Models · Updated 27 Oct, 10:46 pm IST
Hugging Face launches LFM3 multimodal 2B-class family for edge specialization
Why it matters for readers: It shows progress toward smaller models that handle text, vision and audio while running on everyday hardware.
- LFM3 introduces six 2B-class models (text, vision, audio, omni, and decision) trained from a common 1.7B-base backbone.1
- A new tri-hybrid architecture and training recipe lets LFM3-1.7B outperform the previous LFM2.5-2.6B while using about 35% fewer parameters.1
- Models are built for edge use: the text model fits in roughly 1 GB at 4-bit precision and supports a 256K context window for on-device agents.1
- Every LFM3 model is intended to fine-tune on a single GPU so developers can adapt models to their data and deployment environment.1
- LFM3 includes Hugging Face’s first omni model and adds video input support for the vision model, plus new in-house vision and audio encoders.1
Get a brief like this every morning
Uzha reads hundreds of sources and gives you the stories that matter for your work, with every source linked. Free.
Get started