DPT

DPT is a QA-bot designed to help answer questions about DagsHub. It is a fork of the brilliant buster project. Using DagsHub's documentation as reference and sentence-transformers/all-MiniLM-L6-v2 for sentence similarity, we identify documents that contain relevant information to a given query. This is then passed to OpenAI's GPT-3.5 Turbo, that uses the information and the query given a prompt to return an answer to the user query, that's hopefully helpful.

Currently, we're working on the following changeset:

• Using VectorDBs over pythonic semantic clustering
• Better hallucination detection
• Fine-grained document splitting with lookback window

Using DPT

Via Discord

DPT is available as a bot on Discord. Join our server and check out #support!

Locally

Note: DVC has a bug where it doesn't correctly interpret submodules. Please remove .gitmodules before running DVC commands, then run git checkout HEAD -- /path/to/project/root/.gitmodules to reset it back to the original state.

1. Pull the repository with submodules: clone with the --recurse-submodules flag, or git submodule update --init if the repo is already cloned. The pull from the DVC remote using dvc pull -r origin.
2. To use DPT locally, you will need an OpenAI Access Key.
3. Set that in a .env file with the format OPENAI_API_KEY=<key> in the root of the project.
4. Run python -m src.chat.converse from the project root. This sets up a very minimal conversational client that you can use to play with DPT.
5. (Optional) If you would like to use it as a discord bot, create a discord application through their developer portal.
6. Add the bot's token to the .env file with the format DISCORD_TOKEN=<token>.
7. Run python -m src.discord.bot.

Contributing

If you would like to help out, please check our active issues for tasks labelled 'help wanted'. Please fork our project, and create a PR; we will review it as soon as possible.