The Bug in data preparation for How2 Summarization

[teinhonglo]

To Reproduce
Steps to reproduce the behavior:

1. move to a recipe directory, e.g., cd egs2/how2_2000h/summ1
2. local/run_asr.sh --asr_tag asr_pretrain

Error logs

2022-06-19T15:12:30 (asr.sh:252:main) ./asr.sh --lang en --feats_type extracted --token_type bpe --nbpe 1000 --nlsyms_txt data/nlsyms --bpe_nlsyms [hes] --use_lm false --asr_config conf/train_asr_conformer_lf.yaml --inference_config conf/decode_asr.yaml --train_set tr_2000h_utt --valid_set cv05_utt --test_sets dev5_test_utt --bpe_train_text data/tr_2000h_utt/text --asr_tag asr_pretrain
2022-06-19T15:12:34 (asr.sh:443:main) Stage 1: Data preparation for data/tr_2000h_utt, data/cv05_utt, etc.
2022-06-19T15:12:38 (data.sh:23:main) local/data.sh
2022-06-19T15:12:42 (data.sh:36:main) stage 0: Data download
/home/teinhonglo/espnets/espnet/tools/anaconda/envs/espnet/lib/python3.8/site-packages/gdown/cli.py:127: FutureWarning: Option `--id` was deprecated in version 4.3.1 and will be removed in 5.0. You don't need to pass it anymore to use a file ID.
  warnings.warn(
Access denied with the following error:

        Cannot retrieve the public link of the file. You may need to change
        the permission to 'Anyone with the link', or have had many accesses.

You may still be able to access the file from the browser:

         https://drive.google.com/uc?id=sharing

Besides,
The link provided by local/data.sh is not available now.

Thanks in advnace.

[sw005320]

@roshansh-cmu, can you answer it for me?
Google drive downloading is tricky.
You may think of hosting it in the other storage (e.g., huggingface)

[roshansh-cmu]

Thank you for opening this issue.

This issue might take some time to resolve due to server issues. I am investigating alternative storage and will respond here when the issue is resolved.

[roshansh-cmu]

Please request the dataset using the data release form from the How2 data repository : https://github.com/srvk/how2-dataset.

Apologies for the long delay in replying

[teinhonglo]

@roshansh-cmu

Thanks for your reply.
The problem above is solved, but I encountered a runtime error as below.
Any suggestion?

Traceback (most recent call last):
  File "/share/nas165/teinhonglo/espnets/espnet-20220922/tools/anaconda/envs/espnet/lib/python3.8/runpy.py", line 194, in _run_module_as_main
    return _run_code(code, main_globals, None,
  File "/share/nas165/teinhonglo/espnets/espnet-20220922/tools/anaconda/envs/espnet/lib/python3.8/runpy.py", line 87, in _run_code
    exec(code, run_globals)
  File "/share/nas165/teinhonglo/espnets/espnet-20220922/espnet2/bin/asr_train.py", line 23, in <module>
    main()
  File "/share/nas165/teinhonglo/espnets/espnet-20220922/espnet2/bin/asr_train.py", line 19, in main
    ASRTask.main(cmd=cmd)
  File "/share/nas165/teinhonglo/espnets/espnet-20220922/espnet2/tasks/abs_task.py", line 1019, in main
    cls.main_worker(args)
  File "/share/nas165/teinhonglo/espnets/espnet-20220922/espnet2/tasks/abs_task.py", line 1315, in main_worker
    cls.trainer.run(
  File "/share/nas165/teinhonglo/espnets/espnet-20220922/espnet2/train/trainer.py", line 282, in run
    all_steps_are_invalid = cls.train_one_epoch(
  File "/share/nas165/teinhonglo/espnets/espnet-20220922/espnet2/train/trainer.py", line 556, in train_one_epoch
    retval = model(**batch)
  File "/share/nas165/teinhonglo/espnets/espnet-20220922/tools/anaconda/envs/espnet/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1102, in _call_impl
    return forward_call(*input, **kwargs)
  File "/share/nas165/teinhonglo/espnets/espnet-20220922/espnet2/asr/espnet_model.py", line 202, in forward
    encoder_out, encoder_out_lens = self.encode(speech, speech_lengths)
  File "/share/nas165/teinhonglo/espnets/espnet-20220922/espnet2/asr/espnet_model.py", line 368, in encode
    assert encoder_out.size(1) <= encoder_out_lens.max(), (
AssertionError: (torch.Size([1, 3120, 512]), tensor(3036, device='cuda:0'))
Loading tvm binary from: /share/nas165/teinhonglo/espnets/espnet-20220922/tools/anaconda/envs/espnet/lib/python3.8/site-packages/longformer/../longformer/lib/lib_diagonaled_mm_float32_cuda.so
# Accounting: time=13 threads=1

[roshansh-cmu]

Please wait for a day or two- I am in the process of setting up a PR to fix this and other issues with the recipe. Thanks!

[roshansh-cmu]

You may refer to our PR #4805 this when merged should be used to prepare data from the downloaded dataset bz2 file.
The modification made to espnet_model.py should fix the assertion error you are facing

Thank you!