fastxtend
fastxtend copied to clipboard
[WIP] Switch to optimi optimizers
This PR replaces the fastxtend optimizers with optimi optimizers while retaining support for bitsandbytes eight-bit optimizers.
optimizer.utils
adds a mixin class `` to assist in converting any PyTorch optimizer into a fastai-compatible optimizer. Look in optimizer.optimi
for examples of how to use.
ToDo:
- [ ] Finish documentation
- [ ] Remove the fastxtend optimizers
- [ ] Don't automatically import fastxtend optimizers?