LiquidAI/LFM2.5-VL-3B · Hugging Face

LFM2.5-VL-3B is a multimodal variant of LFM2.5, a family of hybrid models designed for on-device deployment. It builds on LFM2-VL-3B with further mid- and post-training. LFM2.5-VL-3B can process both text and images, and uses the LFM2.5-2.6B language model as its backbone, combined with a SigLIP2 NaFlex vision encoder.

- Better OCR: Full page OCR with layout annotation. See layout annotation format for more information.

- Efficient inference: 228 tok/s on an Apple M5 Max and 116 tok/s on an AMD Ryzen AI Max+ 395, in under 3.3 GB of memory.

- Languages: English, Arabic, Chinese, French, German, Italian, Japanese, Korean, Portuguese, Spanish, Vietnamese, Thai, Indonesian, Hindi, Russian, Polish

- Native resolution processing: Uses SigLIP2's NaFlex; large images are split into non-overlapping 512×512 patches and a resized whole-image thumbnail.

We recommend using it for single-turn, high-throughput, low-latency tasks; for example, for near-realtime object detection in automotive applications, batch processing scanned documents with OCR with layout information for turning PDFs into searchable text, or for on-device translation of menus and road signs into your native language.

It is not recommended for long-context, reasoning-intensive tasks, such as visual web design, or answering highly technical questions about blueprints.

LFM2.5-VL-3B decodes 228 tokens/s on an Apple M5 Max and 116 tokens/s on an AMD Ryzen AI Max+ 395, and fits in about 3 GB of memory. It even reaches 20 tokens/s on a Galaxy S26 Ultra, so you can run it fully on-device.

On a single NVIDIA H100 with vLLM, LFM2.5-VL-3B reaches the highest output throughput of any model we tested, about 11K tokens per second at high concurrency, or nearly 1B tokens per day.