update

modeling_flexbert.py

@@ -1639,7 +1639,6 @@ class FlexBertForCausalLM(FlexBertPreTrainedModel):
         #
         # Prediction scores are only computed for masked tokens and the (bs,
         # seqlen) dimensions are flattened
-        breakpoint()
         return_dict = return_dict if return_dict is not None else self.config.use_return_dict
         if self.unpad_embeddings and (indices is None and cu_seqlens is None and max_seqlen is None):
             batch_size, seq_len = input_ids.shape[:2]