SERAC Edit Limits

[shariqahn]

I got the following error when trying to train SERAC:

Traceback (most recent call last):
  File "/EasyEdit/serac_train.py", line 16, in <module>
    trainer.run()
  File "/EasyEdit/easyeditor/trainer/BaseTrainer.py", line 195, in run
    train_info = self.train_step(batch)
  File "/EasyEdit/easyeditor/trainer/EditTrainer.py", line 141, in train_step
    l_total, l_edit, l_loc, l_base, info_dict = self.edit_step(
  File "/EasyEdit/easyeditor/trainer/EditTrainer.py", line 54, in edit_step
    l_edit = self.model.edit_loss_fn(
  File "/EasyEdit/easyeditor/trainer/algs/editable_model.py", line 24, in _edit_loss_fn
    return masked_log_probs(config, pred, targ, shift=True, **kwargs)
  File "/EasyEdit/easyeditor/trainer/losses.py", line 135, in masked_log_probs
    return multiclass_log_probs(config, pred, targ, shift=shift, exact_match=exact_match, **kwargs)
  File "/EasyEdit/easyeditor/trainer/losses.py", line 70, in multiclass_log_probs
    unmasked_log_probs = pred.log_softmax(-1).gather(-1, targ.unsqueeze(-1)).squeeze(-1)
RuntimeError: Size does not match at dimension 1 expected index [10, 32, 1] to be smaller than self [10, 31, 32000] apart from dimension 2

I have run the training successfully on a different dataset that had shorter strings for the new target. Is there a limit to how long the new target should be? If so, do you have suggestions on how to handle this? For example, would it be okay to provide a simplified version of the edits that are within the limit for training but provide the full string for edits?