matsjfunke/rag-from-scratch

implemented vector similarity algorithms to understand their inner workings, used local embeddding models

More on how I personally think about the algorithms here

How to run scripts, and experiment

1. clone repo

   git clone https://github.com/matsjfunke/rag-from-scratch.git

2. install dependences

   python3 -m venv env
   source env/bin/activate
   pip install -r requirements.txt

3. run scripts
   • will take some time on the first run because embeddings aren’t cached (have a look at the handle_embeddings() function to understand)

   python dot-product-rag.py

4. tinkering optiions
   • each script contains a main() function this function gives you inside on tweaks you can make
   • different data: upload other .txt files in the root dir of the repo and specify thier name in file_name variable of main()
   • different prompt: change prompt vairable in main()