EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

fix capture_bs when speculative decoding enabled (#10730)

Browse Source
This commit is contained in:
feng397
2025-09-22 15:43:12 +08:00
committed by GitHub
parent 9d7e82a0ab
commit 944f1ea0ec
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
python/sglang/srt/model_executor/cuda_graph_runner.py
View File

@@ -179,7 +179,7 @@ def get_batch_sizes_to_capture(model_runner: ModelRunner):
capture_bs = ( capture_bs = (
list(range(1, 9)) list(range(1, 9))
+ list(range(10, 33, 2)) + list(range(10, 33, 2))
+ list(range(40, 64, 8)) + list(range(40, 65, 8))
+ list(range(80, 161, 16)) + list(range(80, 161, 16))
) )