A comprehensive guide to understanding and implementing LLM finetuning, covering both theoretical foundations and practical implementation.
Refer to the wiki for more details.
For code implementation, refer to this Jupyter Notebook.
sudo apt update
sudo apt install cmake libcurl4-openssl-dev -y
conda create --prefix=venv python=3.11 pytorch-cuda=12.4 pytorch cudatoolkit xformers -c pytorch -c nvidia -c xformers -y
conda activate ./venv
python -m pip install "unsloth[colab-new] @ git+https://github.com/unslothai/unsloth.git"
python -m pip install --no-deps trl peft accelerate bitsandbytes
python -m pip install jupyter