Improve benchmark scripts and error message printing (#2922)

python/sglang/bench_offline_throughput.py

@@ -39,14 +39,15 @@ class BenchArgs:
     dataset_path: str = ""
     num_prompts: int = 1000
     sharegpt_output_len: Optional[int] = None
+    sharegpt_context_len: Optional[int] = None
     random_input_len: int = 1024
     random_output_len: int = 1024
     random_range_ratio: float = 0.0
-    gen_num_groups: int = 64
-    gen_prompts_per_group: int = 16
-    gen_system_prompt_len: int = 2048
-    gen_question_len: int = 128
-    gen_output_len: int = 256
+    gsp_num_groups: int = 64
+    gsp_prompts_per_group: int = 16
+    gsp_system_prompt_len: int = 2048
+    gsp_question_len: int = 128
+    gsp_output_len: int = 256
     disable_ignore_eos: bool = False
     extra_request_body: Optional[str] = None
     seed: int = 1
@@ -82,6 +83,12 @@ class BenchArgs:
             default=BenchArgs.sharegpt_output_len,
             help="Output length for each request. Overrides the output length from the ShareGPT dataset.",
         )
+        parser.add_argument(
+            "--sharegpt-context-len",
+            type=int,
+            default=BenchArgs.sharegpt_context_len,
+            help="The context length of the model for the ShareGPT dataset. Requests longer than the context length will be dropped.",
+        )
         parser.add_argument(
             "--random-input-len",
             type=int,
@@ -102,35 +109,35 @@ class BenchArgs:
             "used only for random dataset.",
         )
         parser.add_argument(
-            "--gen-num-groups",
+            "--gsp-num-groups",
             type=int,
-            default=BenchArgs.gen_num_groups,
+            default=BenchArgs.gsp_num_groups,
             help="Number of groups with shared prefix, used"
             "only for generate-shared-prefix",
         )
         parser.add_argument(
-            "--gen-prompts-per-group",
+            "--gsp-prompts-per-group",
             type=int,
-            default=BenchArgs.gen_prompts_per_group,
+            default=BenchArgs.gsp_prompts_per_group,
             help="Number of prompts per group of shared prefix, used"
             "only for generate-shared-prefix",
         )
         parser.add_argument(
-            "--gen-system-prompt-len",
+            "--gsp-system-prompt-len",
             type=int,
-            default=BenchArgs.gen_system_prompt_len,
+            default=BenchArgs.gsp_system_prompt_len,
             help="System prompt length, used" "only for generate-shared-prefix",
         )
         parser.add_argument(
-            "--gen-question-len",
+            "--gsp-question-len",
             type=int,
-            default=BenchArgs.gen_question_len,
+            default=BenchArgs.gsp_question_len,
             help="Question length, used" "only for generate-shared-prefix",
         )
         parser.add_argument(
-            "--gen-output-len",
+            "--gsp-output-len",
             type=int,
-            default=BenchArgs.gen_output_len,
+            default=BenchArgs.gsp_output_len,
             help="Target length in tokens for outputs in generated-shared-prefix dataset",
         )
         parser.add_argument(

python/sglang/bench_serving.py

@@ -452,6 +452,7 @@ def get_dataset(args, tokenizer):
             num_requests=args.num_prompts,
             tokenizer=tokenizer,
             fixed_output_len=args.sharegpt_output_len,
+            context_len=args.sharegpt_context_len,
         )
     elif args.dataset_name == "random":
         input_requests = sample_random_requests(
@@ -464,11 +465,11 @@ def get_dataset(args, tokenizer):
         )
     elif args.dataset_name == "generated-shared-prefix":
         input_requests = sample_generated_shared_prefix_requests(
-            num_groups=args.gen_num_groups,
-            prompts_per_group=args.gen_prompts_per_group,
-            system_prompt_len=args.gen_system_prompt_len,
-            question_len=args.gen_question_len,
-            output_len=args.gen_output_len,
+            num_groups=args.gsp_num_groups,
+            prompts_per_group=args.gsp_prompts_per_group,
+            system_prompt_len=args.gsp_system_prompt_len,
+            question_len=args.gsp_question_len,
+            output_len=args.gsp_output_len,
             tokenizer=tokenizer,
         )
     else:
@@ -560,6 +561,7 @@ def sample_sharegpt_requests(
     num_requests: int,
     tokenizer: PreTrainedTokenizerBase,
     fixed_output_len: Optional[int] = None,
+    context_len: Optional[int] = None,
 ) -> List[Tuple[str, int, int]]:
     if fixed_output_len is not None and fixed_output_len < 4:
         raise ValueError("output_len too small")
@@ -597,14 +599,15 @@ def sample_sharegpt_requests(
         output_len = (
             len(completion_token_ids) if fixed_output_len is None else fixed_output_len
         )
-        if prompt_len < 4 or output_len < 4:
+
+        if prompt_len < 1 or output_len < 1:
             # Prune too short sequences.
             continue
-        if prompt_len > 1024 or (
-            prompt_len + output_len > 2048 and fixed_output_len is None
-        ):
+
+        if context_len and prompt_len + output_len > context_len:
             # Prune too long sequences.
             continue
+
         filtered_dataset.append((prompt, prompt_len, output_len))
 
     print(f"#Input tokens: {np.sum([x[1] for x in filtered_dataset])}")
@@ -706,8 +709,8 @@ def get_gen_prefix_cache_path(args, tokenizer):
 
     # Create a unique cache filename based on the generation parameters
     cache_key = (
-        f"gen_prefix_{args.gen_num_groups}_{args.gen_prompts_per_group}_"
-        f"{args.gen_system_prompt_len}_{args.gen_question_len}_{args.gen_output_len}_"
+        f"gen_shared_prefix_{args.gsp_num_groups}_{args.gsp_prompts_per_group}_"
+        f"{args.gsp_system_prompt_len}_{args.gsp_question_len}_{args.gsp_output_len}_"
         f"{tokenizer.__class__.__name__}.pkl"
     )
     return cache_dir / cache_key
@@ -1374,6 +1377,12 @@ if __name__ == "__main__":
         default=None,
         help="Output length for each request. Overrides the output length from the ShareGPT dataset.",
     )
+    parser.add_argument(
+        "--sharegpt-context-len",
+        type=int,
+        default=None,
+        help="The context length of the model for the ShareGPT dataset. Requests longer than the context length will be dropped.",
+    )
     parser.add_argument(
         "--random-input-len",
         type=int,
@@ -1453,49 +1462,49 @@ if __name__ == "__main__":
         help="Append given JSON object to the request payload. You can use this to specify"
         "additional generate params like sampling params.",
     )
+    parser.add_argument(
+        "--profile",
+        action="store_true",
+        help="Use Torch Profiler. The endpoint must be launched with "
+        "SGLANG_TORCH_PROFILER_DIR to enable profiler.",
+    )
+    parser.add_argument(
+        "--lora-name",
+        type=str,
+        default=None,
+        help="The name of LoRA adapter",
+    )
 
     group = parser.add_argument_group("generated-shared-prefix dataset arguments")
     group.add_argument(
-        "--gen-num-groups",
+        "--gsp-num-groups",
         type=int,
         default=64,
         help="Number of system prompt groups for generated-shared-prefix dataset",
     )
     group.add_argument(
-        "--gen-prompts-per-group",
+        "--gsp-prompts-per-group",
         type=int,
         default=16,
         help="Number of prompts per system prompt group for generated-shared-prefix dataset",
     )
     group.add_argument(
-        "--gen-system-prompt-len",
+        "--gsp-system-prompt-len",
         type=int,
         default=2048,
         help="Target length in tokens for system prompts in generated-shared-prefix dataset",
     )
     group.add_argument(
-        "--gen-question-len",
+        "--gsp-question-len",
         type=int,
         default=128,
         help="Target length in tokens for questions in generated-shared-prefix dataset",
     )
     group.add_argument(
-        "--gen-output-len",
+        "--gsp-output-len",
         type=int,
         default=256,
         help="Target length in tokens for outputs in generated-shared-prefix dataset",
     )
-    parser.add_argument(
-        "--profile",
-        action="store_true",
-        help="Use Torch Profiler. The endpoint must be launched with "
-        "SGLANG_TORCH_PROFILER_DIR to enable profiler.",
-    )
-    parser.add_argument(
-        "--lora-name",
-        type=str,
-        default=None,
-        help="The name of LoRA adapter",
-    )
     args = parser.parse_args()
     run_benchmark(args)

python/sglang/srt/managers/io_struct.py

@@ -59,6 +59,9 @@ class GenerateReqInput:
     return_text_in_logprobs: bool = False
     # Whether to stream output.
     stream: bool = False
+    # Whether to log metrics for this request (e.g. health_generate calls do not log metrics)
+    log_metrics: bool = True
+
     # The modalities of the image data [image, multi-images, video]
     modalities: Optional[List[str]] = None
     # LoRA related
@@ -196,6 +199,7 @@ class GenerateReqInput:
             top_logprobs_num=self.top_logprobs_num[i],
             return_text_in_logprobs=self.return_text_in_logprobs,
             stream=self.stream,
+            log_metrics=self.log_metrics,
             modalities=self.modalities[i] if self.modalities else None,
             lora_path=self.lora_path[i] if self.lora_path is not None else None,
         )
@@ -243,6 +247,8 @@ class EmbeddingReqInput:
     sampling_params: Union[List[Dict], Dict] = None
     # Dummy input embeds for compatibility
     input_embeds: Optional[Union[List[List[List[float]]], List[List[float]]]] = None
+    # Whether to log metrics for this request (e.g. health_generate calls do not log metrics)
+    log_metrics: bool = True
 
     def normalize_batch_and_arguments(self):
         if (self.text is None and self.input_ids is None) or (

python/sglang/srt/managers/scheduler.py

@@ -631,7 +631,8 @@ class Scheduler:
         if len(req.origin_input_ids) > self.max_req_input_len:
             logger.warning(
                 "Request length is longer than the KV cache pool size or "
-                "the max context length. Truncated!!!"
+                "the max context length. Truncated. "
+                f"{len(req.origin_input_ids)=}, {self.max_req_input_len=}."
             )
             req.origin_input_ids = req.origin_input_ids[: self.max_req_input_len]
 

python/sglang/srt/managers/tokenizer_manager.py

@@ -79,6 +79,7 @@ from sglang.srt.utils import (
     get_zmq_socket,
     kill_process_tree,
 )
+from sglang.utils import get_exception_traceback
 
 asyncio.set_event_loop_policy(uvloop.EventLoopPolicy())
 
@@ -640,7 +641,9 @@ class TokenizerManager:
 
         self.to_create_loop = False
         loop = asyncio.get_event_loop()
-        self.asyncio_tasks.add(loop.create_task(self.handle_loop()))
+        self.asyncio_tasks.add(
+            loop.create_task(print_exception_wrapper(self.handle_loop))
+        )
 
         # We cannot add signal handler when the tokenizer manager is not in
         # the main thread due to the CPython limitation.
@@ -653,7 +656,9 @@ class TokenizerManager:
                 "not in the main thread. This disables graceful shutdown of the "
                 "tokenizer manager when SIGTERM is received."
             )
-        self.asyncio_tasks.add(loop.create_task(self.sigterm_watchdog()))
+        self.asyncio_tasks.add(
+            loop.create_task(print_exception_wrapper(self.sigterm_watchdog))
+        )
 
     async def sigterm_watchdog(self):
         while not self.gracefully_exit:
@@ -738,9 +743,13 @@ class TokenizerManager:
                     state.finished = recv_obj.finished_reasons[i] is not None
                     state.event.set()
 
-                    if self.enable_metrics:
+                    if self.enable_metrics and state.obj.log_metrics:
                         self.collect_metrics(state, recv_obj, i)
-                    if self.dump_requests_folder and state.finished:
+                    if (
+                        self.dump_requests_folder
+                        and state.finished
+                        and state.obj.log_metrics
+                    ):
                         self.dump_requests(state, out_dict)
             elif isinstance(recv_obj, OpenSessionReqOutput):
                 self.session_futures[recv_obj.session_id].set_result(
@@ -887,20 +896,38 @@ class TokenizerManager:
         )
 
         if len(self.dump_request_list) >= self.dump_requests_threshold:
+            filename = os.path.join(
+                self.dump_requests_folder,
+                datetime.now().strftime("%Y-%m-%d_%H-%M-%S") + ".pkl",
+            )
+            logger.info(f"Dump {len(self.dump_request_list)} requests to {filename}")
+
             to_dump = self.dump_request_list
             self.dump_request_list = []
 
             def background_task():
                 os.makedirs(self.dump_requests_folder, exist_ok=True)
-                current_time = datetime.now()
-                filename = current_time.strftime("%Y-%m-%d_%H-%M-%S") + ".pkl"
-                with open(os.path.join(self.dump_requests_folder, filename), "wb") as f:
+                with open(filename, "wb") as f:
                     pickle.dump(to_dump, f)
 
             # Schedule the task to run in the background without awaiting it
             asyncio.create_task(asyncio.to_thread(background_task))
 
 
+async def print_exception_wrapper(func):
+    """
+    Sometimes an asyncio function does not print exception.
+    We do another wrapper to handle the exception.
+    """
+    try:
+        await func()
+    except Exception:
+        traceback = get_exception_traceback()
+        logger.error(f"TokenizerManager hit an exception: {traceback}")
+        kill_process_tree(os.getpid(), include_parent=True)
+        sys.exit(1)
+
+
 class SignalHandler:
     def __init__(self, tokenizer_manager):
         self.tokenizer_manager = tokenizer_manager

python/sglang/srt/server.py

@@ -135,9 +135,13 @@ async def health_generate(request: Request) -> Response:
     sampling_params = {"max_new_tokens": 1, "temperature": 0.7}
 
     if tokenizer_manager.is_generation:
-        gri = GenerateReqInput(input_ids=[0], sampling_params=sampling_params)
+        gri = GenerateReqInput(
+            input_ids=[0], sampling_params=sampling_params, log_metrics=False
+        )
     else:
-        gri = EmbeddingReqInput(input_ids=[0], sampling_params=sampling_params)
+        gri = EmbeddingReqInput(
+            input_ids=[0], sampling_params=sampling_params, log_metrics=False
+        )
 
     try:
         async for _ in tokenizer_manager.generate_request(gri, request):

python/sglang/test/test_utils.py

@@ -560,6 +560,7 @@ def run_bench_serving(
         tokenizer=tokenizer,
         num_prompts=num_prompts,
         sharegpt_output_len=None,
+        sharegpt_context_len=None,
         random_input_len=random_input_len,
         random_output_len=random_output_len,
         random_range_ratio=0.0,

test/srt/test_moe_ep.py

@@ -44,7 +44,7 @@ class TestEpMoE(unittest.TestCase):
         )
 
         metrics = run_eval(args)
-        assert metrics["score"] >= 0.5
+        self.assertGreater(metrics["score"], 0.5)
 
     def test_mgsm_en(self):
         args = SimpleNamespace(
@@ -56,7 +56,7 @@ class TestEpMoE(unittest.TestCase):
         )
 
         metrics = run_eval(args)
-        assert metrics["score"] >= 0.8
+        self.assertGreater(metrics["score"], 0.8)
 
 
 class TestEpMoEFP8(unittest.TestCase):