Merge branch 'v0.9.2-dev' into v0.9.2-dev-ds

vllm/model_executor/layers/fused_moe/fused_moe.py

@@ -1771,8 +1771,9 @@ def fused_experts_impl(
 
         if envs.VLLM_USE_LIGHTOP and not dpsk_fp16_quick: 
             from lightop import op as op
-            op.moe_sum(intermediate_cache3.view(*intermediate_cache3.size()),
-                    out_hidden_states[begin_chunk_idx:end_chunk_idx], shared_output[begin_chunk_idx:end_chunk_idx], None, routed_scaling_factor)
+            op.moe_sum(input=intermediate_cache3.view(*intermediate_cache3.size()),
+                    output=out_hidden_states[begin_chunk_idx:end_chunk_idx], bias=shared_output[begin_chunk_idx:end_chunk_idx], 
+                    expert_mask=None, num_local_tokens=None, factor=routed_scaling_factor)
         # else:
         #     ops.moe_sum(intermediate_cache3.view(*intermediate_cache3.size()),
         #                 out_hidden_states[begin_chunk_idx:end_chunk_idx])