Liquid AI releases LFM3 open-source small model family
The story so far
No overview yet — start with the coverage timeline below.
Developments
3 developments
- 28 Oct, 05:08 · 2 reportsLiquid AI Releases LFM3MarkTechPost:Liquid AI Releases LFM3: 6 Open-Weight 2B-Class Multimodal Models
- 28 Oct, 01:16 · 1 reportsBuild specialized multimodal models with LFM3Hugging Face Blog:Build specialized multimodal models with LFM3
- 27 Oct, 23:47 · 1 reportsv0.40.12-rc1 releaseOllama releases:v0.40.12-rc1
Coverage timeline
Follow the coverage to see every side of the story.
- ITHome · AILiquid AI releases the LFM3 model family, with the first batch of multiple models based on LFM3-1.7B-Base
Liquid AI releases the LFM3 model family, with the first batch of multiple models based on LFM3-1.7B-Base, among which LFM-1.7B has a 256K context window, outperforming the previous generation LFM2.5-2.6B while reducing parameters by 35%
- MarkTechPostFeaturedLiquid AI Releases LFM3: 6 Open-Weight 2B-Class Multimodal Models
According to Hugging Face blog and MarkTechPost, each model can be fine-tuned on a single GPU; the text model LFM3-1.7B has 1.73B parameters and a 256K context window, requiring about 1 GB of memory with 4-bit quantization to run, and can be deployed on phones, laptops, private servers, or the cloud. MarkTechPost states the free commercial limit is $10 million in annual revenue, and a full 128K context requires 1.6 GB of KV cache. IT Home reports that the parameters of LFM-1.7B are 35% less than the previous generation LFM2.5-2.6B, yet offer better performance.
- Hugging Face BlogFeaturedBuild specialized multimodal models with LFM3
Liquid AI releases the new generation Liquid Foundation Models, LFM3, deriving six 2B-level models from the shared backbone LFM3-1.7B-Base, covering text, vision, audio, multimodal, and decision, all open weights on Hugging Face.
- Ollama releasesv0.40.12-rc1
Liquid AI has released the LFM3 small model family, including the 1.7B parameter text model lfm3 and variants supporting image and audio functions, with a context window of up to 256K tokens; the d1 decision model has been upgraded to 2B parameters, supporting text, image, and audio inference, with Apple hardware already supporting the rc1 version.