fix linter

Author: charlifu

vllm/model_executor/models/mixtral.py

@@ -25,6 +25,7 @@
 from typing import Iterable, List, Optional, Tuple
 
 import torch
+import torch.nn.functional as F
 from torch import nn
 from transformers import MixtralConfig
 
@@ -183,12 +184,12 @@ def process_weights_after_loading(self):
         # Fp8 is the only case where we need to process after loading.
         if not self.use_fp8:
             if os.getenv("VLLM_MOE_PADDING", "1") == "1":
-                self.w13_weight = nn.Parameter(
-                    torch.nn.functional.pad(self.w13_weight.data, (0, 128),
-                                            "constant", 0))
-                self.w2_weight = nn.Parameter(
-                    torch.nn.functional.pad(self.w2_weight.data, (0, 128),
-                                            "constant", 0))
+                self.w13_weight = nn.Parameter(F.pad(self.w13_weight.data,
+                                                     (0, 128), "constant", 0),
+                                               requires_grad=False)
+                self.w2_weight = nn.Parameter(F.pad(self.w2_weight.data,
+                                                    (0, 128), "constant", 0),
+                                              requires_grad=False)
             return
 
         # If checkpoint is fp16, quantize here.