LiquidAI/LFM3-VL-2B
AI model · https://huggingface.co/LiquidAI/LFM3-VL-2B
First seen by Oriel on .
AI summary
This model card introduces LFM3-VL-2B, an on-device vision-language model whose reasoning, video understanding, and spatial capabilities reportedly outperform similarly sized models across 62 benchmarks.
- The 1.92B-parameter model combines a 1.73B LFM3 language model, 164M vision encoder, and 25M projector, with a 128K-token context window.
- It accepts native-resolution text, images, and video; video defaults to 2 fps and 4–768 frames, while each 2×2 patch block becomes one language token.
- Deployment options include BF16 for Transformers, GGUF for llama.cpp, 4-bit MLX for Apple Silicon, and ONNX; vLLM and SGLang support remains unfinished.
- LFM3-VL-2B averaged 71.4 across 62 benchmarks versus 62.1 without thinking, scoring at least eight points above all similarly sized models.
- It targets documents, charts, app screens, object finding, video questions, and robotics, but is not recommended for coding, multi-turn conversations, or complex reasoning.
#vision-language models #on-device ai #video understanding #multimodal reasoning
Generated by gpt-5.6-sol. Check the original for details.
Metric trends
Last 30 days · Up to 120 recorded samples per source and metric, with reposts kept separate · First and latest retained
Likes
Total likes reported by the source.
- Hugging Face Models: 258, latest observation ; +150 over observed period since
Downloads (30d)
Rolling 30-day downloads. A change in this value is not new downloads during the observed period.
- Hugging Face Models: 12,688, latest observation ; +11,092 over observed period since
Aliases
| Source | External ID | URL |
|---|---|---|
| Hugging Face Models | LiquidAI/LFM3-VL-2B | https://huggingface.co/LiquidAI/LFM3-VL-2B Read discussion |