The open-source TypeScript framework for fine-tuning open-weight models.
Train, deploy, and call your own model from TypeScript.
What you can do with it.
A model tuned to your task
Fine-tune an open-weight model on your own data: endpoint logs, a file, or a Hugging Face dataset. Managed GPUs run the training; there is no infrastructure to set up.
An endpoint that serves it
Every training run produces checkpoints. Promote one to an OpenAI-compatible endpoint with its own URL and API keys, and call it from your app.
A workflow you can keep improving
Training runs are TypeScript files in your repo. Version them, re-run them on new data, and roll back to any earlier checkpoint.
How you use it.
Scaffold a project
pnpm create arkor@alpha gives you a working training script to start from.
Write the training run
Pick a base model, point at your data, set the training parameters. All in TypeScript.
Run it and deploy
Run the script. The fine-tune executes on managed GPUs and streams progress back. Deploy any checkpoint to your endpoint.
Everything else is on GitHub.
Code examples, the full API reference, and cookbook recipes live in the repository and the docs.
View on GitHub