TimDettmers / bitsandbytes

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

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

Pytorch XLA/PJRT TPU support

opooladz opened this issue 3 months ago · comments

Omead Pooladzandi commented 3 months ago

Feature request

Pytorch XLA/PJRT TPU support for bitsandbytes

Motivation

Would allow for faster and more memory efficient training of models on TPUs.

Your contribution

Happy to provide TPUs.