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

[AMD] Enable draft_extend CUDA graph for HIP DSA backend (#35672)

Browse Source
This commit is contained in:
jiaryang
2026-08-25 02:18:08 -07:00
committed by GitHub
parent 8f096b853a
commit 3bc1c580c6
1 changed files with 8 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/speculative/eagle_worker_v2.py
+8 -3
View File
@@ -416,10 +416,15 @@ class EagleDraftWorker(EagleDraftWorkerBase):
from sglang.srt.layers.attention.deepseek_v4_backend_hip_radix import ( from sglang.srt.layers.attention.deepseek_v4_backend_hip_radix import (
DeepseekV4HipRadixBackend, DeepseekV4HipRadixBackend,
) )
from sglang.srt.layers.attention.dsa_backend import (
DeepseekSparseAttnBackend,
)
supports_hip_draft_extend_graph = isinstance( supports_hip_draft_extend_graph = (
self.draft_attn_backend, AiterMultiStepDraftBackend isinstance(self.draft_attn_backend, AiterMultiStepDraftBackend)
) or isinstance(self.draft_extend_attn_backend, DeepseekV4HipRadixBackend) or isinstance(self.draft_extend_attn_backend, DeepseekV4HipRadixBackend)
or isinstance(self.draft_extend_attn_backend, DeepseekSparseAttnBackend)
)
graph_supported_backend_types = [ graph_supported_backend_types = [
TritonAttnBackend, TritonAttnBackend,