chenchenghao/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 11 Packages Projects Releases Wiki Activity

HiCache Fix (#8288)

Browse Source 
Co-authored-by: pansicheng <sicheng.pan.chn@gmail.com>
This commit is contained in:
Zhiqiang Xie
2025-07-23 01:51:32 -07:00
committed by GitHub
parent ce86e201df
commit f39037fffb
2 changed files with 5 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
python/sglang/srt/model_executor/model_runner.py
View File

@@ -378,6 +378,7 @@ class ModelRunner:
is_hopper_with_cuda_12_3()
and is_no_spec_infer_or_topk_one(server_args)
and is_fa3_default_architecture(self.model_config.hf_config)
and (not server_args.enable_hierarchical_cache)
):
server_args.attention_backend = "fa3"
elif _is_hip:
@@ -390,7 +391,9 @@ class ModelRunner:
)
else:
# MLA architecture
if is_hopper_with_cuda_12_3():
if is_hopper_with_cuda_12_3() and (
not server_args.enable_hierarchical_cache
):
server_args.attention_backend = "fa3"
elif is_sm100_supported():
server_args.attention_backend = "flashinfer"