-
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
I started disabling validation because I got better results before it was introduced
Validation has no effect on training quality (it sometimes will eat up your total iteration count and…
Oh cringe. The upstream version of torch is now 2.0, whcih I suppose I have to either hand update DLAS to do what #135 suggests, or…
I might have broke it again (this time accidental). I'll need to do some debugging whenever I get the chance.
It's the normalized version of the first text. The LJSpeech dataset has it too (albeit not as simplified). I added it in for some other add-in that required it (and abandoned that venture), but…
That should be it. It doesn't have a line for !source ./venv/bin/activate, so whatever you're using is most definitely not the most recent version.
Actually, it's already a dependency. Are you sure you ran setup first before starting?
.\venv\Scripts\activate.bat
pip install psutil
deactivate
Oh, I forgot to add it back in when I was cleaning things up when ripping metrics. The cleanup is called on training start, however.
desu the first finetune test has a much smaller size (dataset size of 4.5k for 11 epochs). Granted, all of the hyperparameters play a role in…
Commit 5e4f6808cee03732b9610521d734d063a802f731 should fix it. I tested it after manually setting the tokenizer setting, but not when it's not defined in the setting (defaulting to null).