xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-21 18:55:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Bugfix] Fix Shared Expert/Zero expert code in FusedMoE.process_chunk (#25698)

Browse Source 
Signed-off-by: Sage Moore <sage@neuralmagic.com>
Co-authored-by: Robert Shaw <114415538+robertgshaw2-redhat@users.noreply.github.com>
Signed-off-by: yewentao256 <zhyanwentao@126.com>
This commit is contained in:
Sage Moore 2025-09-26 01:25:28 -07:00 committed by yewentao256
parent c761b84d5f
commit fa55373af1
1 changed files with 3 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
vllm/model_executor/layers/fused_moe/layer.py
View File

@ -1925,7 +1925,9 @@ class FusedMoE(CustomOp):
assert self.shared_experts is None or isinstance( assert self.shared_experts is None or isinstance(
final_hidden_states, tuple) final_hidden_states, tuple)
if isinstance(final_hidden_states, tuple): if self.zero_expert_num is not None and self.zero_expert_num > 0:
assert isinstance(final_hidden_states, tuple)
assert self.shared_experts is None
final_hidden_states, zero_expert_result = final_hidden_states final_hidden_states, zero_expert_result = final_hidden_states
if zero_expert_result is not None: if zero_expert_result is not None:
final_hidden_states += zero_expert_result final_hidden_states += zero_expert_result