Edenzzzz's fix for min_8bit_size functionality in Optimizer base classes

bitsandbytes-foundation / bitsandbytes

Accessible large language models via k-bit quantization for PyTorch.

https://huggingface.co/docs/bitsandbytes/main/en/index

MIT License

6.14k stars 616 forks source link

Edenzzzz's fix for min_8bit_size functionality in Optimizer base classes #1286

Closed Titus-von-Koeller closed 2 months ago

Titus-von-Koeller commented 2 months ago

This PR supersedes and closes #1181.

I couldn't push to Edenzzzz's branch as he used the main branch and that had branch protection rules applied. So I couldn't continue in that PR as we need to release by the end of the day and want to include this valuable change.

Thanks @Edenzzzz for providing this fix, I just had to apply the same to the other optimizer base class.

Titus-von-Koeller commented 2 months ago

Optimizer tests are all green, despite the usual flakiness, but checked those cases one-by-one: everything fine. cc @matthewdouglas