This is a document question answering implementation that combines several concepts together to produce an arbitrary document question answering machine.
- Sentence Transformers: used to index text into a vector database
- Question Answering Model: used to find answers in text most likely to contain them
- ChatGPT: used to post process the answers
pip3 install wisdomweaver
Index a directory of asciidoc files:
wisdom -v index --path /path/to/data
Search the index in interactive mode:
wisdom ask "What practice did many ancient C projects use?" --max 5 --threshold 0.2 --interactive --gpt
Apache 2.0
- Learn techniques used in this project: https://swedishembedded.com/tag/gpt/