piper_multilingual_training_notebook.ipynb Notebook doesn't work, there is no model. It doesn't save anything, neither intermediate variants, nor the final model, and the shit learning process itself, at least if you look at the logs, it's complete nonsense.

[nvjob]

There is no last.ckpt file after training

https://github.com/rhasspy/piper/blob/master/notebooks/piper_multilingual_training_notebook.ipynb

[nvjob]

Notepad is completely non-working, I spent a lot of time on it, the model is not saved either intermediate, final, or. It is not clear what is being trained, there is no model.
The settings are also absolutely stupid, as are the comments on them. For example, it says “For how many epochs to save training checkpoints?”, and below “The larger your dataset, you should set this saving interval to a smaller value, as epochs can progress longer time.” And whatever I put there, what value, there is no model anywhere.
If you use the - save_last parameter
Then the model is created at the start of training and constantly changes its size, then smaller, then larger, and at the end of training it remains the same size as the training ended. T. e. it should be 860 MB, and it can be 513, 120, or any number of megabytes. In general, full of nonsense, set up and did not work out normally.

[nvjob]

Once again last tried, same thing, just the cell ends up at 5999 epochs out of 6000. There are no models anywhere.
Epoch 5999: 100% 2/2 [00:03<00:00, 1.92s/it, v_num=2, step=7440.0]Trainer.fit stopped: max_epochs=6000 reached.
Epoch 5999: 100% 2/2 [00:03<00:00, 1.92s/it, v_num=2, step=7440.0]

[9NINE21]

Anyone solve this issue help

[owenmckeith]

Try changing the output_path: to something local in colab instead of your googledrive. This seems to have done the trick for me.

[rmcpantoja]

Hi.
@nvjob @9NINE21
Try this:
https://github.com/rmcpantoja/piper/blob/master/notebooks/piper_multilingual_training_notebook.ipynb

These notebooks from this repo are too obsolete. I changed all the training pipeline some time ago with newer dependencies. There is #177 which never has been merged.

If this still seems shitty for you, train locally.

Cheers.