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

Fix gpu-fault issue when run deepseek-r1 and enable dp (#20841)

Browse Source 
Co-authored-by: wunhuang <wunhuang@amd.com>
This commit is contained in:
kk
2026-03-19 02:36:12 -07:00
committed by GitHub
co-authored by wunhuang
parent cb8105fe28
commit c8f0122acf
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/layers/attention/aiter_backend.py
+1 -1
View File
@@ -1952,7 +1952,7 @@ class AiterAttnBackend(AttentionBackend):
raise ValueError("Invalid forward mode") raise ValueError("Invalid forward mode")
def get_cuda_graph_seq_len_fill_value(self): def get_cuda_graph_seq_len_fill_value(self):
return 1 return 1 if self.num_draft_tokens is None else self.num_draft_tokens
def update_verify_buffers_to_fill_after_draft( def update_verify_buffers_to_fill_after_draft(
self, spec_info: SpecInput, cuda_graph_bs: Optional[int] self, spec_info: SpecInput, cuda_graph_bs: Optional[int]