add option to shorten prompt print in log (#991)

Signed-off-by: Lei Wen <wenlei03@qiyi.com>
Co-authored-by: Lei Wen <wenlei03@qiyi.com>
Co-authored-by: Zhuohan Li <zhuohan123@gmail.com>

vllm/engine/arg_utils.py

@@ -171,6 +171,7 @@ class AsyncEngineArgs(EngineArgs):
     """Arguments for asynchronous vLLM engine."""
     engine_use_ray: bool = False
     disable_log_requests: bool = False
+    max_log_len: Optional[int] = None
 
     @staticmethod
     def add_cli_args(
@@ -183,4 +184,10 @@ class AsyncEngineArgs(EngineArgs):
         parser.add_argument('--disable-log-requests',
                             action='store_true',
                             help='disable logging requests')
+        parser.add_argument('--max-log-len',
+                            type=int,
+                            default=None,
+                            help='max number of prompt characters or prompt '
+                            'ID numbers being printed in log. '
+                            'Default: unlimited.')
         return parser

vllm/engine/async_llm_engine.py

@@ -242,11 +242,13 @@ class AsyncLLMEngine:
                  engine_use_ray: bool,
                  *args,
                  log_requests: bool = True,
+                 max_log_len: Optional[int] = None,
                  start_engine_loop: bool = True,
                  **kwargs) -> None:
         self.worker_use_ray = worker_use_ray
         self.engine_use_ray = engine_use_ray
         self.log_requests = log_requests
+        self.max_log_len = max_log_len
         self.engine = self._init_engine(*args, **kwargs)
 
         self.request_tracker: RequestTracker = RequestTracker()
@@ -325,10 +327,18 @@ class AsyncLLMEngine:
         arrival_time: Optional[float] = None,
     ) -> AsyncStream:
         if self.log_requests:
+            shortened_prompt = prompt
+            shortened_token_ids = prompt_token_ids
+            if self.max_log_len is not None:
+                if shortened_prompt is not None:
+                    shortened_prompt = shortened_prompt[:self.max_log_len]
+                if shortened_token_ids is not None:
+                    shortened_token_ids = shortened_token_ids[:self.
+                                                              max_log_len]
             logger.info(f"Received request {request_id}: "
-                        f"prompt: {prompt!r}, "
+                        f"prompt: {shortened_prompt!r}, "
                         f"sampling params: {sampling_params}, "
-                        f"prompt token ids: {prompt_token_ids}.")
+                        f"prompt token ids: {shortened_token_ids}.")
 
         if not self.is_running:
             if self.start_engine_loop:
@@ -446,5 +456,6 @@ class AsyncLLMEngine:
                      placement_group,
                      log_requests=not engine_args.disable_log_requests,
                      log_stats=not engine_args.disable_log_stats,
+                     max_log_len=engine_args.max_log_len,
                      start_engine_loop=start_engine_loop)
         return engine