supertrainer2000
supertrainer2000 copied to clipboard
euclaise
With the exception of PROWrapper, most of the ranking loss wrappers are untested and possibly unfinished. - [ ] DPO (maybe should be integrated into Pairwise?) - [ ] Pairwise...
- [ ] [Elastic step decay](https://arxiv.org/abs/2110.14109) - [ ] Explore-exploit/WSD (and triangular, as a subset of this) - [ ] [Lookahead](https://proceedings.neurips.cc/paper/2019/hash/90fd4f88f588ae64038134f1eeaa023f-Abstract.html) - [ ] Variational scheduler from RWKV - [...
Currently this project entirely lacks documentation and doesn't have a useful README.