From 6cbe448eed751824d608faf9078ef84724d621c1 Mon Sep 17 00:00:00 2001 From: weizhoublue <45163302+weizhoublue@users.noreply.github.com> Date: Mon, 25 May 2026 12:03:05 +0800 Subject: [PATCH] fix: MoE model using shared routed experts crashes on AMD GPUs (#42373) Signed-off-by: weizhou.lan@daocloud.io --- .../fused_moe/router/aiter_shared_routed_fused_moe_router.py | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/vllm/model_executor/layers/fused_moe/router/aiter_shared_routed_fused_moe_router.py b/vllm/model_executor/layers/fused_moe/router/aiter_shared_routed_fused_moe_router.py index 46d69cf909..8c17ac4d01 100644 --- a/vllm/model_executor/layers/fused_moe/router/aiter_shared_routed_fused_moe_router.py +++ b/vllm/model_executor/layers/fused_moe/router/aiter_shared_routed_fused_moe_router.py @@ -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, )