diff --git a/python/sglang/srt/models/qwen2_moe.py b/python/sglang/srt/models/qwen2_moe.py index bb1719595e..72ceef49c1 100644 --- a/python/sglang/srt/models/qwen2_moe.py +++ b/python/sglang/srt/models/qwen2_moe.py @@ -372,6 +372,7 @@ class Qwen2MoeSparseMoeBlock(nn.Module): or get_moe_a2a_backend().is_flashinfer() or get_moe_a2a_backend().is_flashinfer_megamoe() or get_moe_a2a_backend().is_megamoe() + or get_moe_a2a_backend().is_ascend_fuseep() ) else {} ), @@ -393,6 +394,7 @@ class Qwen2MoeSparseMoeBlock(nn.Module): get_moe_a2a_backend().is_deepep() or get_moe_a2a_backend().is_deepep_v2() or get_moe_a2a_backend().is_mori() + or get_moe_a2a_backend().is_ascend_fuseep() ): self.num_experts = ( config.num_experts + get_exec().moe.ep_num_redundant_experts @@ -827,6 +829,7 @@ class Qwen2MoeSparseMoeBlock(nn.Module): get_moe_a2a_backend().is_deepep() or get_moe_a2a_backend().is_deepep_v2() or get_moe_a2a_backend().is_mori() + or get_moe_a2a_backend().is_ascend_fuseep() ): return self._forward_deepep(hidden_states, forward_batch)