Liquid AI releases LFM3: 2B-parameter multimodal models for edge fine-tuning
What is Build specialized multimodal models with LFM3?
LFM3 is a family of 2-billion-parameter models that handle text, images, audio, video and decision tasks from a single backbone. They can be fine-tuned on a single GPU and run on devices from phones to laptops, with the text model fitting in about 1 GB at 4-bit precision and a 256 K token context window.
Who it is for
Developers building on-device AI assistants can adapt a multimodal model to their own data without needing a server farm.
How to try it
Visit https://huggingface.co/liquidai/lfm3 to download the models