A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 9, 14:45 UTC
[Model Runner V2] Minor simplification for finish_requests (#33048)
Signed-off-by: Woosuk Kwon <woosuk@inferact.ai>
Woosuk Kwon committed
Jan 25, 2026 at 18:35 UTC
2f0d3ba745ad56052d837747cfd1fec5e8d3e31b
1 file changed
+6
-7
vllm/v1/worker/gpu/model_runner.py
+6
-7
@@ -416,13 +416,12 @@ class GPUModelRunner(LoRAModelRunnerMixin):
416
torch.cuda.synchronize()
417
418
def finish_requests(self, scheduler_output: SchedulerOutput) -> None:
419
- if scheduler_output.preempted_req_ids is not None:
420
- for req_id in scheduler_output.preempted_req_ids:
421
- self.req_states.remove_request(req_id)
422
- if self.supports_mm_inputs:
423
- self.encoder_runner.remove_request(req_id)
424
- self.prompt_logprobs_worker.remove_request(req_id)
425
- for req_id in scheduler_output.finished_req_ids:
419
+ finished_req_ids = scheduler_output.finished_req_ids
420
+ if scheduler_output.preempted_req_ids:
421
+ finished_req_ids = finished_req_ids.union(
422
+ scheduler_output.preempted_req_ids
423
+ )
424
+ for req_id in finished_req_ids:
425
self.req_states.remove_request(req_id)
426
if self.supports_mm_inputs:
427
self.encoder_runner.remove_request(req_id)