merve#674
Introducing LFM2.5-VL-1.6B-Extract and LFM2.5-VL-450M-Extract: Vision-language models that return structured JSON, not free-form text.
Pass in an image and a list of fields. Get back a clean JSON object.
> Two sizes: 1.6B parameters and 450M > open-weight > run on any device SoC
🧵
merve#674Introducing LFM2.5-VL-1.6B-Extract and LFM2.5-VL-450M-Extract: Vision-language models that return structured JSON, not free-form text.
Pass in an image and a list of fields. Get back a clean JSON object.
> Two sizes: 1.6B parameters and 450M > open-weight > run on any device SoC
🧵
Many users are excited about Liquid AI's LFM2.5-VL models because they enable impressive on-device structured JSON extraction from images using very small parameter counts, while one noted hallucination of extra schema keys.
LFM2.5-VL-450M-Extract and LFM2.5-VL-1.6B-Extract are available now.
> Weights: https://huggingface.co/LiquidAI/LFM2.5-VL-450M-Extract and https://huggingface.co/LiquidAI/LFM2.5-VL-1.6B-Extract > Docs: http://docs.liquid.ai
Most vision models require a parsing layer between model output and your pipeline. These don't.
You define the fields you want:
> "clock time shown on dashboard" > "number of people in scene" > "invoice total"
The model returns a populated JSON object natively.
(2/n)