benchmark_serving support --served-model-name param (vllm-project#12109)

Signed-off-by: zibai <[email protected]>
Co-authored-by: Roger Wang <[email protected]>

benchmarks/backend_request_func.py

@@ -22,6 +22,7 @@ class RequestFuncInput:
     prompt_len: int
     output_len: int
     model: str
+    model_name: Optional[str] = None
     best_of: int = 1
     logprobs: Optional[int] = None
     extra_body: Optional[dict] = None
@@ -78,7 +79,7 @@ async def async_request_tgi(
                             continue
                         chunk_bytes = chunk_bytes.decode("utf-8")
 
-                        #NOTE: Sometimes TGI returns a ping response without
+                        # NOTE: Sometimes TGI returns a ping response without
                         # any data, we should skip it.
                         if chunk_bytes.startswith(":"):
                             continue
@@ -235,7 +236,8 @@ async def async_request_openai_completions(
 
     async with aiohttp.ClientSession(timeout=AIOHTTP_TIMEOUT) as session:
         payload = {
-            "model": request_func_input.model,
+            "model": request_func_input.model_name \
+                if request_func_input.model_name else request_func_input.model,
             "prompt": request_func_input.prompt,
             "temperature": 0.0,
             "best_of": request_func_input.best_of,
@@ -328,7 +330,8 @@ async def async_request_openai_chat_completions(
         if request_func_input.multi_modal_content:
             content.append(request_func_input.multi_modal_content)
         payload = {
-            "model": request_func_input.model,
+            "model": request_func_input.model_name \
+                if request_func_input.model_name else request_func_input.model,
             "messages": [
                 {
                     "role": "user",

benchmarks/benchmark_serving.py

@@ -525,6 +525,7 @@ async def benchmark(
     api_url: str,
     base_url: str,
     model_id: str,
+    model_name: str,
     tokenizer: PreTrainedTokenizerBase,
     input_requests: List[Tuple[str, int, int]],
     logprobs: Optional[int],
@@ -553,6 +554,7 @@ async def benchmark(
             "Multi-modal content is only supported on 'openai-chat' backend.")
     test_input = RequestFuncInput(
         model=model_id,
+        model_name=model_name,
         prompt=test_prompt,
         api_url=api_url,
         prompt_len=test_prompt_len,
@@ -573,6 +575,7 @@ async def benchmark(
     if profile:
         print("Starting profiler...")
         profile_input = RequestFuncInput(model=model_id,
+                                         model_name=model_name,
                                          prompt=test_prompt,
                                          api_url=base_url + "/start_profile",
                                          prompt_len=test_prompt_len,
@@ -616,6 +619,7 @@ async def limited_request_func(request_func_input, pbar):
     async for request in get_request(input_requests, request_rate, burstiness):
         prompt, prompt_len, output_len, mm_content = request
         request_func_input = RequestFuncInput(model=model_id,
+                                              model_name=model_name,
                                               prompt=prompt,
                                               api_url=api_url,
                                               prompt_len=prompt_len,
@@ -780,6 +784,7 @@ def main(args: argparse.Namespace):
 
     backend = args.backend
     model_id = args.model
+    model_name = args.served_model_name
     tokenizer_id = args.tokenizer if args.tokenizer is not None else args.model
     tokenizer_mode = args.tokenizer_mode
 
@@ -877,6 +882,7 @@ def main(args: argparse.Namespace):
             api_url=api_url,
             base_url=base_url,
             model_id=model_id,
+            model_name=model_name,
             tokenizer=tokenizer,
             input_requests=input_requests,
             logprobs=args.logprobs,
@@ -1222,5 +1228,12 @@ def main(args: argparse.Namespace):
         'always use the slow tokenizer. \n* '
         '"mistral" will always use the `mistral_common` tokenizer.')
 
+    parser.add_argument("--served-model-name",
+                        type=str,
+                        default=None,
+                        help="The model name used in the API. "
+                        "If not specified, the model name will be the "
+                        "same as the ``--model`` argument. ")
+
     args = parser.parse_args()
     main(args)