add breakpoint:

modeling_flexbert.py

@@ -1705,6 +1705,7 @@ class FlexBertForCausalLM(FlexBertPreTrainedModel):
         #     # reshape to batch size
         #     logits = logits.view(-1, self.vocab_size)
         #     # NOTE: error from here above
+        breakpoint()
 
         if self.pad_logits:
             print(f"Padding logits: {logits.shape}")