Issues with score.sh on streaming transformer(mma) models

[cjw414]

Hello Hiro, after I tained streaming transformer model(mma) on Librispeech corpus, I tried to decode the results by score.sh.
It seemed to go well at first, but the decoding went wrong after few steps, with the error message as followed:

# This part is just a warning.
  0%|          | 0/2703 [00:00<?, ?it/s]/mnt/data1/jungwonchang/projects/neural_sp/neural_sp/models/modules/mocha.py:815: UserWarning: This overload of nonzero is deprecated:
  nonzero()
Consider using one of the following signatures instead:
  nonzero(*, bool as_tuple) (Triggered internally at  /opt/conda/conda-bld/pytorch_1595629427478/work/torch/csrc/utils/python_arg_parser.cpp:766.)
  boundary = alpha[b, h, 0, 0].nonzero()[:, -1].min().item()
  9%|▊         | 232/2703 [15:06<1:49:35,  2.66s/it]Original utterance num: 2703
Removed 0 empty utterances

# This part is where I get errors
Traceback (most recent call last):
  File "/home/jungwonchang/projects1/neural_sp/examples/librispeech/s5/../../../neural_sp/bin/asr/eval.py", line 247, in <module>
    main()
  File "/home/jungwonchang/projects1/neural_sp/examples/librispeech/s5/../../../neural_sp/bin/asr/eval.py", line 182, in main
    oracle=True)
  File "/mnt/data1/jungwonchang/projects/neural_sp/neural_sp/evaluators/wordpiece.py", line 85, in eval_wordpiece
    ensemble_models=models[1:] if len(models) > 1 else [])[0]
  File "/mnt/data1/jungwonchang/projects/neural_sp/neural_sp/models/seq2seq/speech2text.py", line 763, in decode
    ensmbl_eouts, ensmbl_elens, ensmbl_decs)
  File "/mnt/data1/jungwonchang/projects/neural_sp/neural_sp/models/seq2seq/decoders/transformer.py", line 896, in beam_search
    rightmost_frame = max(0, aws_last_success[0, :, 0].nonzero()[:, -1].max().item()) + 1
RuntimeError: operation does not have an identity.
  9%|▊         | 232/2703 [15:07<2:41:09,  3.91s/it]

the configuration I used was
conf/asr/mma/streaming/lc_transformer_mma_subsample8_ma4H_ca4H_w16_from4L_64_128_64.yaml

Also, I found out in the decode.log that the streamable feature for my model was False

2021-01-11 14:29:08,749 neural_sp.models.seq2seq.decoders.transformer line:888 INFO: streamable: False
2021-01-11 14:29:08,749 neural_sp.models.seq2seq.decoders.transformer line:889 INFO: streaming failed point: 1

Any idea or advice on this issue?