[Bugfix] Fix race in non-blocking num_accepted_tokens GPU->CPU copy (#36013)

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

vllm/v1/worker/gpu_model_runner.py

@@ -727,8 +727,10 @@ class GPUModelRunner(
         self.draft_token_ids_copy_stream: torch.cuda.Stream | None = None
         self.valid_sampled_token_count_cpu: torch.Tensor | None = None
         self.draft_token_ids_cpu: torch.Tensor | None = None
+        self.num_accepted_tokens_event: torch.Event | None = None
         if self.num_spec_tokens:
             self.draft_token_ids_event = torch.Event()
+            self.num_accepted_tokens_event = torch.Event()
             self.draft_token_ids_copy_stream = torch.cuda.Stream()
             self.draft_token_ids_cpu = torch.empty(
                 (self.max_num_reqs, self.num_spec_tokens),
@@ -1229,6 +1231,8 @@ class GPUModelRunner(
             self.input_batch.num_accepted_tokens_cpu_tensor[:num_reqs].copy_(
                 self.num_accepted_tokens.gpu[:num_reqs], non_blocking=True
             )
+            assert self.num_accepted_tokens_event is not None
+            self.num_accepted_tokens_event.record()
 
     def _update_streaming_request(
         self, req_id: str, new_req_data: NewRequestData
@@ -1773,6 +1777,8 @@ class GPUModelRunner(
             max_seq_len = self.seq_lens.np[:num_reqs].max().item()
 
         if use_spec_decode:
+            if self.num_accepted_tokens_event is not None:
+                self.num_accepted_tokens_event.synchronize()
             self.num_accepted_tokens.np[:num_reqs] = (
                 self.input_batch.num_accepted_tokens_cpu[:num_reqs]
             )