Today we are announcing Lilac Garden, our new cloud service for accelerating AI dataset transforms.
The first service is LLM-powered clustering, enabling a birds eye view of data, 100x faster than running locally.
Read more and sign up for the waitlist: docs.lilacml.com/blog/introduci…
We just released 0.2.3 which allows you to share Lilac-processed datasets via @huggingface from the CLI.
Expensive embeddings, signals, and map results can now be shared with others.
We'll be releasing some Lilac-processed datasets soon.
Docs: docs.lilacml.com/datasets/datas…
This clustering pipeline is all open-sourced, so you can run it locally, but may be prohibitively slow.
Lilac Garden clustering is powered by @modal_labs, allowing us to scale up to 4M documents in < 1 hour.
Sign up for the waitlist: docs.google.com/forms/d/e/1FAI…
Brian Lee just joined Lilac as our first hire, welcome :)
Brian worked at Google Brain on projects like: TensorFlow AutoGraph, a smell digitization project, and MiniGo.
He is also a terrific writer and ML engineer. Excited to have you on board!
moderndescartes.com
We're excited to announce a new way to ✨edit data✨ for AI models.
We use the @astral_sh ruff formatter to edit code in the @GlaiveAI dataset to align a coding assistant to our style.
Blog: docs.lilacml.com/blog/curate-co…
Video: youtu.be/bw8JUpAOSZQ
We just made it much easier to load and host any HuggingFace dataset in Lilac, with no code!
Just provide the Lilac Deployer UI:
✅ HuggingFace dataset name
✅ Your HF access token
We will spin up a HuggingFace space with Lilac on that dataset.
huggingface.co/spaces/lilacai…
We just released [email protected]!
This release brings in Monaco (the VSCode engine) to render documents, with powerful context menus for searching, labeling concepts, and more.
We also added UI support for common ChatML formats like ShareGPT
Release notes: github.com/lilacai/lilac/…
What should you do if you want to effectively and cheaply “instruction finetune” an LLM? @aditi_jh and @JacobianNeuro share some important insights. (1/5)