Merge pull request #602 from ROCm/deepseek_085_fixbug

 add expert mask for fused_moe when running in epmoe mode

Author: valarLip

vllm/model_executor/layers/fused_moe/rocm_aiter_fused_moe.py

@@ -249,6 +249,7 @@ def rocm_aiter_asm_moe_impl(
         w2_scale=fc2_scale,
         quant_type=quant_type,
         activation=activation,
+        expert_mask=expert_mask,
     )