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

[Fix] Fix --disable-flashinfer (#1389)

Browse Source
This commit is contained in:
Lianmin Zheng
2024-09-11 04:36:21 -07:00
committed by GitHub
parent 224200e3c2
commit 15c75e4146
1 changed files with 2 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
python/sglang/srt/server_args.py
View File

@@ -140,11 +140,13 @@ class ServerArgs:
"The option '--disable-flashinfer' will be deprecated in the next release. " "The option '--disable-flashinfer' will be deprecated in the next release. "
"Please use '--attention-backend triton' instead." "Please use '--attention-backend triton' instead."
) )
self.attention_backend = "triton"
if self.disable_flashinfer_sampling: if self.disable_flashinfer_sampling:
logger.warning( logger.warning(
"The option '--disable-flashinfer-sampling' will be deprecated in the next release. " "The option '--disable-flashinfer-sampling' will be deprecated in the next release. "
"Please use '--sampling-backend pytorch' instead. " "Please use '--sampling-backend pytorch' instead. "
) )
self.sampling_backend = "pytorch"
# Model-specific patches # Model-specific patches
if "Alibaba-NLP/gte-Qwen2-1.5B-instruct" == self.model_path: if "Alibaba-NLP/gte-Qwen2-1.5B-instruct" == self.model_path: