Liquid AI releases on-device omni-modal model LFM3-Omni-2B
Original title: LiquidAI/LFM3-Omni-2B — new model trending #30 on Hugging Face
The 2.17B checkpoint accepts text, images, video and 16 kHz audio, and can reply with text, 24 kHz speech, or both.
Liquid AI published LFM3-Omni-2B, the first omni-modal model in its LFM3 family, now trending on Hugging Face. One end-to-end checkpoint is meant to replace separate ASR, vision-language, LLM and TTS models on phones, laptops and other edge hardware. It uses an LFM3 backbone with vision and causal audio encoders plus the LiquidCodec tokenizer, and lists 16 text languages.
Unconfirmed: The model card states both 2.2B and 2.17B parameters.
- The card lists 2.17B total parameters and also describes the model as 2.2B.
- Context length is 32K tokens, with a 128,000-token text vocabulary.
- Video is sampled at 2 fps, from 4 to 384 frames by default.
- Audio input is 16 kHz; speech output is 24 kHz.
- Images and video are processed at native resolution.
- Vendors
- Liquid AI
- AI models
- LFM3-Omni-2B
From the original article
9 min read at huggingface.cosrc="https://huggingface. co/LiquidAI/LFM3-1. 7B/resolve/main/assets/liquid-ai-banner.
By LiquidAI · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.