no output_len")
+                print("no output_len", outputs[i])
                 # We use the tokenizer to count the number of output tokens
                 # for some serving backends instead of looking at
                 # len(outputs[i].itl) since multiple output tokens may be