frozenleaves/vllm
1
0
Fork 0
mirror of https://github.com/vllm-project/vllm.git synced 2025-10-20 14:53:52 +08:00
Code Issues Packages Projects Releases Wiki Activity

[BugFix] Fix DBO hang (#25625)

Browse Source 
Signed-off-by: Lucas Wilkinson <lwilkins@redhat.com>
This commit is contained in:
Lucas Wilkinson
2025-09-25 13:04:48 -04:00
committed by GitHub
parent 916bd9204d
commit 13cc7f5370
1 changed files with 2 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
vllm/v1/worker/gpu_ubatch_wrapper.py
View File

@ -366,7 +366,8 @@ class UBatchWrapper:
cudagraph_runtime_mode=CUDAGraphMode.NONE)
with self.sm_control:
return self._capture_ubatches(ubatch_metadata, self.model)
elif num_tokens in self.cudagraphs:
elif num_tokens in self.cudagraphs \
and cudagraph_runtime_mode is CUDAGraphMode.FULL:
cudagraph_metadata = self.cudagraphs[num_tokens]
cudagraph_metadata.cudagraph.replay()
return cudagraph_metadata.outputs