xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-06-01 15:17:05 +08:00
Code Issues Packages Projects Releases Wiki Activity

[BugFix] Fix spec decode + structured outputs + preemption edge case (#30916)

Browse Source 
Signed-off-by: Nick Hill <nhill@redhat.com>
This commit is contained in:
Nick Hill 2025-12-18 12:59:55 -08:00 committed by GitHub
parent 634a14bd7d
commit b0b77c4655
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 5 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
vllm/v1/worker/gpu_model_runner.py
View File

@ -3393,9 +3393,13 @@ class GPUModelRunner(
return async_output return async_output
def take_draft_token_ids(self) -> DraftTokenIds | None: def take_draft_token_ids(self) -> DraftTokenIds | None:
if self._draft_token_ids is None: if not self.num_spec_tokens:
return None return None
req_ids = self.input_batch.req_ids req_ids = self.input_batch.req_ids
if self._draft_token_ids is None:
return DraftTokenIds(req_ids, [[] for _ in req_ids])
if isinstance(self._draft_token_ids, torch.Tensor): if isinstance(self._draft_token_ids, torch.Tensor):
draft_token_ids = self._draft_token_ids.tolist() draft_token_ids = self._draft_token_ids.tolist()
else: else: