Fix: remove only added forward hooks

bdpy/dl/torch/torch.py

@@ -95,10 +95,17 @@ def run(self, x: _tensor_t) -> Dict[str, _tensor_t]:
         return features
 
     def __del__(self):
-        '''Remove forward hooks from the encoder.'''
+        '''
+        Remove forward hooks for the FeatureExtractor while keeping
+        other forward hooks in the model.
+        '''
         for layer in self.__layers:
+            if self.__layer_map is not None:
+                layer = self.__layer_map[layer]
             layer_object = models._parse_layer_name(self._encoder, layer)
-            layer_object._forward_hooks = OrderedDict()
+            for key, hook in layer_object._forward_hooks.items():
+                if hook == self._extractor:
+                    del layer_object._forward_hooks[key]
 
 
 class FeatureExtractorHandle(object):