xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-24 18:55:37 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Bug] R1 Accuracy: Fix routed_scaling_factor Double Mul Issue (#24119)

Browse Source 
Signed-off-by: yewentao256 <zhyanwentao@126.com>
This commit is contained in:
Wentao Ye 2025-09-02 18:22:30 -04:00 committed by GitHub
parent 457e471971
commit 930a24144c
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 2 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
vllm/model_executor/models/deepseek_v2.py
View File

@ -160,7 +160,8 @@ class DeepseekV2MoE(nn.Module):
topk_group=config.topk_group,
prefix=f"{prefix}.experts",
scoring_func=config.scoring_func,
routed_scaling_factor=self.routed_scaling_factor,
# we do scaling outside, set factor to 1.0 to avoid double mul
routed_scaling_factor=1.0,
e_score_correction_bias=self.gate.e_score_correction_bias,
enable_eplb=self.enable_eplb,
num_redundant_experts=self.n_redundant_experts)