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

a tiny fix for support deepseek bf16 weights (#12313)

Browse Source 
Co-authored-by: gaochang <gaochang@U-19PX2WQ1-0350.local>
This commit is contained in:
Gao016
2025-10-29 11:46:44 +08:00
committed by GitHub
parent 587deb15a7
commit 0297773a2f
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/models/deepseek_v2.py
View File

@@ -2998,7 +2998,7 @@ class DeepseekV2ForCausalLM(nn.Module):
disable_reason = "Only Deepseek V3/R1 on NV-platform with capability >= 80 can use shared experts fusion optimization."
elif get_moe_expert_parallel_world_size() > 1:
disable_reason = "Deepseek V3/R1 can not use shared experts fusion optimization under expert parallelism."
elif self.quant_config.get_name() == "w4afp8":
elif self.quant_config and self.quant_config.get_name() == "w4afp8":
disable_reason = "Deepseek V3/R1 W4AFP8 model uses different quant method for routed experts and shared experts."
if disable_reason is not None: