Handkerchief is the classy and sophisticated alternative to RAG. Stop being GPU-poor and give openai your money by shoving your entire text corpus into parallelized gpt-3.5-turbo-16K calls! Handkerchief does everything you want:

• ✅ Misses less information
• ✅ Understands more of the surrounding context
• ✅ Gives better answers when nothing is found
• ✅ Uses more GPU's!!

To use handkerchief.py, run:

pip install openai tiktoken

Get your API Key and add this line to your ~/.bashrc or ~/.zshrc

export OPENAI_API_KEY='your-api-key'

See the test() function for an example of running handkerchief. RAG is implemented with handkerchief.sneeze(), which does two things:

1. It searches through all 15K token chunks of text for relevant information.
2. It generates a response based on the information found.

NOTE: The prompts used in the script are just examples. You should engineer the prompts to suit your specific use case.

made by Ivan Yevenko