Eric Buehler

Thank you for your comments! I have restructured the XLoraModel to not inherit from LoraModel, and added serveral tests which test all methods: ``` src/peft/tuners/xlora/classifier.py 134 37 72% 83-86, 88-97,...

I have fixed some bugs, and the tests all pass on my machine. I will add the new training and saving/loading tests shortly, and will investigate the structure refactoring you...

I have refactored the forward pass to be more clear and remove that hack with the classifier. Additionally, I will add a saving/loading test. However, I am not sure what...

@BenjaminBossan thanks for opening the PR. I made some changes to fix some changes, and I also implemented one of the TODOs. You can find my changes [here](https://github.com/EricLBuehler/peft/tree/xlora-proposal-ben), and I...

I was able to remove the hacking. I think this is a more elegant solution, what do you think?

@BenjaminBossan, I have updated this PR to roughly match your suggestions, thank you for laying them out. All tests are passing, and this is the coverage: ``` src/peft/tuners/xlora/__init__.py 3 0...

@davidhewitt, @Icxolu thank you for implementing this!

Perhaps links to projects that use Inkwell could be added? In my opinion, a real use case is the best example :)

@Pat-Lafon, that really looks interesting! Currently, I am developing [Kestrel](https://github.com/EricLBuehler/kestrel), a programming language similar to Rust, with a rich type system and safety guarantees, but a different view on generics....

@jett06, I just added quantized GGUF Phi-3 support in #276! That is without LongRope support currently, but you can use a `plain` model with ISQ.