LiquidAI six new models LFM3

by Nate: Tech

What got me about LiquidAI's drop is the fine-tuning claim: a single GPU is all it takes. Six models landed at once, spread across decision work, omni, audio, vision and text. The vision one runs 1.9B parameters and is claimed to hang with models six times bigger. They'll run on GPUs, CPUs and NPUs alike. LFM3 is the piece being framed as a new bar for on-device AI, and honestly local development is moving quicker than anyone called #Tech