minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 13 Packages Projects Releases Wiki Activity

[Spec] Enable draft extend cuda graph for DeepSeek-V4 attention backend (#30853)

Browse Source
This commit is contained in:
Liangsheng Yin
2026-07-11 04:29:49 -05:00
committed by GitHub
parent 79f096d43b
commit 7bac9c8cdb
1 changed files with 5 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/speculative/eagle_worker_v2.py
+5
View File
@@ -471,6 +471,11 @@ class EagleDraftWorker(EagleDraftWorkerBase):
) )
graph_supported_backend_types.append(DeepseekSparseAttnBackend) graph_supported_backend_types.append(DeepseekSparseAttnBackend)
from sglang.srt.layers.attention.deepseek_v4_backend import (
DeepseekV4AttnBackend,
)
graph_supported_backend_types.append(DeepseekV4AttnBackend)
graph_supported_backend = isinstance( graph_supported_backend = isinstance(
self.draft_extend_attn_backend, self.draft_extend_attn_backend,