fix(policies): crop losses based on the action dof (#3133)

Co-authored-by: Chenning Yu <rainorangelemon@gmail.com>

src/lerobot/policies/smolvla/modeling_smolvla.py

@@ -377,6 +377,8 @@ class SmolVLAPolicy(PreTrainedPolicy):
         actions_is_pad = batch.get("action_is_pad")
         loss_dict = {}
         losses = self.model.forward(images, img_masks, lang_tokens, lang_masks, state, actions, noise, time)
+        original_action_dim = self.config.action_feature.shape[0]
+        losses = losses[:, :, :original_action_dim]
         loss_dict["losses_after_forward"] = losses.clone().mean().item()
 
         if actions_is_pad is not None: