Unpacking a local model about espnet_model_zoo

[kamo-naoyuki]

If you'll use local file, you don't need to archive file to zip, but directly give it to the API

[Daniel-asr]

I tried following the ESPnet zoo documentation:

The zip file I used was the one generated by the asr.sh script. Could you please expand your sentence "you don't need to archive file to zip"? I don't understand what you mean by that

Thanks

[kamo-naoyuki]

It should work if you correctly did, thus you made some mistakes definitely.
I mean you can also input file to Speech2Text directly. This is straightforward.

[Daniel-asr]

I tried to do that now by following the example in the asr_inference.py file: speech2text = Speech2Text("asr_config.yml", "asr.pth")
But got the same error:
RuntimeError: /home/ubuntu/espnet/egs/librispeech/asr1/<path_to_my_models_exp>/valid.acc.ave_10best.pth is a zip archive (did you mean to use torch.jit.load()?)

[kamo-naoyuki]

Why didn't you read the error? The error message is really clear.

https://github.com/pytorch/pytorch/releases/tag/v1.6.0:

The model format was changed to zip from pytorch=1.6, I guess you used pytorch>=1.6 when saving the modle, but you are using pytorch<1.6.

[Daniel-asr]

I used the same pytorch version (1.0.1.post2) for training (& saving the model) and for running zoo.

I avoided changing my environment as much as possible due to mismatch with other packages and scripts I have but after upgrading pytorch to 1.6.0 (and tensorboard to the latest version, as required) and after a few small fixes it worked.