[PerfFix] Avoid separate thread for MP executor shm spin (take 2) (#28319)

Signed-off-by: Nick Hill <nhill@redhat.com>
2025-11-07 14:11:03 -08:00
parent da786e339e
commit 67a2da890e
9 changed files with 156 additions and 131 deletions
--- a/vllm/v1/worker/gpu_worker.py
+++ b/vllm/v1/worker/gpu_worker.py
@@ -524,7 +524,7 @@ class Worker(WorkerBase):

    @torch.inference_mode()
    def sample_tokens(
-        self, grammar_output: "GrammarOutput"
+        self, grammar_output: "GrammarOutput | None"
    ) -> ModelRunnerOutput | AsyncModelRunnerOutput:
        return self.model_runner.sample_tokens(grammar_output)