fix

Signed-off-by: David9857 <985700846@qq.com>

Author: David9857

vllm_ascend/models/qwen3.py

@@ -37,6 +37,7 @@ def pad(tensor, x):
         return F.pad(tensor, (0, 0, 0, pad_size)), pad_size
     return tensor, pad_size
 
+
 def unpad(tensor, pad_size):
     if pad_size > 0:
         return tensor[:-pad_size, :]
@@ -253,7 +254,7 @@ def pre_attention_process(self, hidden_states, residual, pad_size=0):
         hidden_states = tensor_model_parallel_all_gather(hidden_states, 0)
         hidden_states = unpad(hidden_states, pad_size)
         return hidden_states, residual
-    
+
     def pre_mlp_process(self, hidden_states, residual, pad_size=0):
         token_num = hidden_states.size(0)
         if token_num != residual.size(0):
@@ -298,7 +299,7 @@ def forward(self,
                 hidden_states, residual)
         hidden_states, pad_size = self.mlp(hidden_states)
         return hidden_states, residual, pad_size
-    
+
 
 ALL_DECODER_LAYER_TYPES = {
     "attention": CustomQwen3DecoderLayer,