fix: MoE model using shared routed experts crashes on AMD GPUs (#42373)
Signed-off-by: weizhou.lan@daocloud.io <weizhou.lan@daocloud.io>
This commit is contained in:
@@ -71,7 +71,7 @@ class AiterSharedRoutedFusedMoERouter(BaseRouter):
|
||||
"Number of tokens mismatch"
|
||||
)
|
||||
|
||||
from vllm.model_executor.layers.fused_moe.rocm_aiter_fused_moe import (
|
||||
from vllm.model_executor.layers.fused_moe.experts.rocm_aiter_moe import (
|
||||
aiter_topK_meta_data,
|
||||
)
|
||||
|
||||
@@ -125,7 +125,7 @@ class AiterSharedRoutedFusedMoERouter(BaseRouter):
|
||||
)
|
||||
|
||||
if aiter_topK_meta_data is not None:
|
||||
from vllm.model_executor.layers.fused_moe.rocm_aiter_fused_moe import (
|
||||
from vllm.model_executor.layers.fused_moe.experts.rocm_aiter_moe import (
|
||||
inject_shared_expert_weights,
|
||||
)
|
||||
|
||||
|
||||
Reference in New Issue
Block a user