A straightforward example illustrating how to fine-tune Mistral-7B with QLoRA.
Inspired by this blogpost, which borrowed from this QLoRA notebook.
Work In Progress.
Uses huggingface, trl, peft, bitsandbytes and pytorch (obviously).
Install environment using poetry
then run poetry run python train.py
to SFT.
Note: Requires (Nvidia) GPUs to run.