xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-04-01 23:47:07 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Bugfix] [ROCm]: Remove assertion logic when using AITER fused moe in unquantizedMethod to reenable LLama4 BF16 (#18205)

Browse Source 
Signed-off-by: tjtanaa <tunjian.tan@embeddedllm.com>
This commit is contained in:
TJian 2025-05-16 00:53:18 +08:00 committed by GitHub
parent fadb8d5c2d
commit 92540529c0
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 0 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
vllm/model_executor/layers/fused_moe/layer.py
View File

@ -503,7 +503,6 @@ class UnquantizedFusedMoEMethod(FusedMoEMethodBase, CustomOp):
indices_type=torch.uint32 if self.moe.use_pplx_kernels else None)
if self.rocm_aiter_moe_enabled:
assert not apply_router_weight_on_input
assert expert_map is None
return self.rocm_aiter_fused_experts(
hidden_states=x,