From c8f0122acf481e8712491dcd6aebc0bd97df430c Mon Sep 17 00:00:00 2001 From: kk <43161300+kkHuang-amd@users.noreply.github.com> Date: Thu, 19 Mar 2026 17:36:12 +0800 Subject: [PATCH] Fix gpu-fault issue when run deepseek-r1 and enable dp (#20841) Co-authored-by: wunhuang --- python/sglang/srt/layers/attention/aiter_backend.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/python/sglang/srt/layers/attention/aiter_backend.py b/python/sglang/srt/layers/attention/aiter_backend.py index c12d6487e..eb7006c39 100755 --- a/python/sglang/srt/layers/attention/aiter_backend.py +++ b/python/sglang/srt/layers/attention/aiter_backend.py @@ -1952,7 +1952,7 @@ class AiterAttnBackend(AttentionBackend): raise ValueError("Invalid forward mode") 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( self, spec_info: SpecInput, cuda_graph_bs: Optional[int]