If we chose checkpoint as autoregressive model in advanced tab would that work?
FurkanGozukara opened this issue · 2 comments
FurkanGozukara commented
neonbjb commented
lol what is this!? who built this UI??
I have no idea what it is or how it works, but DLAS who two different types of resume options: you can resume from a "training state", which restores optimizers, grad scalers and model weights, or you can resume from just model weights, which resets the other settings. The latter can be useful for carrying new models across to new training scripts, for example after model surgery.
vootox commented