Debugged use_cache flag for MLP model (which doesn't take in the `u…

…se_cache` argument in its forward call)

pyvene/models/intervenable_base.py

@@ -1440,10 +1440,14 @@ def forward(
                 )
 
             # run intervened forward
-            if labels is not None:
-                counterfactual_outputs = self.model(**base, labels=labels, use_cache=use_cache)
-            else:
-                counterfactual_outputs = self.model(**base, use_cache=use_cache)
+            model_kwargs = {}
+            if labels is not None: # for training
+                model_kwargs["labels"] = labels
+            if 'use_cache' in self.model.config.to_dict(): # for transformer models
+                model_kwargs["use_cache"] = use_cache
+
+            counterfactual_outputs = self.model(**base, **model_kwargs)
+
             set_handlers_to_remove.remove()
 
             self._output_validation()