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

Signed-off-by: tjtanaa <tunjian.tan@embeddedllm.com>
Signed-off-by: Gregory Shtrasberg <Gregory.Shtrasberg@amd.com>

Author: tjtanaa

vllm/model_executor/layers/fused_moe/layer.py

@@ -503,7 +503,6 @@ def forward_cuda(
             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,