Tune Lab is a FastAPI + NextJS application that is purpose built for torchtune.
Important
Tune Lab is in active development and is not yet ready for use
Important
Tune Lab will be built alongside Copilot and Copilot Workspace, and is closed for contributions
Note
Tune Lab is a portfolio project of jxtngx
Tune Lab is intended to be cloned and used locally, on a private cloud, or a managed cloud service.
Important
Access to NVIDIA GPUs is required regardless of deployment method
- a RESTful API for managing and tuning models with torchtune
- a NextJS web interface for managing and tuning models
- gated access to the API and web interface with Supabase
- examples for different compute provider backends
- tuneapi/: FastAPI and compute backend
- tunelab/: NextJS frontend
- copilot-instructions: Copilot system prompt
- .instructions/: Engineering personas for Copilot instructions
- .knowledge/: Knowledge base for Copilot
- .system/: System design charts for Copilot
Note
tunelab/ is a monorepo created from the shadcn/ui template and uses turborepo as the build system
Llama models tuned with Tune Lab are intended to be used with Llama Stack applications in Llama Lab.