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

Fix quantized moe checker fail for Qwen3 dense fp8 model (#13853)

Browse Source
This commit is contained in:
fzyzcjy
2025-11-25 03:16:50 +08:00
committed by GitHub
parent 9535015d05
commit 94216a9cc4
1 changed files with 5 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
python/sglang/srt/model_executor/model_runner.py
View File

@@ -597,9 +597,12 @@ class ModelRunner:
)
moe_tp_size = self.tp_size // self.moe_ep_size
moe_intermediate_size = (
self.model_config.hf_text_config.moe_intermediate_size
moe_intermediate_size = getattr(
self.model_config.hf_text_config, "moe_intermediate_size", None
)
if moe_intermediate_size is None:
return
if moe_intermediate_size % moe_tp_size != 0:
raise ValueError(
f"moe_intermediate_size {moe_intermediate_size} must be divisible by moe_tp_size ({moe_tp_size}) which is tp_size ({self.tp_size}) divided by moe_ep_size ({self.moe_ep_size})."