-
https://git.ecker.tech/ aims to provide a place to share my efforts while maintaining true ownership of my code, as I do not trust GitHub.
XMR: 4B9TQdkAkBFYrbj5ztvTx89e5LpucPeTSPzemCihdDi9EBnx7btn8RDNZTBz2zihWsjMnDkzn5As1LU6gLv3KQy8BLsZ8SG
- Joined on
2022-10-10
c231d842aa
make dependencies after the one in this repo force reinstall to downgrade, i hope, I hav eother things to do than validate this works
When saving the training configuration, an extra ' is added at the end of the path automatically. Manually editing it out of the yaml fixed it.
Ah, I see. Fixed in commit c92b006129bf6b91557e3…
Alight, now that I'm in a slightly better headspace now, I can try to explain what the loss curves mean, but with a brief crash course on what the model does (to my understanding):
The…
and its being super slow than the usual soo idk if i did something wrong with re computing voice latents or something idk...
I reverted my change to the routine that deduces sample batch…
Remedied in d3184004fd08d99fc87ecd3dd15de19705b486a6. Untested, but I'm very confident it works.
Another quick bandaid fix solution was to wrap it yourself, but would have required you to…
./training/21/
Oh, it's something really silly. The way I'm making the config is by string replacement instead of just loading the YAML, setting the keys, and exporting it like a normal…
I wonder if I botched the "cleanup" for the automatic deduction of the autoregressive sample size. I'll validate it in a moment. You can always hard-set the sample batch size under settings.
Found it. I commented out what I thought was an override. Remedied in commit eb1551ee92f64e632935f0677439271f42c89937.
adamw
do not use adamw_zero, it will keep your learning rate fixed and not decay, as I've learned the hard way. A lot of the "do this if you're distributing" (multigpu) comments don't seem…
it creates 70 voice and text dataset for that after validation. i train it to 500 epochs, that save every 50 at a learning rate of 0.00009
Thank you. The graphs would be helpful too, but…
So I did. I suppose I'll have to retrain what I've been training today, since I imagine that's a pretty big problem.