-
Notifications
You must be signed in to change notification settings - Fork 53
Export Finetuned wav2vec2 model #68
Export Finetuned wav2vec2 model #68
Conversation
@vasudevgupta7 please work on the failure logs too. |
I am actually unable to understand the reason for failing logs. Any help is much appreciated. |
Strange that it's failing due to In any case, @MorganR will be able to tell more. |
I would expect this to work too. I can investigate next week. |
Thanks!! Rest all suggestions are done. |
Hey Vasudev, it looks like the model asset is an archive, but it is not compressed. You can compress the model like this: $ cd $SAVED_MODEL_DIR
$ tar -czvf ../saved-model.tar.gz * That should fix the validation issue. |
@MorganR, I have changed the |
Hi @vasudevgupta7
Would you mind removing these files starting with python tools/validator.py vasudevgupta7/models/wav2vec2-960h/1.md Thank you! |
@WGierke, I have deleted extra files & tests passing for me locally. |
Thank you for your contribution. Your pull request has been accepted according to the TensorFlow Hub Terms of Service at www.tfhub.dev/terms and Google's Privacy Policy at https://www.google.com/policies/privacy. Your model should appear on tfhub.dev within a day. |
Any pull request you open is subject to the TensorFlow Hub Terms of Service at www.tfhub.dev/terms and Google's Privacy Policy at https://www.google.com/policies/privacy.
This PR adds fine-tuned wav2vec2 model to TFHub. This model can directly be used for inference. It achieved
~ 6%
WER on LibriSpeech test-clean dataset.@MorganR @sayakpaul