Tune Lab is a FastAPI + NextJS application that is purpose built for torchtune.
Note
Tune Lab is in active development and is not yet ready for use.
Tune Lab contains two main components: a torchtune + FastAPI backend and a NextJS frontend. The backend is responsible for running finetuning experiments, while the frontend allows users to quickly launch torchtune's pre-built recipes from the convience of the browser.
Clone the repo and implement your own Clerk auth application by creating an account on Clerk, and updating the local .env or production environment variables in Vercel.
Warning
the purpose of the Clerk auth wall is to gate access to compute resources behind a protected route in the app.
- Vercel: hosted deployment
- shadcn/ui: UI components
- Modal: serverless compute
- Weights & Biases: experiment tracking
- Clerk: authentication and authorization
- Discord: experiment notifications
- Docusaurus: documentation
- LaunchDarkly: feature flags
- Neon: serverless postgreSQL database
- Drizzle: database ORM
GitHub Copilot instructions can be viewed at copilot-instructions.md and copilot-personas.
Llama models tuned with Tune Lab are intended to be used with Llama Stack applications in Llama Lab.
Tune Lab is closed for contributions at this time. Thank you for your interest!
Note
Tune Lab is a portfolio project of the core maintainer