Note: You can view the video at this location

This streamlit application is a personal tool to help and manage experimentation with GPT3 API. The original GPT3 playground is nice, but lacks version control and ability to save outputs for comparison.

I made a small streamlit application that helps to manage it. Right now one can add primes or small datasets needed to prime the Completion API. This repo contains these primes in src/gpt3_exp/datasets.

• Add your OpenAI key to gpt3_exp/gpt3_config.yml in this format:

GPT3_API: ab-XXXXXXXXXXXXXXXXXXXXXXXX

• Or you can add it via the streamlit app directly.
• Install poetry. Follow the official site or this cookbook
• Once poetry is installed, run poetry install. This will download all the packages needed (ideally in .venv) as well as setup the repository.
• To run migrations: poetry run migrate

• To run the streamlit app: poetry run st-server
• You will now be able to view the application @ localhost:8000

• Create a task.yml file in gpt3_exp/datasets, this file would be available in the streamlit app in the dropdown

• Experimentation dashboard
• Save results to DB
• Add search api to experimentation dashboard
• Create overview dashboard based off DB

• GPT3 Fine Tuning: Github Link

PRs are welcomed. Would be happy to help out with any issues :)