update loss.py and kws_model.py

kws/model/kws_model.py

@@ -56,6 +56,7 @@ class KWSModel(torch.nn.Module):
             x = self.preprocessing(x)
         x, _ = self.backbone(x)
         x = self.classifier(x)
+        x = torch.sigmoid(x)
         return x
 
 

kws/model/loss.py

@@ -37,7 +37,6 @@ def max_polling_loss(logits: torch.Tensor,
         (float): loss of current batch
         (float): accuracy of current batch
     '''
-    logits = torch.sigmoid(logits)
     mask = padding_mask(lengths)
     num_utts = logits.size(0)
     num_keywords = logits.size(2)