Trying attention method

training/train_at.py

@@ -139,7 +139,8 @@ class ShepherdAttentionExtractor(BaseFeaturesExtractor):
             dropout=0.0, batch_first=True,
         )
         self.transformer = nn.TransformerEncoder(encoder_layer,
-                                                 num_layers=n_layers)
+                                                 num_layers=n_layers,
+                                                 enable_nested_tensor=False)
 
     def forward(self, obs: torch.Tensor) -> torch.Tensor:
         B = obs.shape[0]