minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

[Fix] Make the DSpark draft num_token_non_padded host-to-device copy non-blocking (#34782)

Browse Source
This commit is contained in:
Liangsheng Yin
2026-08-13 16:10:54 -07:00
committed by GitHub
parent abdef3c38e
commit 151a314829
1 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/speculative/dspark_components/dspark_draft.py
+2 -2
View File
@@ -355,8 +355,8 @@ class DraftBlockProposer:
spec_algorithm=SpeculativeAlgorithm.DSPARK, spec_algorithm=SpeculativeAlgorithm.DSPARK,
spec_info=self._draft_block_spec_info, spec_info=self._draft_block_spec_info,
capture_hidden_mode=CaptureHiddenMode.NULL, capture_hidden_mode=CaptureHiddenMode.NULL,
num_token_non_padded=torch.tensor( num_token_non_padded=torch.tensor(draft_num_tokens, dtype=torch.int32).to(
draft_num_tokens, dtype=torch.int32, device=device device, non_blocking=True
), ),
num_token_non_padded_cpu=draft_num_tokens, num_token_non_padded_cpu=draft_num_tokens,
) )