Ran some of the available markdown files through a vector embedding with openai stuff using langchain. Keeping source markdown in git for now
The vector db is written to disk (also in git) and used for questions later.
Get an openai api key then:
- Install prereqs:
python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
- Add your api key:
export OPENAI_API_KEY="..."
- Run:
python qa.py "How many box references can I pass in a single app call?"
Yes, submit a pr with a new source directory in source_data
containing markdown documents I'll add it to the vector db locally then push it.
To add a source locally, follow the steps above to set up the environment.
Then run:
python ingest.py -d source_data/$SOURCE_NAME/
- Source manifest file
- Replace current prompts with examples
- Add chat logs with QA from Discord
- Add codex to possible outputs