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

Set the default attention backend for GLM-4.5v to fa3 (#9245)

Browse Source
This commit is contained in:
zifeitong
2025-08-17 16:34:19 -07:00
committed by GitHub
parent ff0cf51c8e
commit 84b30d9e00
1 changed files with 1 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
python/sglang/srt/utils.py
View File

@@ -2345,6 +2345,7 @@ def is_fa3_default_architecture(hf_config):
"Qwen3ForCausalLM", "Qwen3ForCausalLM",
"Qwen3MoeForCausalLM", "Qwen3MoeForCausalLM",
"Glm4MoeForCausalLM", "Glm4MoeForCausalLM",
"Glm4vMoeForConditionalGeneration",
"Step3VLForConditionalGeneration", "Step3VLForConditionalGeneration",
} }
return architectures[0] in default_archs return architectures[0] in default_archs