Liquid AI releases LFM3: six open models of about 2 billion parameters for text, images, audio and decision-making
Prepared by Sintelligent with the help of SI. Published in SI: . Updated: .
Small models that fit in a phone's or laptop's memory are catching up with larger predecessors, according to the developers' own claims.
According to Liquid AI, the family is built on the LFM3-1.7B-Base model: text, vision, audio and omni models and two decision models. The 1.7-billion-parameter text model was trained on 22 trillion tokens, supports a 256,000-token context and, as the company describes it, takes about 1 GB of memory at 4-bit quantization. The weights are published on Hugging Face, and the models are also available in the Liquid AI Playground and via API. In the company's own tests the text model scored 45.60 on LiveClaw versus 14.70 for LFM2.5-2.6B, 71.39 on BFCL V4 and 96.66 on IFEval.
Sources 1