修改错误代码

vllm/_custom_ops.py

@@ -1384,7 +1384,7 @@ def ep_moe_align_block_size(topk_ids: torch.Tensor, num_experts: int,
                          experts_ids: torch.Tensor,
                          num_tokens_post_pad: torch.Tensor,
                          start_expert, end_expert) -> None:
-    torch.ops._C.ep_moe_align_block_size(topk_ids, num_experts, block_size,
+    torch.ops._moe_C.ep_moe_align_block_size(topk_ids, num_experts, block_size,
                                       sorted_token_ids, experts_ids,
                                       num_tokens_post_pad, start_expert, 
                                       end_expert)

vllm/model_executor/models/mixtral.py

@@ -545,13 +545,6 @@ class MixtralForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
             
             for layername in loaded_params:
                 weight = params_dict[layername]
-                if "lm_head.weight" in layername:
-                    lay_key_words.append("lm_head.weight")
-                    combined_words = "|".join(lay_key_words)
-                    os.environ['LM_NN'] = '1'  
-                else:
-                    os.environ['LM_NN'] = '0'
-
                 matches = re.findall(combined_words, layername)
 
                 if matches: