hk6an6 / llm_quickstart Goto Github PK

Exercises from a quickstart to large language models. There are 2 sections in this doc. First, prerequisites configure your development environment. Then, the chapters section has a link to a folder with with the relevant code for each chapter.

Before you begin, install Xcode command line tools on your MAC:

• Get Xcode: https://developer.apple.com/xcode/
• Install command line tools: https://mac.install.guide/commandlinetools/4.html

After confirming you have xcode command line tools, get Conda:

1. Setup Conda with miniforge https://github.com/conda-forge/miniforge#unix-like-platforms-mac-os--linux
2. Create a Conda environment:

mkdir llm_quickstart
conda create -p ./env  
# activate the environment:
conda activate ./env
conda install pytorch-nightly::pytorch torchvision torchaudio -c pytorch-nightly
conda install matplotlib
conda install ipykernel
conda env config vars set PYTORCH_ENABLE_MPS_FALLBACK=1

If you made it this far, you have a working environment. Well done! From this point forward, you can use conda activate ./env whenever you are in the ‘llm_quickstart’ folder.

And if you don't yet have a code editor, like VSCode. You can use Jupyter notebooks within VSCode: https://code.visualstudio.com/docs/datascience/jupyter-notebooks

Finally, install hugging face:

conda install -c huggingface transformers
conda install sentence-transformers

Wrap up by creating a kernel for your jupyter notebooks: python -m ipykernel install --user --name=pytorch211.

• Chapter 2 - semantic search
• Chapter 3 - fine tuning
• Chapter 4 - beginner prompting
• Chapter 5 - adv prompting
• Chapter 6 - recommendation engine