fix：更新 Marlin W16A16 MoE 中支持的断言消息

vllm/model_executor/layers/fused_moe/fuse_moe_w16a16_marlin.py

@@ -240,7 +240,7 @@ def fused_experts_impl_w16a16_marlin(hidden_states: torch.Tensor,
     assert hidden_states.dtype in [torch.bfloat16,torch.float16]
     compute_type = hidden_states.dtype
     assert use_lightop, (
-        "only BW and set LMSLIM_USE_LIGHTOP=1 support Marlin W16A16 MoE")
+        "only BW and set VLLM_USE_LIGHTOP=1 support Marlin W16A16 MoE")
 
     num_tokens, K = hidden_states.shape
     E, twoN, K_w1 = w1.shape