[Scheduler] Add --max-num-active-seqs to cap RUNNING admission (#56758)

Signed-off-by: Li <chuali@amd.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
This commit is contained in:
Chuan (Richard) Li
2026-09-17 10:22:12 +08:00
committed by GitHub
co-authored by Cursor
parent b1f9ebc0c3
commit f730a93d2b
5 changed files with 51 additions and 1 deletions
+6
View File
@@ -556,6 +556,7 @@ class EngineArgs:
max_num_scheduled_tokens: int | None = None
long_prefill_token_threshold: int = SchedulerConfig.long_prefill_token_threshold
max_num_seqs: int | None = None
max_num_active_seqs: int | None = SchedulerConfig.max_num_active_seqs
max_num_queued_reqs: int | None = None
max_num_queued_tokens: int | None = None
max_logprobs: int = ModelConfig.max_logprobs
@@ -1589,6 +1590,10 @@ class EngineArgs:
"default": None,
},
)
scheduler_group.add_argument(
"--max-num-active-seqs",
**scheduler_kwargs["max_num_active_seqs"],
)
scheduler_group.add_argument(
"--max-num-queued-reqs", **scheduler_kwargs["max_num_queued_reqs"]
)
@@ -2421,6 +2426,7 @@ class EngineArgs:
max_num_batched_tokens=self.max_num_batched_tokens,
max_num_scheduled_tokens=self.max_num_scheduled_tokens,
max_num_seqs=self.max_num_seqs,
max_num_active_seqs=self.max_num_active_seqs,
max_num_queued_reqs=self.max_num_queued_reqs,
max_num_queued_tokens=self.max_num_queued_tokens,
max_model_len=model_config.max_model_len,