Predictions result is empty when run within shared checkpoint

I got warning about Could not find trained model in mode_dir: bert_model_ouput and the predictions is empty. I run it with python3. Does any one here got the same issue?

`File "run_nq.py", line 1472, in main
[r._asdict() for r in all_results])
File "run_nq.py", line 1308, in compute_pred_dict
summary = compute_predictions(e)
File "run_nq.py", line 1240, in compute_predictions
score, summary, start_span, end_span = sorted(predictions, reverse=True)[0]
IndexError: list index out of range
`